EDI TO R I A L
An EU landmark for AI governance
A
groundbreaking draft law adopted with over-
whelming majority by the European Parliament
provides a glimpse into the future of governance
of artificial intelligence (AI). The AI Act (AIA)
aims to safeguard fundamental rights and ensure
the ethical development of AI in Europe and be-
yond. It is the most ambitious framework to date
designed to guide the development and usage of AI. The
vote echoes a growing chorus of researchers from vari-
ous disciplines calling for guardrails to govern powerful
AI. Although AIA’s final form will emerge from talks with
the European Council and the Commission, the decision
by Europe’s influential law-making body offers a timely
opportunity for the AI research community to brace for
impact, which is expected to reverberate across borders.
With many countries seeking to
govern AI, the AIA is hailed as the
world’s first comprehensive law reg-
“…proponents
ulating AI across applications and
contexts. Welcomed by many, it also
presents challenges due to its length,
nuanced obligations linked to intri-
contend that
rules promoting
cate risk categories, and reliance on
technical standards still in the mak-
ing. Critics argue that its density and
compliance costs may stifle innova-
trustworthy
tion, particularly among start-ups,
whereas proponents contend that AI will fuel greater
rules promoting trustworthy AI will
fuel greater innovation. innovation.”
Undoubtedly, the AIA will have
far-reaching effects on communities
engaged in applied AI research, within Europe and glob-
ally. Researchers engaged in the development of AI will
directly encounter the intricacies of the AIA. Although
it does not apply to research and development before
market placement, real-world testing conditions trig-
ger compliance. For AI researchers working in the life
sciences, employment, or education, familiarity with
AIA requirements becomes crucial if the goal is to en-
ter the European Union (EU) market. High-risk systems
as defined by AIA necessitate compliance with rigorous
premarket obligations, ranging from risk management
to various control measures and transparency require-
ments. Providers of (generative) foundation models must
meet separate obligations.
Recognizing the need to balance between regulating
risk and bolstering research, the AIA includes measures
PHOTO: ANDREAS HEDDERGOTT
to promote socially and environmentally beneficial AI de-
velopment. Enabling mechanisms include research fund-
ing but also priority access to AI regulatory sandboxes,
SCIENCE science.org
which offer innovation-friendly “test beds” for AI ex-
perimentation by using the flexibility built into regu-
lations. Furthermore, the regulation exempts free and
open-source components, unless they are part of a high-
risk AI system placed on the EU market by a provider.
The AIA also affects researchers working with
high-risk AI systems within organizations, including
hospitals and financial institutions, for professional Urs Gasser
purposes within the EU. Deployers of such systems are
is a professor
responsible for the use of relevant and unbiased input
of Public Policy,
data. For instance, a team of medical researchers de-
Governance,
ploying a customized AI in the context of personalized
p
medicine must ensure that the training data are free and Innovative
from racial, social, or religious biases. Human over- Technology and
sight and log-keeping obligations are among the re- dean of the School
quirements that may influence the of Social Sciences
operational practices of research- and Technology at
ers using high-risk AI systems. Technical University
g
The AIA also carries implica- of Munich, Munich,
tions for scholars studying the Germany. urs.
wide-ranging impacts of AI. Trans-
parency requirements, such as
gasser@tum.de
y
marking content generated by AI,
will facilitate empirical research
and the development of technical
and policy tools to address gover-
nance challenges, including deep-
fakes. That said, it is regrettable
that the European legislature has
yet to incorporate rules granting
researchers tiered access to data,
parameters, and models, particu-
y g
larly for foundation models. Such an access regime
could foster accountability and enable in-depth re-
search on social impact, as a similar access provision
in the EU Digital Services Act has shown.
The anticipated extraterritorial effects of the AIA are
reminiscent of those of the General Data Protection
,
Regulation (GDPR) and expected to reshape commer-
cial AI research practices globally. It is also a wake-up
call to the science and technology community more
broadly, highlighting the importance of integrating AI
ethics and regulation into the core curriculum of as-
piring AI researchers and educating decision-makers
about the promise and limits of AI. Initiatives such as
the launch of the School of Social Sciences and Tech-
nology at the Technical University of Munich, where
I serve as dean, exemplify strategic commitments to
such bridge-building and human-centered engineering,
inspiring reform within and beyond Europe.
–Urs Gasser
Published online 15 June 2023; 10.1126/science.adj1627
23 JUNE 2023 • VOL 380 ISSUE 6651 1203
 NEWS
It’s very sad.
“
All we were trying to do is get a cake for Father’s Day.
”
Vaccine researcher Peter Hotez, in The Washington Post, about protesters who pressed him at his home
to debate presidential candidate and vaccine opponent Robert F. Kennedy Jr. about vaccine misinformation.

IN BRIEF
Edited by
Jeffrey Brainard

p
g
y
ATMOSPHERIC SCIENCE

Volcano triggered world’s most intense lightning storm

T
he January 2022 eruption of an underwater vol- recorded in previous storms. The lightning included
cano in Tonga produced the most extreme light- record-setting bolts up to 30 kilometers tall, and some of
ning storm ever recorded. The giant, electrically them appeared to “surf” the ejected material, spreading

charged ash plume of Hunga Tonga-Hunga Ha’apai
(above) spawned nearly 200,000 lightning flashes,
researchers report this week in Geophysical
Research Letters. At its peak, lightning struck about
2600 times per minute—more than twice the peak rate
EU nature law survives key vote
E N V I R O N M E N T | The European Parliament’s
environment committee last week narrowly
rejected an attempt to kill a proposed law
aimed at restoring degraded ecosystems.
The Nature Restoration Law could restore at
least 20% of the European Union’s land and
sea areas by 2030 and all degraded ecosys-
tems by 2050. To do so, it would set multiple
binding targets, such as reversing the decline
in bees and other pollinators by 2030. The
proposal’s opponents say the restrictions
would threaten food security, jobs, farming,
y g
outward in 250-kilometer-wide rings. The analysis was
based on satellite imagery and data from radio antenna
networks, and it yielded new insights into how the erup-
tion unfolded. The research team identified four distinct
phases lasting 11 hours, far longer than was known.
,
and fishing. But days before the vote, more by 2020, nearly one-third of the HIV/AIDS
than 3300 researchers sent members of papers in the PubMed database had a lead
Parliament references to scientific findings author at an institution in Africa, up from
they said countered these claims. The com- only 5.1% in 1986, the year HIV was officially
mittee must still vote on many proposed named, reports a study this week in PLOS
amendments before the full Parliament Global Health. South Africa alone accounted
considers the measure next month. for one-third of the 83,527 HIV articles that
came from African countries during that
35-year period; another 20% came from
Africa increases HIV research Kenya, Nigeria, and Uganda. Factors that
| Early in the AIDS
I N F E CT I O U S D I S E A S E S correlated with the share of HIV publica-
epidemic, relatively few published studies tions included the size of the country’s
about the disease came from African coun- economy and population, and how many
tries, where HIV has taken a heavy toll. But residents were infected with HIV. Despite

1204 23 JUNE 2023 • VOL 380 ISSUE 6651 science.org SCIENCE

 the surge in publications from Africa, its
share in 2020 was still “relatively low” given
that 67% of the people living with HIV in the
world reside there, the researchers note.
U.K. shakes up evaluation rubric
RESEARCH METRICS | Evaluators will grade
U.K. universities more heavily on work
environment and diversity—and less on
research output—to determine their share of
higher education block grants, four national
funding bodies announced last week. The
shifts are part of a plan to change the way
the Research Excellence Framework (REF)
defines “excellence” and address criticisms
that the REF encourages universities to pri-
oritize quantity over quality in their research
outputs. The changes aim to tackle inequity
and incentivize a broader range of “contribu-
tions to knowledge,” such as data sets and
based on their potential to cause harm.
To avoid stifling innovation, lawmakers
exempted R&D before an AI system is
released for public use. The law would
ban some applications outright, including
intrusive and discriminatory uses of AI
in biometric surveillance and predictive
policing. High-risk AI systems would be
assessed before and after being put on the
market. Members of Parliament say they
hope the final version will set the tone for
global AI standards.
Open-access funders cut journals
PUBLISHING | Two-thirds of the more
than 2300 scholarly journals in a program
designed to flip them to open access
(OA)—making their articles immediately
free to read when published—failed to
meet prescribed targets for progress in
2022. As a result, the Coalition S group
of research funders, which sponsors the
program, will remove them from it at
year’s end. The funders, most of whom are
based in Europe, will no longer pay the
fees that these journals charge authors
for OA publication. The funders’ grantees
may still publish manuscripts OA in these
titles if they pay using other funds. “For
some publishers, the transition to full
and immediate open access is unlikely
to happen in a reasonable time frame,” a
Coalition S official said. The titles to be
dropped at the end of this year include
77% of those enrolled in the coalition’s
Transformational Journal program by
publishing giant Springer Nature. Almost
all members of its elite Nature family of
journals met their goals and can remain
in the program throughout 2024, when it
is scheduled to end.

p
software. They will be implemented during
the next round of evaluation, set to conclude
in 2028. Consultations over the next A fragment from Iraq
18 months will refine the assessment criteria. depicts the goddess
The push comes amid international efforts Ishtar Sharrat-niphi in a
to reshape research assessment. starburst with rosettes.

g
Groundwater use moves Earth axis
| The immense weight
P L A N E TA RY S C I E N C E

y
of groundwater pumped from aquifers has
helped shift Earth’s rotational axis, scientists
report. The effect has exceeded that from the
release of meltwater from ice in Antarctica,
they found. Despite the fixed position of the
North and South poles as shown on maps,
the orientation of the spin axis constantly
drifts because of shifts in ocean currents, the
sloshing of molten iron in Earth’s core, and
changes in the amount of water stored in ice

y g
sheets. But when researchers tried to model
this polar wandering, it didn’t match the
path observed between 1993 and 2010—until
they accounted for 2150 gigatons of water
extracted from underground, according to
a report last week in Geophysical Research

,
Letters. Most of the pumped water ended ARCHAEOLOGY
up in the ocean, a redistribution of the
planet’s mass that caused the axis to move
Goddess images found at terrorist-damaged Iraq site

78 centimeters in the direction of Iceland
during that period. The finding could help
scientists confirm how much groundwater
pumping has contributed to sea level rise.
EU’s draft AI law exempts research
POLICY | Amid worries about rapid
advances in artificial intelligence (AI), the
PHOTO: PENN MUSEUM
R
esearchers have discovered valuable fragments of a monument to a
Mesopotamian goddess among 3000-year-old ruins in northern Iraq. The site
was severely damaged in 2015 by the Islamic State extremist group, which
was largely ousted from the country by 2017. Using bombs and bulldozers, the
group destroyed aboveground structures in Nimrud, once the capital of the
ancient Assyrian Empire and one of the most important Mesopotamian heritage sites.
The destroyed buildings included a modern reconstruction of the Temple of Ishtar
that encased original historic remains. Researchers from Iraq and the University of
Pennsylvania uncovered the newly discovered relics inside the temple’s ruins, the

European Parliament last week approved
a legal framework governing its use. The
AI Act—which must now go through nego-
tiations with EU member states before
becoming law—will regulate AI systems
institution’s Museum of Archaeology and Anthropology announced last week. The finds
include fragments of a large stone monument that depict Ishtar, the Mesopotamian
goddess of love and war. One fragment (above) contains the first known representation
of Ishtar Sharrat-niphi, an aspect of the goddess associated with the planet Venus.

SCIENCE science.org 23 JUNE 2023 • VOL 380 ISSUE 6651 1205


IN DEP TH
BIOLOGY
Human stem cells turned into
detailed lab replicas of embryos
Mock embryos created by multiple groups recapitulate
developmental events beyond implantation
By Mitch Leslie
B
iologists trying to create laboratory
models of the early growth of a hu-
man embryo have taken a major step
forward. In preprints posted online
last week, four research teams re-
ported using various kinds of human
stem cells, some genetically modified, to
create ersatz embryos that closely resemble
real embryos that are up to 14 days old. The
creations replicate the period after the em-
bryo implants in the uterus, which is very
difficult to study.
The rush of papers was triggered when
the leader of one group, developmental bio-
logist Magdalena Zernicka-Goetz, briefly
described her team’s results at the Inter-
1206 23 JUNE 2023 • VOL 380 ISSUE 6651
national Society for Stem Cell Research
(ISSCR) meeting in Boston on 14 June.
Other scientists are still assessing the
four groups’ claims but say the embryo
stand-ins are significant. “This is a very
important stage of development” that re-
searchers need to be able to model, says
stem cell biologist Aryeh Warmflash of
Rice University, who wasn’t connected to
the work. “These are the most complete
models of the early postimplantation hu-
man embryo.”
Besides clarifying early human develop-
ment, the new embryo models could help
researchers better understand birth defects
and probe the safety of drugs used during
pregnancy. But they pose fraught issues.
U.K. law, for example, prohibits research on
Placenta precursor cells (pink) highlight an 8-day-old
embryo-like structure made from human stem cells.
donated in vitro fertilization (IVF) embryos
that are beyond 14 days old. But because
these new models are formed from human
stem cells, not eggs and sperm, that law
does not apply to them. The ethical and
moral status of creations like these also
remains unresolved. Embryo models are
“a matter of significant moral discussion
and of significant moral concern,” says
ethicist J. Benjamin Hurlbut of Arizona
State University.
Scientists have previously spurred hu-
man stem cells to form structures that
closely resemble the blastocyst, the embry-
onic stage that begins about 5 days after
fertilization and implants in the womb.
These imitation blastocysts, known as
blastoids, help researchers probe events
p
around implantation, which may be key to
some cases of infertility. But blastoids stag-
nate developmentally, making it hard to in-
vestigate changes that occur slightly later.
The period of human development imme-
diately after implantation is “truly a black
g
box,” says stem cell biologist Jacob Hanna
of the Weizmann Institute of Science.
One way in which researchers have tried
to delve into events after the blastocyst
y
stage is by producing ersatz embryos from
mouse stem cells. Last year, for example,
Hanna and colleagues grew such cells into
embryo mimics that sported a beating
heart, the rudiments of a brain and spinal
cord, and incipient muscles.
To create similar models of human de-
velopment, Hanna and his team started
with cell lines previously cultured from
early human embryos and with stem cells
y g
converted from adult cells. A good mimic
needs to contain not just the cells of the
embryo proper, but also the cells that pro-
duce the so-called extraembryonic tissues
that help nurture the growing offspring,
such as the placenta and umbilical cord.
,
Using cell culture media developed by
other scientists and mixtures they crafted
themselves, the researchers nudged
their stem cells to specialize into extra-
embryonic cell lineages found in a genuine
embryo. They then allowed these lineages
to mingle with stem cells. The resulting cell
clusters showed placenta precursor cells
called trophoblasts and other hallmarks of
postblastocyst embryos, the team reports
in its paper on the bioRxiv preprint server.
Zernicka-Goetz, who divides her re-
search between the University of Cam-
bridge and the California Institute of
Technology, and her colleagues took a
different approach. They, too, had previ-
ously grown mouse stem cells into cellular
science.org SCIENCE

structures that recapture postimplantation
events. In the new work, led by Cambridge
Ph.D. student Bailey Weatherbee, they gen-
erated extraembryonic cell lineages from
two kinds of human embryonic stem cells
that had been genetically modified to dif-
ferentiate into the tissue types when ex-
posed to the antibiotic doxycycline. The
scientists then mixed the results with
unmodified human embryonic stem cells,
and the three types of cells coalesced to
produce embryo-like clumps. “These struc-
tures do not recapitulate all aspects of the
embryo, but rather serve as a complemen-
tary tool for us to study … key stages of
development,” Weatherbee says.
A third group, led by University of Pitts-
burgh’s Mo Ebrahimkhani, used geneti-
cally modified induced pluripotent stem
cells—the embryo-like cells created from
adult human cells—to form embryonic and
extraembryonic tissues in cell clusters they
call iDiscoids. The fourth group, a Chi-
nese team led by Tianqing Li of Kunming
University of Science and Technology, cre-
ated its E-assembloids by mixing freshly
derived human embryonic stem cells and
extraembryonic cells.
None of the lab creations perfectly re-
produces the organization of natural post-
implantation embryos, says developmental
biologist Janet Rossant of the Hospital for
Sick Children. But that isn’t a bad thing,
she adds. “The issue is to use the right
model for the system you want to address.”
Researchers now want to improve their
procedures to produce even more accurate
embryo stand-ins, Hanna says, so they can
obtain a clearer picture of how early hu-
man development unfolds. His team and
the Cambridge team stopped their experi-
ments when the ersatz embryos were the
equivalent of natural 14-day-old human
embryos. That cutoff was long the agreed-
on limit for growing IVF-derived human
embryos. However, Zernicka-Goetz and
other researchers have argued that study-
ing IVF embryos for longer would be valu-
able. ISSCR, the organization that sets
permissible research practices in the field,
dropped its strict 14-day limit on culture of
IVF embryos in a 2021 update to its guide-
lines. The guidelines do not set a specific
time limit for embryo models like the ones
generated from stem cells.
The University of Cambridge last week
PHOTO: SKYNESHER/GETTYIMAGES
launched a project, involving scientists, bio-
ethicists, and legal scholars, that aims to re-
lease recommendations on the appropriate
use of such models of human development
later this year. “The big question is how the
boundary between a tissue culture and a hu-
man organism is going to be drawn and on
what criteria,” Hurlbut says. j
SCIENCE science.org
WORKFORCE
Conferences confront abortion
bans and anti-LGBTQ+ laws
Survey shows few societies are pulling out of affected
U.S. states, but many are implementing safety measures
By Amanda Heidt
W
hen Claire Kouba heard that the
American Geophysical Union’s
(AGU’s) annual meeting would
be held in New Orleans in 2025,
she was worried. After the U.S.
Supreme Court’s June 2022 deci-
sion overturning Roe v. Wade, which guar-
anteed a constitutional right to abortion,
Louisiana had banned the procedure with
few exceptions. Kouba—a hydrogeologist
at the University of California, Davis, and
an AGU member since 2011—had attended
the 2022 meeting in Chicago while early
in her first pregnancy and may be trying
to grow her family again in 2025. As she
contemplated traveling to Louisiana, she
found herself fearing for her own safety
and that of other attendees who might
need emergency medical care for compli-
cations such as miscarriage, ectopic preg-
nancy, and septic uterus.
Other professional organizations are
also planning conferences in states that
have enacted restrictive abortion laws,
anti-LGBTQ+ legislation, or measures seen
as racially discriminatory, raising similar
concerns. Some have relocated their meet-
ings: The American College of Obstetricians
and Gynecologists moved its May meeting
this year, from Louisiana to Maryland, af-
Some pregnant attendees fear they will be at risk if they travel to conferences in certain U.S. states.
N E WS
ter members worried they could be arrested
for describing their work, and the American
Association of Immunologists relocated its
2024 meeting from Arizona to Illinois, issu-
ing a statement decrying the “dramatic and
deleterious impact” the Supreme Court’s
Dobbs v. Jackson Women’s Health Organiza-
tion decision was having on patients.
But they are exceptions, according to a
p
Science survey of U.S. conferences across
STEM disciplines. Of the 15 organizations
that responded, 10 currently have meetings
planned in states with restrictive abortion
policies or anti-LGBTQ+ legislation; none
is relocating. Four cited the steep financial
g
costs associated with pulling out of exist-
ing contracts typically negotiated years in
advance. But at least two-thirds of the or-
ganizations are taking other measures to
y
address members’ concerns. Among them:
new guidelines for choosing future meeting
venues and safety measures to protect vul-
nerable groups. Organizers and attendees
alike have also expressed hope that the hy-
brid and virtual offerings popularized during
COVID-19, which can increase participation
and diversity, will remain an option.
AGU is among those not relocating,
though Kouba asked leaders to do so in
y g
an open letter that garnered more than
800 signatories. The letter pointed out that
AGU had shifted to a virtual platform dur-
,
23 JUNE 2023 • VOL 380 ISSUE 6651 1207

structures that recapture postimplantation
events. In the new work, led by Cambridge
Ph.D. student Bailey Weatherbee, they gen-
erated extraembryonic cell lineages from
two kinds of human embryonic stem cells
that had been genetically modified to dif-
ferentiate into the tissue types when ex-
posed to the antibiotic doxycycline. The
scientists then mixed the results with
unmodified human embryonic stem cells,
and the three types of cells coalesced to
produce embryo-like clumps. “These struc-
tures do not recapitulate all aspects of the
embryo, but rather serve as a complemen-
tary tool for us to study … key stages of
development,” Weatherbee says.
A third group, led by University of Pitts-
burgh’s Mo Ebrahimkhani, used geneti-
cally modified induced pluripotent stem
cells—the embryo-like cells created from
adult human cells—to form embryonic and
extraembryonic tissues in cell clusters they
call iDiscoids. The fourth group, a Chi-
nese team led by Tianqing Li of Kunming
University of Science and Technology, cre-
ated its E-assembloids by mixing freshly
derived human embryonic stem cells and
extraembryonic cells.
None of the lab creations perfectly re-
produces the organization of natural post-
implantation embryos, says developmental
biologist Janet Rossant of the Hospital for
Sick Children. But that isn’t a bad thing,
she adds. “The issue is to use the right
model for the system you want to address.”
Researchers now want to improve their
procedures to produce even more accurate
embryo stand-ins, Hanna says, so they can
obtain a clearer picture of how early hu-
man development unfolds. His team and
the Cambridge team stopped their experi-
ments when the ersatz embryos were the
equivalent of natural 14-day-old human
embryos. That cutoff was long the agreed-
on limit for growing IVF-derived human
embryos. However, Zernicka-Goetz and
other researchers have argued that study-
ing IVF embryos for longer would be valu-
able. ISSCR, the organization that sets
permissible research practices in the field,
dropped its strict 14-day limit on culture of
IVF embryos in a 2021 update to its guide-
lines. The guidelines do not set a specific
time limit for embryo models like the ones
generated from stem cells.
The University of Cambridge last week
PHOTO: SKYNESHER/GETTYIMAGES
launched a project, involving scientists, bio-
ethicists, and legal scholars, that aims to re-
lease recommendations on the appropriate
use of such models of human development
later this year. “The big question is how the
boundary between a tissue culture and a hu-
man organism is going to be drawn and on
what criteria,” Hurlbut says. j
SCIENCE science.org
WORKFORCE
Conferences confront abortion
bans and anti-LGBTQ+ laws
Survey shows few societies are pulling out of affected
U.S. states, but many are implementing safety measures
By Amanda Heidt
W
hen Claire Kouba heard that the
American Geophysical Union’s
(AGU’s) annual meeting would
be held in New Orleans in 2025,
she was worried. After the U.S.
Supreme Court’s June 2022 deci-
sion overturning Roe v. Wade, which guar-
anteed a constitutional right to abortion,
Louisiana had banned the procedure with
few exceptions. Kouba—a hydrogeologist
at the University of California, Davis, and
an AGU member since 2011—had attended
the 2022 meeting in Chicago while early
in her first pregnancy and may be trying
to grow her family again in 2025. As she
contemplated traveling to Louisiana, she
found herself fearing for her own safety
and that of other attendees who might
need emergency medical care for compli-
cations such as miscarriage, ectopic preg-
nancy, and septic uterus.
Other professional organizations are
also planning conferences in states that
have enacted restrictive abortion laws,
anti-LGBTQ+ legislation, or measures seen
as racially discriminatory, raising similar
concerns. Some have relocated their meet-
ings: The American College of Obstetricians
and Gynecologists moved its May meeting
this year, from Louisiana to Maryland, af-
Some pregnant attendees fear they will be at risk if they travel to conferences in certain U.S. states.
N E WS
ter members worried they could be arrested
for describing their work, and the American
Association of Immunologists relocated its
2024 meeting from Arizona to Illinois, issu-
ing a statement decrying the “dramatic and
deleterious impact” the Supreme Court’s
Dobbs v. Jackson Women’s Health Organiza-
tion decision was having on patients.
But they are exceptions, according to a
p
Science survey of U.S. conferences across
STEM disciplines. Of the 15 organizations
that responded, 10 currently have meetings
planned in states with restrictive abortion
policies or anti-LGBTQ+ legislation; none
is relocating. Four cited the steep financial
g
costs associated with pulling out of exist-
ing contracts typically negotiated years in
advance. But at least two-thirds of the or-
ganizations are taking other measures to
y
address members’ concerns. Among them:
new guidelines for choosing future meeting
venues and safety measures to protect vul-
nerable groups. Organizers and attendees
alike have also expressed hope that the hy-
brid and virtual offerings popularized during
COVID-19, which can increase participation
and diversity, will remain an option.
AGU is among those not relocating,
though Kouba asked leaders to do so in
y g
an open letter that garnered more than
800 signatories. The letter pointed out that
AGU had shifted to a virtual platform dur-
,
23 JUNE 2023 • VOL 380 ISSUE 6651 1207
NE WS | I N D E P T H
ing COVID-19 and said the health risks to
pregnant people warrant similar consider-
ation. “AGU members can gain significant
professional benefits through in-person
conference attendance, but pregnant AGU
members should not face disproportionate
medical risks to do so,” Kouba says.
The letter spurred much activity and
conversation within the organization, in-
cluding a roundtable to discuss the Dobbs
decision and a member survey. Lauren
Parr, AGU’s senior vice president of meet-
ings and learning, says the experience
brought home how “the current climate in
the United States” was challenging AGU’s
goal of making meetings feel safe and wel-
coming. In preparation for the New Or-
leans meeting, Parr has been in touch with
medical professionals at local hospitals,
who have guaranteed “safe and equitable
health care” for attendees.
Other societies have taken their own
steps. The American Chemical Society
(ACS), which held a meeting in Indiana
in March, asked speakers in one session
to dedicate a few minutes of their talk to
issues related to diversity after the state
banned all forms of gender-affirming care
for minors. And the American Anthropo-
logical Association (AAA) responded after
NAACP released a travel advisory for Mis-
souri, citing “looming danger” for Black
people there following the passage of a
law making it more difficult to sue busi-
nesses over racial discrimination, among
other concerns. AAA, which has a meeting
scheduled in Missouri in 2026, began con-
versations with the state’s NAACP chapter,
city and county officials, and the Conven-
tion and Visitors Bureau. (A similar advi-
sory has since been released for Florida,
where AAA also has a future meeting.)
For Kouba, AGU’s response has been
gratifying, though she says she’s likely to
remain home. Some botanists will soon
face the same decision: Next month, Bot-
any 2023—a joint meeting for six plant
science societies—is to be held in Idaho,
which has some of the most extreme anti-
abortion laws in the country. On 30 March,
the planning committee released a state-
ment noting that breaking the contract
would be “fiscally irresponsible and finan-
cially devastating for most—if not all—our
partnering societies.” The meeting will
have a virtual component, and the loca-
tion is a deal breaker for some. University
of Oregon botanist Hilary Rose Dawson,
a queer woman, wrote on Twitter that al-
though she appreciated the conference’s
statement, “I still feel unsafe attending.”
Some societies have decided to plan fu-
ture meetings in states where attendees will
feel safer. Before the Dobbs ruling, the So-
1208 23 JUNE 2023 • VOL 380 ISSUE 6651
ciety for Integrative and Comparative Bio-
logy (SICB) made plans to hold this year’s
meeting in Texas and its 2025 meeting in
Georgia. Both states have passed laws that
limit or ban abortion, criminalize gender-
affirming care for minors, and tamp down
discussion of race in schools. This year’s
meeting went forward as planned in Janu-
ary, with a virtual component, and the
society is not pulling out of Georgia for
2025. But in planning the society’s 2026
gathering, executives surveyed members
and held member forums to discuss anti-
LGBTQ+ policies. The feedback led SICB
to limit the states for consideration. “We
drew a line at the notion that our members
could be arrested at a meeting we were
having,” says society president Patricia
Hernandez, referring to laws that ban pro-
viding gender-affirming care, as some po-
tential attendees do for dependents.
But ruling out certain locations can have
downsides, others say. Any location—even
one in a seemingly “safe” state—will have
drawbacks, Parr says, and excluding cer-
tain states risks alienating large swaths
of an organization’s membership. Pulling
out of a planned venue can also damage
relationships with the local community,
which may rely on the tourism dollars and
jobs that conferences bring, Parr points
out. “Instead, can we potentially use our
voice for good while we’re there? Can we
be more inclusive? Can we make sure that
we’re hiring minority-owned businesses to
participate with us?”
Others agree it can be a mistake to re-
treat. “We need to make sure not to let
people who are currently living under the
harshest circumstances feel abandoned,”
says Daniel Muratore, an oceanographer
at the Santa Fe Institute in New Mexico
who co-authored a preprint about building
queer- and trans-inclusive conferences.
Co-author Rachel Gregor, a marine micro-
biologist at the Massachusetts Institute of
Technology, says discussions about swap-
ping venues were among the most “spir-
ited and nuanced” she and her colleagues
had, but many agreed that blanket bans
can be problematic.
Amanda Morris, a chemist at the Vir-
ginia Polytechnic Institute and State
University who touched on legislation
targeting transgender people in Indiana
during her presentation at the ACS confer-
ence there, acknowledges that in the fast-
changing U.S. political environment, the
challenges facing conference organizers
might have been hard to predict. But now,
she says, “I do expect them to show up for
members and support them.” j
Amanda Heidt is a freelance journalist in Moab, Utah.
ASTROPHYSICS
Infrastructure
woes could
slow South Pole
experiments
New telescopes and
neutrino detectors must wait
for work on polar facilities
By Adrian Cho
p
F
or decades, the South Pole has been a
wonderland for physicists. They have
stared into its exquisitely clear sky to
study the afterglow of the big bang—
the cosmic microwave background
(CMB)—and used the ice itself to spot
g
nearly undetectable particles called neutri-
nos streaking from other galaxies. But plans
for two big new projects at the pole could be
delayed for years by less lofty infrastructure
y
issues and a limited electrical supply.
In a 12 June “Dear Colleague” letter, the
National Science Foundation (NSF) told re-
searchers that, after the pandemic lockdown,
it must attend to a backlog of maintenance
at multiple Antarctic sites. At the South Pole,
for the next several years the agency will
only support experiments that have already
been approved to go forward. “[P]roposers
seeking support for new projects at South
y g
Pole Station should consult the cognizant
program officer to discuss alternative loca-
tions,” the letter says.
That sentence has physicists wondering
when NSF will go forward with two major
projects that must be built at the pole. One
,
would deploy new CMB telescopes there and
the other would expand the gargantuan Ice-
Cube neutrino detector. The projects, known
as CMB-S4 and IceCube-Gen2, aim to start
building within 5 years. “Maybe in the next
few weeks we’ll get a little more clarity what
the Dear Colleague letter actually means,”
says Mary Hall Reno, a theoretical physicist
at the University of Iowa.
NSF’s warning comes at a delicate time, as
by October an ad hoc Particle Physics Proj-
ect Prioritization Panel (P5) must formulate
a new 10-year plan for U.S. particle physics.
“The problem is nobody knows” the true
time frame for the projects, says Hitoshi
Murayama, a theorist at the University of
California, Berkeley, who chairs P5.
science.org SCIENCE
NE WS | I N D E P T H
ing COVID-19 and said the health risks to
pregnant people warrant similar consider-
ation. “AGU members can gain significant
professional benefits through in-person
conference attendance, but pregnant AGU
members should not face disproportionate
medical risks to do so,” Kouba says.
The letter spurred much activity and
conversation within the organization, in-
cluding a roundtable to discuss the Dobbs
decision and a member survey. Lauren
Parr, AGU’s senior vice president of meet-
ings and learning, says the experience
brought home how “the current climate in
the United States” was challenging AGU’s
goal of making meetings feel safe and wel-
coming. In preparation for the New Or-
leans meeting, Parr has been in touch with
medical professionals at local hospitals,
who have guaranteed “safe and equitable
health care” for attendees.
Other societies have taken their own
steps. The American Chemical Society
(ACS), which held a meeting in Indiana
in March, asked speakers in one session
to dedicate a few minutes of their talk to
issues related to diversity after the state
banned all forms of gender-affirming care
for minors. And the American Anthropo-
logical Association (AAA) responded after
NAACP released a travel advisory for Mis-
souri, citing “looming danger” for Black
people there following the passage of a
law making it more difficult to sue busi-
nesses over racial discrimination, among
other concerns. AAA, which has a meeting
scheduled in Missouri in 2026, began con-
versations with the state’s NAACP chapter,
city and county officials, and the Conven-
tion and Visitors Bureau. (A similar advi-
sory has since been released for Florida,
where AAA also has a future meeting.)
For Kouba, AGU’s response has been
gratifying, though she says she’s likely to
remain home. Some botanists will soon
face the same decision: Next month, Bot-
any 2023—a joint meeting for six plant
science societies—is to be held in Idaho,
which has some of the most extreme anti-
abortion laws in the country. On 30 March,
the planning committee released a state-
ment noting that breaking the contract
would be “fiscally irresponsible and finan-
cially devastating for most—if not all—our
partnering societies.” The meeting will
have a virtual component, and the loca-
tion is a deal breaker for some. University
of Oregon botanist Hilary Rose Dawson,
a queer woman, wrote on Twitter that al-
though she appreciated the conference’s
statement, “I still feel unsafe attending.”
Some societies have decided to plan fu-
ture meetings in states where attendees will
feel safer. Before the Dobbs ruling, the So-
1208 23 JUNE 2023 • VOL 380 ISSUE 6651
ciety for Integrative and Comparative Bio-
logy (SICB) made plans to hold this year’s
meeting in Texas and its 2025 meeting in
Georgia. Both states have passed laws that
limit or ban abortion, criminalize gender-
affirming care for minors, and tamp down
discussion of race in schools. This year’s
meeting went forward as planned in Janu-
ary, with a virtual component, and the
society is not pulling out of Georgia for
2025. But in planning the society’s 2026
gathering, executives surveyed members
and held member forums to discuss anti-
LGBTQ+ policies. The feedback led SICB
to limit the states for consideration. “We
drew a line at the notion that our members
could be arrested at a meeting we were
having,” says society president Patricia
Hernandez, referring to laws that ban pro-
viding gender-affirming care, as some po-
tential attendees do for dependents.
But ruling out certain locations can have
downsides, others say. Any location—even
one in a seemingly “safe” state—will have
drawbacks, Parr says, and excluding cer-
tain states risks alienating large swaths
of an organization’s membership. Pulling
out of a planned venue can also damage
relationships with the local community,
which may rely on the tourism dollars and
jobs that conferences bring, Parr points
out. “Instead, can we potentially use our
voice for good while we’re there? Can we
be more inclusive? Can we make sure that
we’re hiring minority-owned businesses to
participate with us?”
Others agree it can be a mistake to re-
treat. “We need to make sure not to let
people who are currently living under the
harshest circumstances feel abandoned,”
says Daniel Muratore, an oceanographer
at the Santa Fe Institute in New Mexico
who co-authored a preprint about building
queer- and trans-inclusive conferences.
Co-author Rachel Gregor, a marine micro-
biologist at the Massachusetts Institute of
Technology, says discussions about swap-
ping venues were among the most “spir-
ited and nuanced” she and her colleagues
had, but many agreed that blanket bans
can be problematic.
Amanda Morris, a chemist at the Vir-
ginia Polytechnic Institute and State
University who touched on legislation
targeting transgender people in Indiana
during her presentation at the ACS confer-
ence there, acknowledges that in the fast-
changing U.S. political environment, the
challenges facing conference organizers
might have been hard to predict. But now,
she says, “I do expect them to show up for
members and support them.” j
Amanda Heidt is a freelance journalist in Moab, Utah.
ASTROPHYSICS
Infrastructure
woes could
slow South Pole
experiments
New telescopes and
neutrino detectors must wait
for work on polar facilities
By Adrian Cho
p
F
or decades, the South Pole has been a
wonderland for physicists. They have
stared into its exquisitely clear sky to
study the afterglow of the big bang—
the cosmic microwave background
(CMB)—and used the ice itself to spot
g
nearly undetectable particles called neutri-
nos streaking from other galaxies. But plans
for two big new projects at the pole could be
delayed for years by less lofty infrastructure
y
issues and a limited electrical supply.
In a 12 June “Dear Colleague” letter, the
National Science Foundation (NSF) told re-
searchers that, after the pandemic lockdown,
it must attend to a backlog of maintenance
at multiple Antarctic sites. At the South Pole,
for the next several years the agency will
only support experiments that have already
been approved to go forward. “[P]roposers
seeking support for new projects at South
y g
Pole Station should consult the cognizant
program officer to discuss alternative loca-
tions,” the letter says.
That sentence has physicists wondering
when NSF will go forward with two major
projects that must be built at the pole. One
,
would deploy new CMB telescopes there and
the other would expand the gargantuan Ice-
Cube neutrino detector. The projects, known
as CMB-S4 and IceCube-Gen2, aim to start
building within 5 years. “Maybe in the next
few weeks we’ll get a little more clarity what
the Dear Colleague letter actually means,”
says Mary Hall Reno, a theoretical physicist
at the University of Iowa.
NSF’s warning comes at a delicate time, as
by October an ad hoc Particle Physics Proj-
ect Prioritization Panel (P5) must formulate
a new 10-year plan for U.S. particle physics.
“The problem is nobody knows” the true
time frame for the projects, says Hitoshi
Murayama, a theorist at the University of
California, Berkeley, who chairs P5.
science.org SCIENCE
 The Amundsen-Scott South Pole Station is sinking into the accumulating snow and needs to be elevated before new science projects can be built.
But physicists may be getting ahead of
themselves, says James Ulvestad, acting di-
rector of NSF’s office of polar programs. Nei-
ther project has advanced beyond the R&D
phase, he notes, much less gained approval
for construction or funding from Congress.
“Just getting through that process and then
preparing for construction almost inevitably
takes you nearer to the end of the decade any-
way,” he says. He adds that as currently con-
ceived, the projects would likely require more
electrical power than the South Pole Station
can provide.
CMB-S4 aims to be the definitive ground-
based study of the big bang’s lingering
radiation, which is rich in clues to the uni-
verse’s origin and structure. An $840 mil-
lion collaboration between NSF and the
Department of Energy, CMB-S4 would erect
two 6-meter microwave telescopes in Chile
and one 5-meter and nine 0.5-meter tele-
scopes at the South Pole. Those at the pole
would be essential for a key goal: detecting
pinwheel-like swirls in the polarization of
the CMB known as primordial B-modes.
Those swirls would be a signal of distor-
tions of space called gravitational waves
rippling through the early universe, which
in turn would be evidence that the newborn
cosmos underwent an exponential growth
spurt known as inflation.
The CMB-S4 team is ready to start prep-
ping the site as soon as it gets the go-ahead,
hopefully in 2028, says John Carlstrom, an
astrophysicist and team member at the Uni-
PHOTO: RAFFAELA BUSSE CC BY-NC-ND
versity of Chicago. There is some urgency, he
notes, as that year Japan plans to launch a
spacecraft called LightBIRD that also aims to
detect the primordial B modes. “To be hon-
est, late in the decade works,” Carlstrom says.
“But if late in the decade becomes late in the
next decade, we’re in trouble.”
IceCube-Gen2 would increase the size of
SCIENCE science.org
the world’s biggest neutrino detector, which
spots ultra–high energy neutrinos from space
with strings of optical sensors embedded
deep in the ice. The 5160 detectors watch
for a flash of light—an optical shock wave—
radiated by charged particles generated when
a neutrino collides with an atomic nucleus.
Last year, IceCube physicists reported 79 neu-
trinos coming from the heart of a nearby gal-
axy called NGC 1068, the first steady source
of extragalactic neutrinos pinpointed.
The Gen2 project would deploy another
9600 optical sensors to increase the detec-
tor’s volume from 1 to 8 cubic kilometers, en-
abling it to detect more neutrino sources and
ushering in an era of neutrino astronomy. Re-
searchers hope to deploy the first new strings
in 2028 and complete the task in 7 or 8 years.
If the project slips several years, collaborators
might look for other projects, Reno notes.
“We’ve already had a 3-year delay with CO-
VID,” she says. “With increased delays will
you be losing the human resources you need
to keep this sophisticated program going?”
But neither project can start before NSF
fixes the infrastructure, Ulvestad says. The
7400-square-meter Amundsen-Scott South
Pole Station and the support buildings for
IceCube and the current CMB telescopes are
all sinking into accumulating snow and need
to be elevated. And the 48-year-old, snow-
covered Quonset huts used to store fuel, sup-
plies, and vehicles need reinforcing.
Congress has appropriated $60 million
this year for such work, and NSF plans to
spend similar amounts in the next 2 years.
But the logistics of getting materials and
people to the pole mean work there won’t
begin until late 2026 and will likely take 2 or
3 years, Ulvestad says. That pushes construc-
tion for both science projects to 2028 or 2029
at the earliest.
It could come significantly later. Before
p
either project can be presented to NSF’s
governing National Science Board for ap-
proval, it must go through a design stage
that typically lasts 3 to 6 years, says Linnea
Avallone, NSF’s chief officer for research
facilities. “Congress in particular has been
g
very outspoken about wanting us to follow
our processes,” she says. So, approval might
not come until the end of the decade, with
polar work to start a few years later.
y
Among the unresolved design issues is
that of electrical power. CMB-S4 would re-
quire 170 kilowatts, 20% more than the
current CMB telescopes. Eventually, IceCube-
Gen2 would require 190 kilowatts, 2.5 times
as much as IceCube. But power at the pole is
limited, supplied by a diesel generator pro-
ducing between 600 and 700 kilowatts.
In principle, NSF could install more gen-
erators and haul more fuel to the pole, says
y g
Albrecht Karle, a physicist and IceCube team
member at the University of Wisconsin-
Madison. Most fuel is towed overland
1600 kilometers from McMurdo Station on
the coast by tractors in the so-called South
Pole Traverse, which has run since 2006. “The
,
traverse is generally scalable,” Karle says.
That’s not necessarily so, Ulvestad says.
NSF runs three traverses per season, and for
reasons of safety, increasing that number is
unlikely, he says. Boosting the grid capac-
ity at the South Pole would also require a
separate project, not even proposed so far,
that likely couldn’t be realized much before
2040, Ulvestad says.
Getting pushed late into the next decade
is exactly what physicists hope to avoid. “It
would be a shame to think after all these
years, decades even, that we would get
beat” to discovering primordial B modes,
Carlstrom says. “It would feel like, ‘Really,
we didn’t do it because we’re waiting for a
building to be elevated?’” j
23 JUNE 2023 • VOL 380 ISSUE 6651 1209
NE WS | I N D E P T H
GEOPHYSICS
Undersea mountains help slicken tectonic plates
Swallowed up seamounts play pivotal role in “slow slip” earthquakes at subduction zones
By Paul Voosen
I
n 2001, geoscientists reported a com-
pletely new kind of earthquake at a sub-
duction zone, a seam where a tectonic
plate of ocean crust dives under a conti-
nent. Subduction zones were previously
thought to behave in one of two ways:
either creeping along steadily and smoothly,
without any tremors at all, or sticking for
decades or centuries and then rupturing
catastrophically in the world’s largest earth-
quakes. But geoscientists now
know subduction zones often
take a middle path: GPS sen-
sors have shown they can slip
in quiet, nearly imperceptible
earthquakes that last weeks
or months. What they haven’t
known is why.
Evidence from the Hi-
kurangi subduction zone in
New Zealand now suggests
these “slow slip” events often
depend on seamounts, the
underwater volcanoes that
stud the sea floor in large num-
bers (Science, 21 April, p. 226).
You might expect a mountain
swallowed up by a fault to act
as a sticking point. But it turns lubricate New Zealand’s Hikurangi subduction zone as they are devoured.
out that many seamounts pro-
vide the grease, says Nathan Bangs, a marine
geophysicist at the University of Texas (UT)
at Austin and lead author of a new 3D survey
of Hikurangi, published earlier this month
in Nature Geoscience. “A lot of the intuition
doesn’t seem to be applying here,” he says.
“You think it’s doing one thing, and it’s doing
something else.”
Slow slip events are not simply an aca-
demic curiosity, adds Laura Wallace, a UT
Austin geodesist. “Understanding where slow
slip is happening versus where faults are
locked up is important for seismic hazard as-
sessments,” she says. And more mysteriously,
two recent subduction zone earthquakes,
including the giant 2011 Tohoku quake in
Japan, were preceded by slow slip events—
and perhaps triggered by them. But the
relationships are murky. “We are pretty con-
fused about this issue,” says Emily Brodsky,
an earthquake physicist at the University of
California, Santa Cruz.
The Hikurangi subduction zone, which
dives under the east coast of New Zealand’s
1210 23 JUNE 2023 • VOL 380 ISSUE 6651
north island along a seafloor trench, is a
hot spot for slow slip research because the
motion occurs in a shallow zone just a few
kilometers below the sea floor. In 2018, the
Marcus Langseth, a U.S. seismic imaging
ship, mapped part of the fault with a degree
of detail that provided “an unprecedented
look at a subduction zone,” Bangs says.
The Langseth crisscrossed the region, tow-
ing long strings of hydrophones to catch re-
flections from nearly 150,000 airgun blasts,
which penetrate the water and bounce off
Seamounts (seen as bumps on the smooth ocean floor) seem to
layers of sediment and rock under the sea
floor. In addition, Japanese scientists covered
the sea floor with 97 ocean-bottom seismo-
meters. The campaign produced a torrent of
data, which took more than a year for a con-
tractor to stitch together. When the results
debuted at a conference in 2022, “people’s
jaws were on the floor,” Wallace says.
The 3D images revealed a 2-kilometer-tall
seamount wedged in the subduction zone,
4.5 kilometers below the sea floor. “We caught
it in the act of subducting,” Bangs says. “We
can see the structures it’s creating and how
it’s being accommodated.” Notable was not
only the seamount itself, but also the shadow
it cast. Like a lead blocker ahead of a running
back, the seamount shielded a pile of water-
logged seafloor sediments nearly 20 kilo-
meters long, keeping it from being scoured
away by the top plate. This water can lubri-
cate parts of the plate boundary at depths
it would otherwise be unable to reach, pre-
venting a full lock up, says Christine Chesley,
a marine geophysicist at the Woods Hole
Oceanographic Institution.
The sediment shadow is a “huge result,”
Chesley says. But it’s not the only way for sea-
mounts to smuggle water down to the depths.
Chesley participated in an electromagnetic
survey of Hikurangi in 2018 and 2019. The
team’s array of seafloor electrodes, sensi-
tive to the conductivity of water, captured a
subducting seamount that appeared to have
a conductive interior covered by a thin resis-
tive crust. That suggested the volcanic rocks
inside the seamount were also rich in wa-
ter, trapped by an imperme-
p
able cap, the team concluded.
“That stuff is wet and weak,”
says Ake Fagereng, a geologist
at Cardiff University.
Perhaps even more impor-
tant than the water carried by
g
seamounts are the slick sedi-
ments they shed as they slowly
erode over millions of years.
In 2018, Wallace co-led an off-
y
shore drilling expedition to Hi-
kurangi, which pulled up a lot
of clay sediments rich in the
mineral smectite, often used
by the oil and gas industry as
a lubricant in drilling muds. In
lab experiments meant to re-
create the pressure and heat
of the subduction zone, these
clays don’t grow stiff enough to lock and rup-
y g
ture in large earthquakes, but could host slow
slip, Wallace and her co-authors found in a
paper published earlier this year in Science.
Whether seamounts play a similar role in
other subduction zones remains to be seen.
Many of Hikurangi’s seamounts were once
,
exposed above sea level, which may have led
to more erosion and the creation of more
clays. The shallow subduction zone might
also encourage slow slip in a way that deeper
zones do not, Fagereng says.
This month, U.S. scientists are visiting
colleagues in Chile to discuss a planned
collaboration, called SZ4D, that would
study slow slip events in the Chilean sub-
duction zone. But that project, if funded, is
still years away. In the meantime, Bangs’s
3D seismic data are now open for other sci-
entists to use, and sensors in the boreholes
drilled at Hikurangi have captured several
new slow slip events. More insights are
coming, Bangs says. “It will really open our
eyes to what is happening here.” j
science.org SCIENCE

BIOSECURITY
Could chatbots help devise
the next pandemic virus?
An MIT class exercise suggests AI tools can be used
to order a bioweapon, but some are skeptical
By Robert F. Service
T
ech experts have been warning that
artificial intelligence (AI) could turn
against humanity by taking over every-
thing from business to warfare. Now,
Kevin Esvelt is adding another worry:
AI could help someone with no science
background and evil intentions order a virus
capable of unleashing a pandemic.
Esvelt, a biosecurity expert at the Mas-
sachusetts Institute of Technology, recently
asked students to create a dangerous virus
PHOTO: OTIS HISTORICAL ARCHIVES/NATIONAL MUSEUM OF HEALTH AND MEDICINE/WIKIMEDIA COMMONS
with the help of ChatGPT or other so-called
large language models, systems that can gen-
erate humanlike responses to broad ques-
tions based on vast training sets of internet
data. After only an hour, the class came up
with lists of candidate viruses and companies
that could synthesize their genetic code and
assemble the pieces.
Esvelt and others say the exercise, de-
scribed in an arXiv preprint posted on
6 June, underscores that AI systems may
soon allow nonscientists to design bioweap-
ons. Says Jaime Yassif of the Nuclear Threat
Initiative, a nonprofit focused on reducing
nuclear and biosecurity threats, “This is dra-
matically increasing the risk in ways that are
really alarming.” But at least one virologist
believes the concerns are overblown.
Biosecurity experts were already worried
that biology’s culture of openly exchang-
ing information, including virus sequences,
could be useful to bioterrorists. In principle,
papers describing a deadly extinct or circulat-
SCIENCE science.org
ing virus could provide a blueprint for a new
bioweapon. But to date, pulling this off has
required considerable expertise. The would-
be terrorist would need to identify a candi-
date virus as a starting point, synthesize the
viral genetic material, and mix it with other
reagents to “boot up” a virus capable of in-
fecting cells and reproducing.
All those steps are rapidly becoming easier,
Yassif says. For example, new benchtop DNA
printers might allow researchers to circum-
vent the screening that most synthetic bio-
logy companies now do to ensure no orders
include genetic material for potential bio-
weapons. Someone with malicious intent
could then send these genetic blueprints to
one of dozens of contract research companies
to be assembled into the target viruses.
AI could further lower the barriers,
Esvelt says. To investigate, he divided a class
of graduate students without life sciences
expertise into three groups, each with three
or four members. All groups had access to
GPT-4, Bard, and other AI chatbots, and were
given 1 hour to ask the chatbots to help them
design and acquire agents capable of causing
a pandemic.
Within the hour, the chatbots had sug-
gested four viruses: the 1918 H1N1 influenza
virus, a 2012 avian H5N1 influenza virus, the
smallpox virus variola major, and a strain of
the Nipah virus. In some cases, the chatbots
even pointed to genetic mutations reported
in the literature to increase transmission.
The AI engines also described techniques
to assemble a virus from its genetic sequence,
Biosafety experts warn that artificial intelligence
chatbots could make it easier for terrorists to launch
a pandemic as deadly as the 1918 flu outbreak.
as well as the necessary lab supplies and
companies that could provide them. Finally,
the chatbots even suggested companies that
might be willing to print genetic material
without screening it, and contract labs that
could help put the pieces together.
Esvelt doubts that these chatbots’ sugges-
tions pose much of a pandemic threat. Many
people, for example, have some level of im-
munity to previous pandemic flu viruses. And
variola’s large genome is difficult for even ex-
perts to assemble. (Before assigning it to his
class, Esvelt ran the experiment himself to
ensure it wouldn’t come up with truly threat-
ening suggestions, and he ran his plans by
other biosecurity experts.)
Yet Esvelt believes the experiment under-
p
scores how AI and other tools could make it
easier for would-be terrorists to unleash new
threats as the growing literature on biological
threats is incorporated into AI training data.
And Yassif notes that the technology will
likely be open source, so in everyone’s hands.
g
Restricting the information available to
chatbots and other AI engines could help,
Esvelt thinks. Among his proposals: exclud-
ing from training sets the very small number
y
of papers online that describe recipes for
creating and enhancing pathogens. Remov-
ing these papers, which Esvelt’s team esti-
mates make up less than 1% of the PubMed
abstracts database, “would suffice to elimi-
nate nearly all the risk,” they write in the
preprint. It would carry a cost, the authors
acknowledge—the AI engines could not use
these papers to advance biology in positive
ways—but the benefit of preventing misuse
y g
would be “practical and immediate.”
“In principle that is a very good sugges-
tion,” says Atoosa Kasirzadeh, an AI safety
expert at the University of Edinburgh. But
she notes, “At the moment we don’t have
good protocols to allow large language mod-
,
els to train on some parts of the internet and
not others.”
Other safeguards could include requir-
ing all DNA synthesis companies and future
benchtop DNA printers to screen genetic ma-
terial against known pathogens and toxins,
and requiring contract research organiza-
tions (CROs) to verify the safety of any genetic
material they are requested to assemble.
Virologist Gustavo Palacios of the Mount
Sinai Health System sees no cause for alarm.
He says booting up viruses is more challeng-
ing than described, and the idea that a CRO
that’s not some sort of rogue organization
will create a biological weapon is “preposter-
ous.” Still, Yassif concludes: “We need better
controls at all the chokepoints.” j
23 JUNE 2023 • VOL 380 ISSUE 6651 1211
NE WS
FEATURES
INTO THE
DARK
A European space telescope sets off to discover
dis
of dark energy—the biggest ingredient in the universe By Daniel Clery
W
hen the Euclid space tele-
scope blasts off from Cape
Canaveral in Florida early
next month, it will embark
on an unprecedented effort
to survey 1 billion galaxies—
and perhaps solve cosmo-
logy’s greatest mystery. The
search will cover more than
one-third of the sky and look back in time
to galaxies shining when the universe
was just one-quarter of its current age of
13.8 billion years. Although the task is im-
mense, Euclid’s primary goal is surprisingly
simple. The data it collects will be boiled
down to a single number, denoted by w.
And cosmologists are hoping, maybe even a
bit desperately, that it is not –1.
w describes the effect of dark energy, the
mysterious antigravitational force that is
accelerating the expansion of the universe.
All measures so far suggest that w is close
to –1. If it proves to be exactly that, it will
confirm the vanilla solution to dark energy:
that it’s a simple tweak—a cosmological
constant—added to Albert Einstein’s the-
1212 23 JUNE 2023 • VOL 380 ISSUE 6651
the nature
ory of gravity, which bestows empty space
with an innate springiness of its own. As
the universe expands, giving birth to more
space, the total amount of dark energy also
grows—so that the energy density always
remains constant.
That solution is anathema to cosmo-
logists because it is simply a fudge factor
that does not explain where dark energy
comes from and why it has that value. “If
w does equal –1, we still don’t know what
it is,” says astrophysicist Ofer Lahav of
University College London. A slight devia-
tion from –1 or, even better, a value of w
that changes with time could point cosmo-
logists toward a new, overarching theory
that would entail a fresh understanding
of physics. “The hope is to find a wrinkle,”
says Mark Halpern, a cosmologist at the
University of British Columbia.
The €1.4 billion Euclid mission, developed
by the European Space Agency (ESA), is not
tackling this problem alone. Many smaller
studies preceded it and Euclid is the first
of several multibillion-dollar dark energy
projects that will soon debut, including the
p
g
y
Vera C. Rubin Observatory, a U.S.-funded
survey telescope in Chile that will open
its eye in 2025, and NASA’s Nancy Grace
y g
Roman Space Telescope, to launch in 2026.
“The whole community is spending billions
of dollars to see if w is –1. If it is not, there
will be another Nobel Prize,” Lahav says.
At the very least, researchers hope the
telescopes will narrow a growing list of
,
alternatives to a cosmological constant,
including revisions of Einstein’s theory of
gravity and theories invoking a new physi-
cal force that could come in various flavors
and vary in time. “Every week there are new
theories,” says cosmologist Celia Escamilla-
Rivera of the National Autonomous Uni-
versity of Mexico. “But this soup of models
needs data.” For Jason Rhodes, an astro-
physicist at NASA’s Jet Propulsion Labo-
ratory (JPL), it’s an exciting time to be a
cosmologist. “We might be on the precipice
of discovering new physics.”
IT WAS 25 YEARS AGO that a team of as-
tronomers shocked the world with a report
that the expansion of the universe set in
science.org SCIENCE

motion by the big bang was not slowing
because of gravity, as everyone expected.
Instead, it was somehow speeding up. The
researchers relied on measurements of
50 supernovae of a particular kind, known
as type Ia: white dwarf stars that explode
with an inherently predictable brightness.
The team found that the most distant ones
were dimmer—and hence farther away—
than they should have been if the cosmic
expansion was steady or slowing down.
“The community was completely con-
fused,” Lahav remembers.
What came to be called dark energy
seemed to be at work. Since then, the evi-
dence for it has continued to mount, from
hundreds more supernovae, observations of
how galaxies clustered over time, and pat-
terns in the big bang’s afterglow, the cosmic
microwave background (CMB). A consensus
model, called lambda-CDM, has emerged in
which the universe has three components:
5% normal matter, mostly atoms; 27% cold
dark matter (CDM) made up of some as yet
undetected particle; and 68% dark energy
from a mysterious source.
SCIENCE science.org
Dark energy is speeding up the expansion of space,
and shaping the cosmic web of galaxies.
The Greek letter lambda in the cosmic
recipe represents one possible explanation,
the cosmological constant, which Einstein
himself had proposed in 1917. He was at-
tempting to fix a problem with his theory
of gravity, general relativity. At the time,
astronomers believed the universe was
static so, without something to oppose it,
gravity would pull the universe toward col-
lapse. Adding a constant amount of energy
to empty space did the job, but Einstein
was never comfortable with the idea. He
dropped it a little over a decade later when
Edwin Hubble and others showed that dis-
tant galaxies were in fact being flung apart
in the wake of the big bang.
Now, lambda is enjoying a second life,
but it remains a fudge factor without any
physical explanation, and one possible
basis for it doesn’t hold up. According to
quantum mechanics, the theory of the
atomic world, the vacuum should be abuzz
with energy from fluctuations that cause
particle-antiparticle pairs to constantly pop
in and out of existence. But the predicted
energy in that quantum fizz is much too big:
10120 (that’s 1 followed by 120 zeros) times
the value of lambda astronomers have ob-
served. “It’s incompatible with us existing,”
says Ue-Li Pen, a theoretical astrophysicist
at the Perimeter Institute. “The universe
would blow up.”
Cosmologists describe space as a perfect
fluid and w is defined as the ratio of the flu-
id’s pressure to its energy density. A negative
value indicates an outward pressure, and
a value of exactly –1 signifies that the pres-
sure is a constant, unchanging feature of the
cosmos. A deviation from –1 would point to
a dark energy density that is growing or de-
clining with time—and a universe that could
end up accelerating even faster, or eventu-
ally start contracting. But all estimates so
far—from the CMB and the universe’s earli-
est days, and from its most recent few billion
years—suggest that w is close enough to –1
for it to sit within the error bars.
To get a more precise reading, astrono-
mers want to fill in the middle part of cos-
mic history, in particular the period more
than 7 billion years ago when the universe
was smaller and gravity still dominated
over dark energy. They want to observe the
transition from deceleration to acceleration,
but have not had the tools to look back that
far. Earth’s atmosphere limits ground-based
telescopes from making sufficiently precise
measurements beyond about 3 billion years
into the past. “We have to get this right, and
you can only do it from space,” says Euclid
Project Scientist René Laureijs at ESA.
N E WS
IN 2007, ESA RECEIVED proposals for two mis-
sions that, using different techniques, would
take the search for dark energy into space for
the first time. One proposal, called the Dark
Universe Explorer, would rely on a technique
called weak gravitational lensing: tiny dis-
tortions in the shapes of distant galaxies that
result as the gravity of intervening matter
bends their light. Unlike strong gravitational
lensing, where one sees smeared arcs or cop-
ies of celestial objects, these distortions are
imperceptible to the eye; it is only by statisti-
cally comparing large samples of galaxy im-
ages that weak lensing can be measured.
Those measurements give astronomers
a handle on the clumpiness of matter be-
tween the imaged galaxies and Earth. And
by making those measurements for galax-
ies at different distances, they can see how
the fight between gravity and dark energy
evolved over time.
p
The second proposal, called the Spectro-
scopic All-Sky Cosmic Explorer (SPACE),
would exploit so-called baryon acoustic os-
cillations (BAOs). These originated soon
after the big bang when the universe was
a roiling soup of gas and photons. Denser
g
clumps of matter emitted pressure waves
through the soup, concentrating matter in
ripples around each clump. These ripples
can be detected statistically in the CMB, and
y
today their imprint can be seen as a peak
in the distribution of galaxy separations at
490 million light-years.
By mapping millions of galaxies and
calculating the size of this standard yard-
stick at different times, astronomers can
chart the expansion rate and see the effect
of dark energy. The galaxy distribution “is
a cosmological laboratory,” says Andrea
Cimatti of the University of Bologna, who
y g
led the SPACE proposal. “But to achieve the
accuracy needed requires a very large set of
data, a very large volume of the universe.”
With two compelling cosmology mis-
sions to choose from, ESA suggested merg-
ing them onto a single spacecraft and, in
,
2011, Euclid was approved. But it wasn’t
an easy marriage. “It was complicated,”
says John Peacock of the University of Ed-
inburgh, one of the founding fathers of
the Euclid mission. “There were a pile of
experimental trade-offs.”
For example, there were competing de-
mands for the light from Euclid’s 1.2-meter
main mirror. The weak lensing survey re-
quires visible light to make sharp images,
whereas the BAO survey relies on infrared
light to map more distant galaxies, whose
light is “redshifted” by the expansion of the
universe. Euclid managers commissioned
optics-maker Zeiss to cast a complex, layered
piece of glass that allows infrared light to
pass through to one set of instruments while
23 JUNE 2023 • VOL 380 ISSUE 6651 1213

Clashing forces
mold the cosmos
The expansion of the
universe has been shaped
by dark matter and dark
energy, both poorly
understood. Early on, dark
matter’s gravity slowed
cosmic growth. Later, as
galaxies spread apart,
dark energy began to
accelerate the expansion.
The Euclid mission will
chart the changing
expansion rate further
back in time than ground-
based telescopes.
Probing dark energy
Dark energy has been quantified by studying radiation left from the big bang and mapping galaxies in more
recent times. But astronomers want to know what happened in between. They rely on three main techniques.
Cosmic distance ladder
Supernovae with predictable
brightness provide distances
to galaxies. Their Doppler-
shifted light shows how
fast those galaxies recede,
revealing the cosmic
expansion rate.
Galactic vibrations
The seeds of galaxies cast
ripples in the gas of the
early universe. These ripples
have grown with cosmic
expansion, providing a
yardstick to measure it.
A distorting lens
Measuring how dark
matter distorts images of
distant galaxies reveals the
clumpiness of matter and
how it has evolved. Dark
energy acts as a brake on
growing clumpiness.
What lies ahead
Determining dark energy to a level of 1% or better could help theorists explain its origin.
Contending theories fall into three main categories.
Constant push of
Distance
vacuum energy
Present
Time
Cosmological constant
The simplest explanation for
dark energy fits current data.
The vacuum has an innate
antigravitational energy that
grows as more space is spawned,
driving cosmic acceleration.
Acceleration from
dark energy
Deceleration reflecting visible light along a different path
from dark matter to another set of instruments.
Engineers also elected to build Euclid’s
mirrors and other components from sili-
Big con carbide, a material used in car brakes
bang
and bulletproof vests, because temperature
swings don’t alter its shape—critical if per-
fect weak lensing pictures are to be taken
reliably over a 6-year survey. But silicon car-
bide is difficult to work with: Components
must be molded from a powder and baked
Telescope reach into ceramic. A baseplate that holds the tele-
Euclid scope’s detectors was especially tricky, says
Euclid Project Manager Giuseppe Racca at
Ground-based
ESA. Technicians kept finding tiny cracks,
13.8 billion 10 billion 5 billion Present and casting a flawless one took 2 years lon-
years ago years ago years ago ger than planned. “We had two or three fail-
ures,” Racca says.
The BAO survey team had to make some
compromises. To measure precise distances
to galaxies, researchers need accurate red-
p
shifts that can only be gleaned from spec-
tra. The team had wanted to use an array
Supernova
Sup of thousands of tiny switchable mirrors to
Earth deflect light from individual galaxies into a
light-splitting spectrograph, gathering spec-
Distance
Distan
Dis c
too Earth
Eartt tra from many galaxies at once. But ESA
g
deemed the device too risky as it had never
been tested in orbit. Instead, the survey re-
lies on a grating prism, or grism, to spread
out the light from every source in the field
y
of view and record all the spectra simultane-
Baryon ously with a camera. “It’s simple and brutal,”
acoustic Peacock says. But that approach requires
oscillation
complex image processing to untangle over-
lapping spectra and remove spectra from
Cosmic microwave
background
unwanted objects, such as foreground stars.
The detectors for the infrared camera also
Warped light path proved to be a headache. Euclid’s designers
wanted to use detectors made by Teledyne
Earth
Technologies, in California, but because they
y g
have military applications, they are subject to
U.S. export controls. NASA played the inter-
mediary, acquiring the sensors and carrying
Distant galaxies Dark matter halo Distorted
out testing and packaging at JPL in return
galaxy
images for more than 100 seats for U.S. researchers
in the Euclid Consortium, the 2000-strong
,
body that built the instruments and will pro-
cess the data. But the extra bureaucracy was
Big rip burdensome. “Even talking to my European
Local colleagues, I needed to get the conversation
pull
cleared,” Rhodes says.
Distance
By 2022, the spacecraft was almost com-
Big plete. Then Russia invaded Ukraine and
crunch Western sanctions scuttled Euclid’s sched-
uled launch on a Russian Soyuz rocket.
Cosmic
Spacecraft are designed to withstand the
Time push
launch stresses of their chosen rocket, so
Quintessence Modified gravity switching is not simple. But by late last
If the acceleration is not If gravity at cosmic year, ESA had found a suitable alterna-
constant, dark energy could be scales works differently tive in SpaceX’s Falcon 9. “It was a period
a fifth fundamental force that from in the local
is sometimes attractive and universe, the need for
of anxiety and uncertainty,” says Bhuvnesh
sometimes repulsive, leading to a dark energy and dark Jain, a cosmologist at the University
variety of fates for the universe. matter can disappear. of Pennsylvania.
science.org SCIENCE

EUCLID NOW SITS in a Florida cleanroom,
ready for its early-July launch and month-
long journey to L2, a gravitational balance
point 1.5 million kilometers from Earth
where NASA’s JWST space telescope also
resides. Euclid’s U.S. rivals, the Roman tele-
scope and the Rubin observatory, will join
the dark energy hunt later. The three proj-
ects will look at different parts of the sky, at
different wavelengths, and with overlapping
combinations of the three primary tech-
niques (weak lensing, BAO, supernovae).
Roman, with a larger mirror than Euclid, will
peer further into the past but over a smaller
sky area. The ground-based Rubin will not
see as far as the other two, but its optical and
infrared survey will be the widest.
Each team wants to be first to solve the
dark energy mystery, but they are likely to
need supporting evidence from
one another to convince the
community that any discovery
is real, Rhodes says. “Ten years
down the road, joint analyses of
data from all three—analyzing at
pixel level in clever ways—will
put the most compelling con-
straints on dark energy.”
Theorists have reason to hope
that the data will quickly chal-
lenge the current picture, noting
that current observations offer
two clues that all is not right
with lambda-CDM. The first is a
discrepancy in measurements of
the Hubble Constant, H0, which
describes how fast the universe
is expanding. Studies of the CMB
allow researchers to calculate
that today a galaxy 1 megaparsec
(3.26 million light-years) from
Earth should be receding at In a 6-year survey, Euclid’s 1.2-meter main mirror will scan more than one-third
68 kilometers per second (km/s), of the sky and look back 10 billion years. It aims to record 10 billion objects.
assuming the particular mix of
matter, dark matter, and dark energy pre-
scribed by lambda-CDM. But when astrono-
mers measure the actual recession velocity
of nearby galaxies, using type Ia supernova
distances and other techniques, they clock
a speed of 73 km/s. Researchers initially as-
sumed it was simple experimental error—
these are hard things to measure—but as
methods were refined and new techniques
added, the error bars shrank but the so-
called Hubble tension remained.
A second tension is over an independent
parameter called S8, which describes the
PHOTO: STEPHANE CORVAJA/ESA
clumpiness of matter in the cosmic web
of galaxies. Again, given the lambda-CDM
recipe, measures of clumpiness in the CMB
predict a value of S8 that we should see today.
But the nearby universe actually appears to
be 10% less clumpy than the prediction. “You
should be able to draw a line between the
SCIENCE science.org
two but when you draw the line, it doesn’t
match up,” Rhodes says. The gap is another
chink in the armor of lambda-CDM.
To resolve the Hubble tension, theorists
have concocted a raft of models, collectively
known as “early dark energy.” They suggest
that a form of dark energy gave the uni-
verse an early push before the creation of
the CMB. This would raise the value of H0
derived from CMB data. Escamilla-Rivera
considers it a “very good candidate.” Recent
observations of the way the ancient light of
the CMB is polarized—a tendency to vibrate
in certain directions—have hinted at that
early growth spurt. But it wouldn’t resolve
the S8 tension, and the evidence is nowhere
near firm enough to convince Jain. “Dark
energy appeared, disappeared, reappeared
again—it doesn’t sound pretty,” he says.
Early dark energy, like many other alter-
natives to the cosmological constant, re-
quires an entirely new force. Some theorists
describe the force as a field or fluid pervad-
ing the whole universe and give it the name
quintessence. The key to quintessence is
that it changes in value through cosmic
history and can be attractive or repulsive.
It might have switched from attraction to
repulsion about 10 billion years ago, gradu-
ally pushing cosmic expansion toward ac-
celeration. It could be fickle again: It might
cause the expansion to grow at a faster than
exponential rate, tearing apart the cosmos
in a “big rip,” or flip into an attractive force
pulling the universe into a big crunch.
Data from Euclid and other dark energy
probes could boost this picture. If they can
detect even a tiny deviation in w away from
–1—evidence that dark energy is not con-
N E WS | F E AT U R E S
stant in time—that would be a nail in the
coffin of the cosmological constant. “The
key goal we all want to know is: Does dark
energy evolve?” Peacock says.
Another possibility is that the theory of
gravity needs a revamp. General relativity
solves puzzles near and far, from the orbit of
Mercury to gravitational lensing by galaxy
clusters, but it has never been tested across
truly cosmic distances, where some think
it might break down. In addition to mea-
suring the BAO yardstick, Euclid’s spectro-
scopic survey can chronicle how gravity
pulled clusters of galaxies into sheets and fil-
aments over time, forming the cosmic web.
If the growth of these structures deviates
from the predictions of general relativity, it
could indicate that Einstein’s theory itself
needs revision, eliminating the need for
dark energy. “If Einstein’s the-
ory is incorrect, there’s a huge
p
industry of alternative theories,”
Peacock says.
AFTER ARRIVAL at L2 and a few
months of calibration, Euclid
will begin its survey, covering
g
36% of the sky while avoiding
areas crowded with the stars
and gas of the Milky Way and
looking back 10 billion years. In
y
just 2 days it will survey as much
of the sky as the Hubble Space
Telescope has in more than
3 decades. It will record an esti-
mated 10 billion galaxies, stars,
and Solar System objects, identi-
fying new targets for JWST and
large ground-based telescopes
to inspect. “It’s a fantastic gold
mine for any kind of science,”
y g
says astrophysicist Yannick
Mellier of the Astrophysics In-
stitute of Paris and leader of
the Euclid Consortium.
Out of that flood, the Euclid Consortium
will analyze images of 1 billion galaxies for
,
weak lensing and spectra from tens of mil-
lions of galaxies to fix their positions for
the BAO survey. The aim is to calculate the
value of w with an accuracy of better than
1%, an order of magnitude better than cur-
rent estimates.
Humbled that they haven’t figured out
what makes up more than two-thirds of the
universe, cosmologists are now collectively
holding their breath. “We still don’t know
very much. There’s no evidence that it’s not
a cosmological constant,” says cosmologist
Jo Dunkley of Princeton University. “In
5 years, I hope we’ll have firm evidence that
lambda-CDM is broken in some way.” With
luck, she may also have a glimmer of what
might take its place. j
23 JUNE 2023 • VOL 380 ISSUE 6651 1215
 INSIGHTS
PERSPECTIVES
MATERIALS SCIENCE

A bendable
biological ceramic
A fatigue-resistant
mollusk hinge could

p
inspire the development
of new materials

g
y
y g
,

The freshwater mussel

Cristaria plicata has a
fatigue-resistant hinge that,
when the valves are closed,
stores elastic energy to
allow them to spring open.

1216 23 JUNE 2023 • VOL 380 ISSUE 6651 science.org SCIENCE

 By Rachel L. Crane1 and Mark W. Denny2 can live hundreds of years, so fatigue is a by the outer ligament requires the folding
potential danger. fan region to deform circumferentially but

F
rom a catastrophic bridge collapse to
broken industrial equipment to the
mundane snapping of a plastic latch,
structures can break when their com-
ponents fail because of fatigue—dam-
age accumulation caused by repeated
stresses. Fatigue is a problem not just for
artificial structures but also for organisms.
Running, jumping, chewing, flying: Many
of life’s activities involve repeated loads
that can cause fatigue failure (1), leading
to injury or death. This places a high evo-
lutionary pressure on avoiding and repair-
ing fatigue-caused damage. Biomineralized
tissues—such as bone, teeth, and mollusk
shell—are often notably tough, fatigue-re-
sistant structures that are built primarily
from brittle ceramic components and thus
Meng et al. show that the hinge of C. pli-
cata can withstand 1,500,000 typical load-
ing cycles (equivalent to one cycle a minute
for almost 3 years) without fatigue damage,
and they reveal how this fatigue resistance
works. The hinge is a thick, semicircular
arch with an inner portion that forms a
“folding fan region” and an elastic “outer
ligament” along its peripheral edge. At each
end, the arch is connected to a rigid foun-
dation extending from one of the valves. As
the mussel’s muscles contract and the valves
close, the arch’s foundations rotate. As the
foundations rotate, the arched folding fan
region maintains its dimensions radially
but deforms circumferentially, with the in-
ner edge compressing and the outer edge
expanding (see the figure). This stretches
maintain form radially, and both of these
features must be fatigue resistant.
Materials fatigue when, with loading,
small flaws grow into propagating cracks. If,
after repeated load cycles, the cracks extend
across a structure, it breaks catastrophi-
cally. Using brittle calcium carbonate crys-
tals in an organic matrix, mollusks (includ-
ing bivalves) build shells that resist fatigue
through multiscale strategies that slow the
spread of accumulating damage (3, 6). For
example, nacre (mother-of-pearl, a layer in
many mollusk shells) is remarkably tough
and strong because the brick-and-mortar
structure of its aragonite crystals deflects
propagating cracks through the matrix, and
features of adjacent crystals such as micro-
bridges and waviness cause them to inter-

have provided inspiration to materials
scientists seeking to overcome the usual
trade-off between strength and tough-
ness (2–4). Unlike these relatively rigid
structures, on page 1252 of
this issue, Meng et al. (5)
the outer ligament and allows it to act as
a spring, storing elastic energy that returns
the hinge to its initial configuration when
the muscles relax. Thus, energy storage
p
lock as they pull apart (6, 7).
The arch of the hinge of C. plicata is also
constructed from calcium carbonate crystals
in an organic matrix, but unlike shell, it is
flexible, and the fatigue-resis-
tance mechanisms reported

reveal multiscale, fatigue-
resistance mechanisms of a
robust, bendable biominer-
alized structure: the hinge
g
A multiscale deformable ceramic by Meng et al. operate pri-
Bivalves use the muscles inside their shell to clamp shut (red arrows), but they depend marily not by limiting dam-
on the hinge to elastically reopen them (blue arrows). As the valves close, the hinge age but by avoiding damage
deforms, stretching the outer ligament across the radially rigid and fracture-resistant in the first place. The fold-

of a freshwater mussel,
Cristaria plicata.
Mussels and other bivalves
(such as oysters, clams, and
scallops) build an external
shell comprising two domed
valves that are joined by a
hinge, which faces a consid-
erable structural challenge.
The valves open for feeding
y
folding fan region. ing fan region is composed
of ceramic aragonite nano-
Valves open wires oriented radially and
Outer ligament tightly packed in the organic
Bivalve shell matrix. The flexible matrix
contributes to circumferen-
Folding fan tial deformation, and the
region
Hinge nanowires provide the radial
rigidity to support the outer
Connection to valve
ligament. However, repeated

y g
and reproduction and close compressive loading of the
when the animal is threat- Valves closed nanowires introduces a po-
ened. Using muscles inside tential problem: How do the
the shell, a bivalve can pull stiff, thin nanowires avoid
CREDITS: (PHOTO, OPPOSITE PAGE) X.-S. MENG ET AL. (5); (GRAPHIC) K. HOLOSKI/SCIENCE

the valves together, safely en- breaking? The organic ma-

closing the organism’s body. trix helps prevent nanowire

However, because muscles
cannot push, the animal
cannot use them to force
the valves open. Instead, the
hinge stores elastic energy
when the valves are closed so
that when the animal relaxes
its muscles, elastic rebound
causes the valves to reopen.
Bivalves open and close their
valves repeatedly throughout
the day and, in some species,
1
University of California, Davis, Davis, protected from fracture
CA, USA. 2Stanford University, Stanford,
CA, USA. Email: rlcrane@ucdavis.edu;
mwdenny@stanford.edu
,
Muscle bending and reduces shear
stress caused by nanowires
sliding. Additionally, the
nanowires are aligned along
a specific crystallographic
axis that confers several
Fatigue resistance benefits: It is the stiffest
The folding fan region orientation, it is the axis
comprises rigid aligned Twin of fastest growth (8), and it
ceramic aragonite Organic boundary is associated with the forma-
nanowires that are matrix
embedded in an organic
tion of twin boundaries, so
Nanowire that many of the nanowires
matrix that protects the
nanowires from breaking. Aligned are formed as twinned crys-
The nanowires are further crystallo- tals joined along one face,
graphic
because they are aligned
a configuration that pro-
orientation
crystals (black arrows) that tects against bending and
often form twinned pairs. fracture (9).

SCIENCE science.org 23 JUNE 2023 • VOL 380 ISSUE 6651 1217

I NS I GHTS | P E R S P E C T I V E S

By integrating principles across scales— IMMUNOLOGY

from the overall structure of the hinge to
the atomic structure of individual crystals—
Meng et al. reveal how nature creates a fa-
tigue-resistant, bendable, elastic structure
Channeling antigens
primarily from brittle components. These
cross-scale principles require precision at
the finest scales, and the cellular and mo-
to CD8+ T cells
lecular mechanisms by which mollusks de- Perforin-2 facilitates antigen translocation
posit shell with such precision is an area of to the cytosol in cross-presenting dendritic cells
ongoing exploration (7, 10). It is intriguing
to compare these processes in the valves
and hinge because they have similar com- By Kavita Rawat and Claudia V. Jakubzick conventional dendritic cells (cDCs), which
ponents but different structures (11). bridge innate and adaptive immunity by cap-

Matching biological fine-scale control
is a particular challenge for human en-
gineers interested in bioinspired materi-
als, as evidenced by the difficulties faced
in developing composites that mimic the
strength and toughness of nacre (7, 12, 13).
As a preliminary proof of concept, Meng et
C
ytotoxic CD8+ T cells (CTLs) can kill
any virus-infected cell or tumor cell,
providing they display antigenic pep-
tides on major histocompatibility
complex class I (MHC-I) molecules.
However, tumor cells or virus-infected
cells cannot present antigens in a sufficiently
turing exogenous cell-associated antigens on
MHC-I molecules and presenting them to
naïve CD8+ T cells, while also providing co-
stimulation (1). This process is called cross-
presentation (2). Despite the importance of
cross-presentation for antitumor and antivi-
ral immunity and vaccine design, the mecha-

al. simulated the use of brittle nanowires
aligned in a resilient matrix by embedding
glass fibers in a matrix of polydimethylsi-
loxane (PDMS) polymer. How to best ap-
ply the full multiscale principles that they
develop, though, remains an ongoing ques-
p
stimulating manner to also prime naïve CD8+ nisms remain unclear. On page 1258 of this
T cells to become CTLs. This is the role of issue, Rodríguez-Silvestre et al. (3) report that
Coordinating phagocytosis and antigen presentation
In cross-presenting conventional dendritic cells (cDC1s), exogenous cell-associated antigens are recognized

tion. Furthermore, these principles were
developed from a single species with a
specific natural history and evolutionary
trajectory. What can be learned by exam-
g
by CLEC9A. Upon phagocytosis, pH modulation by NOX2, ROS production, and aquaporin-3 facilitate perforin-2
maturation and phagosome-to-cytosol antigen transport. This leads to perforin-2–mediated localized
rupture of the phagosome membrane through lipid peroxidation; ESCRT-III may repair ruptured phagosomes.
Cytosolic antigens use classical MHC-I pathway molecules for cross-presentation (ubiquitination, proteasome

y
ining species with different demands and degradation, and TAP transport into the endoplasmic reticulum).
constraints? It is unknown whether fa-
tigue becomes more relevant in a longer- MHC-I antigen Conventional
lived species, or whether calcium carbon- presentation dendritic cell
ate availability affects hinge structure in
saltwater bivalves. Scallop hinges undergo Exogenous antigen (dead cells) with
CLEC9A exposed F-actin and myosin II
high-frequency deformations as the animal
swims (14); do their fatigue characteristics
differ in revealing ways? Although the me- SYK
chanical properties Meng et al. examine Ub Proteasome

y g
match the needs of this particular organ- Ub
Ub
ism, there is exciting promise in how these
Ub
principles might be refined with a broader
phylogenetic scope. j
Phagosome Perforin-2
R E F E R E N C ES A N D N OT ES ? maturation Golgi
apparatus

,
1. J. M. Gosline, Mechanical Design of Structural Materials NOX2 ROS pH?
in Animals (Princeton Univ. Press, 2018). Ub
H2O2 Aquaporin-3 H2O2 Lipid Ub
2. J. W. Pro, F. Barthelat, MRS Bull. 44, 46 (2019). peroxidation
3. R. L. Crane, M. W. Denny, J. Exp. Biol. 223, jeb220277
(2020).
? Ub
4. D. Arola et al., Int. J. Fatigue 32, 1400 (2010).
5. X.-S. Meng et al., Science 380, 1252 (2023).
6. H. Kakisawa, T. Sumitomo, Sci. Technol. Adv. Mater. 12, ESCRT-III
064710 (2011). for repair
7. L.-B. Mao et al., J. Am. Chem. Soc. 144, 18175 (2022).
8. A. G. Checa, T. Okamoto, J. Ramírez, Proc. R. .Soc. B 273, Nucleus
1329 (2006).
9. Y. Tian et al., Nature 493, 385 (2013).
10. M. S. Clark, J. Exp. Biol. 223, jeb206961 (2020). TAP
11. K. Kubota et al., J. Struct. Biol. 199, 216 (2017).
sin
12. L.-B. Mao et al., Science 354, 107 (2016). CLEC9A, C-type lectin domain pa b2 microglobulin
containing 9A; ESCRT-III, endosomal Ta
13. M. Grossman et al., Adv. Mater. 29, 1605039 sorting complex required for transport III; Calreticulin
(2017). MHC-I, major histocompatibility complex class I;
14. M. Denny, L. Miller, J. Exp. Biol. 209, 4503 (2006). NOX2, NADPH oxidase 2; ROS, reactive oxygen MHC-I
species; TAP, transporter associated with antigen
processing; Ub, ubiquitin. Endoplasmic reticulum
10.1126/science.adi5939

1218 23 JUNE 2023 • VOL 380 ISSUE 6651 science.org SCIENCE

I NS I GHTS | P E R S P E C T I V E S

By integrating principles across scales— IMMUNOLOGY

from the overall structure of the hinge to
the atomic structure of individual crystals—
Meng et al. reveal how nature creates a fa-
tigue-resistant, bendable, elastic structure
Channeling antigens
primarily from brittle components. These
cross-scale principles require precision at
the finest scales, and the cellular and mo-
to CD8+ T cells
lecular mechanisms by which mollusks de- Perforin-2 facilitates antigen translocation
posit shell with such precision is an area of to the cytosol in cross-presenting dendritic cells
ongoing exploration (7, 10). It is intriguing
to compare these processes in the valves
and hinge because they have similar com- By Kavita Rawat and Claudia V. Jakubzick conventional dendritic cells (cDCs), which
ponents but different structures (11). bridge innate and adaptive immunity by cap-

Matching biological fine-scale control
is a particular challenge for human en-
gineers interested in bioinspired materi-
als, as evidenced by the difficulties faced
in developing composites that mimic the
strength and toughness of nacre (7, 12, 13).
As a preliminary proof of concept, Meng et
C
ytotoxic CD8+ T cells (CTLs) can kill
any virus-infected cell or tumor cell,
providing they display antigenic pep-
tides on major histocompatibility
complex class I (MHC-I) molecules.
However, tumor cells or virus-infected
cells cannot present antigens in a sufficiently
turing exogenous cell-associated antigens on
MHC-I molecules and presenting them to
naïve CD8+ T cells, while also providing co-
stimulation (1). This process is called cross-
presentation (2). Despite the importance of
cross-presentation for antitumor and antivi-
ral immunity and vaccine design, the mecha-

al. simulated the use of brittle nanowires
aligned in a resilient matrix by embedding
glass fibers in a matrix of polydimethylsi-
loxane (PDMS) polymer. How to best ap-
ply the full multiscale principles that they
develop, though, remains an ongoing ques-
p
stimulating manner to also prime naïve CD8+ nisms remain unclear. On page 1258 of this
T cells to become CTLs. This is the role of issue, Rodríguez-Silvestre et al. (3) report that
Coordinating phagocytosis and antigen presentation
In cross-presenting conventional dendritic cells (cDC1s), exogenous cell-associated antigens are recognized

tion. Furthermore, these principles were
developed from a single species with a
specific natural history and evolutionary
trajectory. What can be learned by exam-
g
by CLEC9A. Upon phagocytosis, pH modulation by NOX2, ROS production, and aquaporin-3 facilitate perforin-2
maturation and phagosome-to-cytosol antigen transport. This leads to perforin-2–mediated localized
rupture of the phagosome membrane through lipid peroxidation; ESCRT-III may repair ruptured phagosomes.
Cytosolic antigens use classical MHC-I pathway molecules for cross-presentation (ubiquitination, proteasome

y
ining species with different demands and degradation, and TAP transport into the endoplasmic reticulum).
constraints? It is unknown whether fa-
tigue becomes more relevant in a longer- MHC-I antigen Conventional
lived species, or whether calcium carbon- presentation dendritic cell
ate availability affects hinge structure in
saltwater bivalves. Scallop hinges undergo Exogenous antigen (dead cells) with
CLEC9A exposed F-actin and myosin II
high-frequency deformations as the animal
swims (14); do their fatigue characteristics
differ in revealing ways? Although the me- SYK
chanical properties Meng et al. examine Ub Proteasome

y g
match the needs of this particular organ- Ub
Ub
ism, there is exciting promise in how these
Ub
principles might be refined with a broader
phylogenetic scope. j
Phagosome Perforin-2
R E F E R E N C ES A N D N OT ES ? maturation Golgi
apparatus

,
1. J. M. Gosline, Mechanical Design of Structural Materials NOX2 ROS pH?
in Animals (Princeton Univ. Press, 2018). Ub
H2O2 Aquaporin-3 H2O2 Lipid Ub
2. J. W. Pro, F. Barthelat, MRS Bull. 44, 46 (2019). peroxidation
3. R. L. Crane, M. W. Denny, J. Exp. Biol. 223, jeb220277
(2020).
? Ub
4. D. Arola et al., Int. J. Fatigue 32, 1400 (2010).
5. X.-S. Meng et al., Science 380, 1252 (2023).
6. H. Kakisawa, T. Sumitomo, Sci. Technol. Adv. Mater. 12, ESCRT-III
064710 (2011). for repair
7. L.-B. Mao et al., J. Am. Chem. Soc. 144, 18175 (2022).
8. A. G. Checa, T. Okamoto, J. Ramírez, Proc. R. .Soc. B 273, Nucleus
1329 (2006).
9. Y. Tian et al., Nature 493, 385 (2013).
10. M. S. Clark, J. Exp. Biol. 223, jeb206961 (2020). TAP
11. K. Kubota et al., J. Struct. Biol. 199, 216 (2017).
sin
12. L.-B. Mao et al., Science 354, 107 (2016). CLEC9A, C-type lectin domain pa b2 microglobulin
containing 9A; ESCRT-III, endosomal Ta
13. M. Grossman et al., Adv. Mater. 29, 1605039 sorting complex required for transport III; Calreticulin
(2017). MHC-I, major histocompatibility complex class I;
14. M. Denny, L. Miller, J. Exp. Biol. 209, 4503 (2006). NOX2, NADPH oxidase 2; ROS, reactive oxygen MHC-I
species; TAP, transporter associated with antigen
processing; Ub, ubiquitin. Endoplasmic reticulum
10.1126/science.adi5939

1218 23 JUNE 2023 • VOL 380 ISSUE 6651 science.org SCIENCE

 endosomal perforin-2 is a channel-forming tive to NOX2 couldco be aquaporin-3. Mice
protein that allows the release
ase of internal- lacking this channel
ch exhibit a reduced
ized exogenous cell-associated d antigens to ability to m mount antiviral responses
the cytosol of cross-presenting
ng cDCs. owing to decreased
dec cross-presentation.
A fundamental yet unresolved ved question Aquaporin-3 directly transports hydro-
is how exogenous antigens, whichhich enter gen peroxid
peroxide into endosomes, which
the cross-presenting DCs (referred
referred could contribute
co to endosomal
to as cDC1) through endosomal omal lipid peroxidation.
p However, a
pathways, are trimmed into to maladaptive
ma side effect is
peptides in the cytosol and d compromised
co membrane
then access MHC-I, which integrity
in and the release of
is primarily loaded with pep-- endosomal
en cargo (9). How
tides in the endoplasmic re- e- aquaporin-3
aqu influences the
ticulum (ER). One theory posits osits maturation
matu of perforin-2, along
that all processing and loadingading with the
th other mediators, will re-
might occur within vacuoles when quire further
f investigation.
MHC-I molecules are recycled, cycled, Toll-like
Toll- receptor (TLR) ligation
thus circumventing the problem lem of enhances
enhanc cross-presentation and
exogenous antigen transport rt from promotes
promote fusion events between
phagosome to cytosol (P2C). Another, phagosomes
phagosom and MHC-I–containing
more widely supported theory ry proposes P2C recycling endoso
endosomes (10, 11). Rodríguez-

shuttling of antigens, followed by protea-
somal protein processing in the cytosol and
transporter associated with antigen process-
ing (TAP)–dependent entry of peptides into
the ER for final processing and loading on
MHC-I (4, 5). The P2C pathway has been
Schematic of a hexadecameric perforin-2 pore,
which forms in phagosome membranes to facilitate
the transportation of antigens to the cytosol.
Rodríguez-Silvestre et al. used dead cells as
an exogenous antigen source in their study,
p
Silvestre et al. demonstrate that TLR stimula-
tion, particularly TLR9 and TLR11, enhanced
perforin-2 maturation, outlining another po-
tential mechanism by which TLR enhances
cross-presentation. Together, it is possible
that TLR agonists boost perforin-2 proteo-

shown to depend either on phagosome-ER
fusion and recruitment of ER retrotransloca-
tion channel machinery, or highly localized
endosomal vesicle damage and rupture (4, 5).
it becomes intriguing to further explore the
interplay between perforin-2 and the cDC1-
specific protein C-type lectin domain con-
taining 9A (CLEC9A), which plays a role in
g
lytic maturation and function through the
induction of lipid peroxidation, resulting
in increased P2C antigen release and cross-
presentation. If this newly discovered P2C

Perforin-2 is a pore-forming antibacterial
protein that is essential in innate immunity.
Perforin-2 shares similarities with the pore-
forming membrane attack complex (MAC)
of complement and with perforin-1 (6).
Perforin-1, which is released by CTLs, targets
the outer cellular membrane, resulting in
cytotoxicity. In macrophages, perforin-2 can
lead to pH-dependent pore formation, which
kills bacteria in phagolysosomes. Rodríguez-
the uptake and subsequent signaling of ne-
crotic cells in cDC1.
Rodríguez-Silvestre et al. demonstrate that
perforin-2 maturation is required for pore
formation on the endosomal membrane and
is pH dependent. However, the pH of cDC1
phagosomes is mostly neutral, so additional
mediators are required to regulate pH in a
timely manner and prevent antigen degrada-
tion after uptake. Perhaps perforin-2 matura-
y
molecular pathway is important, viruses
might have found ways to undermine it. In
the future, it will be intriguing to delve into
the role of perforin-2 in disease models, par-
ticularly in the context of cancer or viral in-
fections. Additionally, investigating whether
other antigen-presenting cells, such as TLR7-
stimulated cDC2 and monocytes, use per-
forin-2 for cross-presentation could reveal
conserved function (12, 13). Lastly, exploring

Silvestre et al. show that perforin-2 pore
formation also occurs in phagosomal mem-
branes in cDC1, allowing the regulated re-
lease of exogenous antigens into the cytosol
while preserving the integrity of the vesicle
(see the figure). Their findings indicated that
tion depends on delivery of cargo by CLEC9A,
which recognizes F-actin and associated
myosin on necrotic cells (7). Additionally,
NADPH oxidase 2 (NOX2) is triggered by
CLEC9A, which generates reactive oxygen
species (ROS) and regulates phagosomal pH.
y g
how the nature of antigens affects the mecha-
nisms of perforin-2–mediated functions has
the potential to reveal missing links within
the P2C pathway. Such investigations could
greatly enhance our understanding of im-
mune responses and make valuable contribu-

the absence of perforin-2 in cDC1 resulted
in a partial decrease in cross-presentation
of cell-associated antigens, whereas no sig-
nificant difference was observed in cross-
presentation of soluble antigens. The de-
creased presentation of cell-associated anti-
gens by perforin-2–deficient cDC1s resulted
from impaired antigen shuttling to the cyto-
IMAGE: P. RODRÍGUES-SILVERTRE ET AL. (3)
Hence, it is critical to comprehend the timing
of the pH shift that enables perforin-2 matu-
ration and the impact on antigens in the en-
dosomes. The authors report no loss in en-
dosome membrane integrity after perforin-2
activation, so another mediator potentially
involved in maintaining phagosomal mem-
brane integrity could be endosomal sorting
,
tions to developing personalized vaccines. j
RE FE REN CES A ND N OT ES
1. J. M. den Haan et al., J. Exp. Med. 192, 1685 (2000).
2. M. Kovacsovics-Bankowski, K. L. Rock, Science 267, 243
(1995).
3. P. Rodríguez-Silvestre et al., Science 380, 1258 (2023).
4. R. A. Ohara, K. M. Murphy, Semin. Immunol. 66, 101711
(2023).

5. F. M. Cruz et al., Semin. Immunol. 66, 101729 (2023).

sol rather than a failure in antigen uptake.
The mechanism by which perforin-2 elic-
its antigen endocytic escape in cDC1s dif-
fers from previously described mechanisms
in the P2C pathway. However, because
Department of Microbiology and Immunology, Geisel
School of Medicine at Dartmouth College, Hanover, NH,
USA. Email: kavita.rawat@dartmouth.edu;
claudia.jakubzick@dartmouth.edu
complex required for transport III (ESCRT-
III), which facilitates the primary repair sys-
tem for intracellular membranes associated
with cDC1 and cross-presentation (8).
A study found that Nox2-deficient mice
display no defect in cross-presentation (9),
raising the question of whether there are
alternative mechanisms for the endosomal
lipid peroxidation in cDC1. An alterna-
6. R. M. McCormack et al., eLife 4, e06508 (2015).
7. D. Sancho et al., Nature 458, 899 (2009).
8. M. Gros et al., Cell Rep. 40, 111205 (2022).
9. S. C. Nalle et al., PLOS ONE 15, e0238484 (2020).
10. P. Nair-Gupta et al., Cell 158, 506 (2014).
11. M. Zehner et al., Immunity 42, 850 (2015).
12. A. N. Desch et al., Nat. Commun. 5, 4674 (2014).
13. S. R. Larson et al., Cell Death Differ. 23, 997 (2016).
10.1126/science.adi5711

SCIENCE science.org 23 JUNE 2023 • VOL 380 ISSUE 6651 1219

I NS I GHTS | P E R S P E C T I V E S
MATERIALS SCIENCE
The prospects of high-
temperature superconductors
Overcoming cost barriers could make high-temperature
superconductors pervasive
By Alexander Molodyk1 and
David C. Larbalestier2
S
uperconductors conduct electric-
ity with essentially zero resistance,
avoiding many of the power losses in
present electric power transmission,
conversion, and use. Strong electro-
magnetic fields have so far been the
principal application of superconductors,
with widespread commercial superconduc-
tivity limited to magnetic resonance imag-
ing (MRI) electromagnets composed of the
low-temperature superconductor (LTS)
Nb47Ti. Broader applications of LTSs have
been hindered by the need to cool them
with liquid helium (at or below 4.2 K).
High-temperature superconductors (HTSs)
(1) that can operate at liquid nitrogen tem-
peratures (between 65 and 80 K)
promised ubiquitous applications
that could escape the constraint of
“…the present outlook for high-temperature
LTSs. Achieving the International
Energy Agency roadmap to carbon-
free economies by 2050 would be their industrial applications is historic…”
greatly facilitated by the use of nu-
clear fusion–generated electricity. HTSs
have been used in prototype nuclear fusion
reactors (2), thereby creating the oppor-
tunity to overcome the cost barriers that
have so far prevented the commercial de-
velopment of HTS technologies.
Since the unexpected report in 1987 of
high-temperature superconductivity at 93
K (1), the idea that HTSs could revolution-
ize the electric power industry (3), going
far beyond the classical application of
superconductivity to electromagnets, has
been pursued. Despite many technical suc-
cesses (4), electric industry applications
of HTSs are few, largely because the high
cost of HTS materials made replacement
of copper and iron electric infrastructure
economically infeasible. HTS research over
the past decade thus fell back to making
ultra–high field electromagnets that were
impossible with “classical” LTS conduc-
1
Faraday Factory Japan, 2-3-19 Hashimotodai, Midori-ku,
Sagamihara, Kanagawa, Japan. 2Applied Superconductivity
Center, National High Magnetic Field Laboratory, Florida
State University, Tallahassee, FL, USA. Email: a.molodyk@
faradaygroup.com; larbalestier@asc.magnet.fsu.edu
1220 23 JUNE 2023 • VOL 380 ISSUE 6651
tors, a scientifically exciting but limited
market (5). However, HTSs could enable
economical compact fusion reactors as
the means to contribute to the 2050 zero-
carbon goal, and the development of fu-
sion synergistically creates a demand for
HTSs, promoting new production capacity
that will result in HTS cost reduction. This
could transform the economics of HTSs,
especially for replacement of copper and
iron in electrotechnology.
Major investments in developing com-
pact tokamak fusion reactors (6, 7) have
successfully yielded an ~1.8 m by 0.5 m
toroidal field coil prototype composed of
the best-performing HTS, REBa2Cu3O7-d
(REBCO; RE, rare-earth element), in
coated conductor (CC) form (8) that oper-
ates at 20 K and 20 T (9). In a compact
superconductor materials and
fusion tokamak, a set of D-shaped coils of
HTS CC generates a toroidal magnetic field
that confines the plasma (see the figure).
This prototype used 270 km of REBCO, sev-
eral times the quantity of all REBCO CCs
used in all high-field magnets made so far.
This achievement required a major expan-
sion of CC production that now provides
the opportunity to make REBCO CC avail-
able by the ton at prices that are suitable
for broad application, not just to fusion
but also to electric utilities and a liquid hy-
drogen economy.
Historically, the high-energy physics
community has provided the dominant
demand for new superconductors, and in-
deed it is now driving the demand for both
LTSs and HTSs as essential components
of ultra–high energy particle colliders.
The demonstration of a 20 T, 20 K toroi-
dal field fusion magnet (9) has created a
strong argument for design and operation
of HTS-enabled dipole magnets for any
Future Circular Collider at CERN (10).
HTSs operating at 15 to 25 K could offer
vast cryogenic savings compared with cur-
rent superfluid helium at 1.8 K, today’s ex-
pensive solution for cooling LTS magnets
at the Large Hadron Collider.
The prospect of operating helium-free,
perhaps even in liquid nitrogen at 65 to 80
K, is what originally drove expectations for
pervasive applications of HTSs, especially
REBCO. But attempts to apply HTSs soon
showed that it is not just their high transi-
tion temperatures (or critical temperatures,
Tc) that make superconductors useful by al-
lowing savings on cryogenics, but more im-
portantly, the ability to carry high current
densities (Jc) in strong magnetic fields. High
Jc depends on how well the quantized vorti-
ces inside the superconductor are “pinned”
from moving by various structural defects.
Attaining high Jc was an intense research
activity for more than 20 years that first
required understanding the pronounced
anisotropy of superconducting properties
p
in REBCO that results from its structural
anisotropy and makes strong vortex pin-
ning difficult (11). The ability of the REBCO
compound to be grown as thin films, while
incorporating high densities of insulating
nanoscale RE2O3 and perovskite compounds
g
such as BaZrO3, enabled the very strong vor-
tex pinning that makes high Jc possible, even
at liquid nitrogen temperatures (12).
A second, independent problem also
y
stood in the way of HTS applica-
tions: the large sensitivity to any
disorder that locally depresses the
superconducting properties, as oc-
curs between virtually all grains
within the polycrystalline conduc-
tor (at grain boundaries, GBs) (13).
Whereas the two principal LTS materials,
Nb47Ti and Nb3Sn, are high-carrier den-
sity, isotropic, s-wave superconductors, the
y g
cuprate HTSs are markedly anisotropic
d-wave superconductors in which GB car-
rier densities and superconducting prop-
erties are strongly depressed in all except
very low-angle GBs (when the difference in
crystallographic orientation between two
,
adjacent grains is very small). This results
in a strong degradation of grain-to-grain
connectivity and Jc of polycrystalline mate-
rial (13). Accordingly, it took more than 15
years for HTS conductors to emerge, even
in lengths of much less than 1 km. By con-
trast, LTS conductors are generally made
in single lengths of much more than 10 km.
For each of the three commercial HTS
materials available today—REBCO and
the bismuth strontium calcium copper
oxide (BSCCO) compounds Bi-2223 and
Bi-2212—the path to high Jc has only been
possible by development of high crystallo-
graphic texture fabrication that minimizes
obstructions to long-range supercurrent
transport. In the case of REBCO, a totally
science.org SCIENCE
 High-temperature superconductors in a tokamak fusion reactor
The development of nuclear fusion power generation, such as with compact tokamak fusion reactors, is driving the growth and commercialization
of high-temperature superconductor (HTS)–coated conductors.
Tokamak fusion generator
Prototype compact tokamak fusion
generators use an HTS-coated
conductor in D-shaped toroidal field
coils to generate a toroidal magnetic
field that confines plasma for nuclear
fusion power generation.
Custom finish
(thickness varies)
new, largely vapor deposition thin-film
production route was needed (12). Notably,
Bi-2212 and Bi-2223 can be made by con-
ventional wire fabrication routes but at the
price of having poorer texture and conse-
quently lower Jc. By contrast, industrializa-
tion of thin-film CC fabrication processes
has resolved both the problem of weak
vortex pinning and GB supercurrent ob-
struction, now enabling mass production
of economical HTSs.
Currently, lengths of 500 to 1000 m of
REBCO CC with almost single-crystal tex-
ture are made worldwide. The dominant
production route uses ion-beam–assisted
deposition (IBAD) to grow a cube-textured
MgO template, 10 to 50 nm thick, onto a
strong, 30- to 100-mm-thick, untextured
metallic substrate such as Hastelloy-C276.
Some intermediate oxide layers ~100 nm
GRAPHIC: A. FISHER/SCIENCE
thick then allow lattice matching of the
MgO to the 1- to 3-mm-thick REBCO layer,
which is then protected by 1 to 2 mm of
sputtered silver and finally by a thicker 5-
to 50-mm copper layer, which is generally
electroplated to provide electrical stabil-
SCIENCE science.org
Toroidal field coils
Outer poloidal field coils
(for plasma positioning
and shaping)
Copper Silver HTS
(5–50 mm) (2 mm) (1–3 mm)
ity and protection against loss of super-
conductivity in the REBCO. Most, or all,
of the process requires multiple physical
vapor deposition chambers, which come at
high capital cost and have relatively slow
throughput, making CC fabrication both
complex and expensive. The increased de-
mand on HTSs for the development of nu-
clear fusion power generation could have a
transformative impact, driving REBCO CC
manufacture into a full industrial opera-
tion with potentially huge cost reductions.
The commonly used cost metric for super-
conductors, dollars per kiloampere-meter
($/kA-m), defines the per-meter cost of con-
ductor needed to transmit 1000 A of electric
current, at 77 K in self-field (that is, with
no magnetic field applied); or, more gener-
ally, at any user-desired temperature and
magnetic field. Present-day volume prices
of HTSs range from $150 to $200/kA-m.
Many analyses of the commercial viability
of superconducting applications show that
a conductor cost of $50/kA-m is the tipping
point for widespread application for electric
power use. A long-range outlook projects
Central solenoid
p
Toroidal magnetic field
g
Coated conductor
The coated conductor comprises
multiple layers to protect the HTS
and ensure electrical stability.
Fabrication involves various
y
physical vapor deposition
processes, which are expensive
and slow. But manufacturing is
becoming more robust and
Buffer Substrate scalable as demand grows, which
(~100 nm) (30–100 mm) is reducing production cost.
HTS costs below $10/kA-m when produced
on a very large scale (14).
It is remarkable that among all manu-
y g
facturable superconductors, Nb47Ti, the
least “potent” (in the sense of its available
operational field and temperature space),
is the only one to have reached commer-
cial, tonnage scale. Despite expensive Nb,
Nb47Ti became the economical enabler of
,
mass-market applications of MRI, because
MRI electromagnets made with Nb47Ti op-
erate in the persistent current mode and
require only a small cryocooler. Because
compact fusion reactors require properties
that are only attainable with REBCO CC,
an opportunity to bring the distinct bene-
fits of HTSs to many new markets that can-
not tolerate the present high cost of HTSs
is provided.
Prototype compact fusion reactors (6, 7)
required a 10-fold increase in HTS supply,
from hundreds to thousands of kilometers
a year in the past 3 years. This demand
enabled recent advances that made HTS
manufacture robust and scalable (8), al-
lowing the required production increase to
23 JUNE 2023 • VOL 380 ISSUE 6651 1221
I NS I GHTS | P E R S P E C T I V E S
an annual multi-ton level. This huge expan-
sion of production scale could soon reduce
conductor costs to ~$100/kA-m. HTS use
cost also depends strongly on the super-
conductor Jc and production yield. Today’s
best laboratory samples have Jc exceeding
that of commercial conductors by a factor
of 2 or more (15), thus providing a further
industrial improvement path. As produc-
tion technology matures, manufacturing
yield will also increase, further reducing
cost. This will allow HTS CCs to become
competitive for applications in which cop-
per and iron are replaced in electric utili-
ties and wind turbines, and perhaps even
enabling electric aircraft with hydrogen-
cooled superconducting motors.
Overall, the present outlook for HTS
materials and their industrial applications
is historic, because of the opportunity for
REBCO superconductor use to expand,
as happened 35 years ago for the produc-
tion of Nb47Ti for MRI electromagnets.
The development of compact nuclear fu-
sion power generation (which is still at the
prototype stage) is the immediate stimulus
that has driven exponential annual volume
increases. The applied superconductivity
community is anticipating the virtuous
cycle of price reduction and further de-
mand from other electrotechnology appli-
cations that are not yet economic at today’s
REBCO CC prices compared with the pres-
ent use of copper, iron, and LTSs. This pro-
spective sustainable market of HTS mate-
rials and applications promises numerous
public benefits for much human activity in
energy production, distribution, and use;
medicine; transportation; and research. j
R E F E R E N C ES A N D N OT ES
1. M. K. Wu et al., Phys. Rev. Lett. 58, 908 (1987).
2. P. Ball, Nature 599, 362 (2021).
3. D. Larbalestier, A. Gurevich, D. M. Feldmann, A.
Polyanskii, Nature 414, 368 (2001).
4. R. Scanlan, A. P. Malozemoff, D. C. Larbalestier, Proc.
IEEE 92, 1639 (2004).
5. S. Hahn et al., Nature 570, 496 (2019).
6. B. N. Sorbom et al., Fusion Eng. Des. 100, 378 (2015).
7. A. Sykes et al., Nucl. Fusion 58, 016039 (2018).
8. A. Molodyk et al., Sci. Rep. 11, 2084 (2021).
9. https://cfs.energy/news-and-media/
cfs-commercial-fusion-power-with-hts-magnet
10. P. Védrine et al., in European Strategy for Particle
Physics—Accelerator R&D Roadmap, N. Mounet, Ed.
(CERN Yellow Reports: Monographs, CERN-2022-001),
chap. 2, pp. 9–59.
11. D. J. Bishop, Nature 365, 394 (1993).
12. J. L. MacManus-Driscoll, S. Wimbush, Nat. Rev. Mater. 6,
587 (2021).
13. H. Hilgenkamp, J. Mannhart, Rev. Mod. Phys. 74, 485
(2002).
14. V. Matias, R. H. Hammond, Phys. Procedia 36, 1440
(2012).
15. G. Majkic et al., Supercond. Sci. Technol. 33, 07LT03
(2020).
AC K N OW L E D G M E N TS
A.M. is R&D director, shareholder, and board member at
Faraday Factory Japan.
10.1126/science.abq4137
1222 23 JUNE 2023 • VOL 380 ISSUE 6651
PSYCHOLOGY
How AI can distort human beliefs
Models can convey biases and false information to users
By Celeste Kidd1 and Abeba Birhane2,3
I
ndividual humans form their beliefs by
sampling a small subset of the available
data in the world. Once those beliefs are
formed with high certainty, they can be-
come stubborn to revise. Fabrication and
bias in generative artificial intelligence
(AI) models are established phenomena that
can occur as part of regular system use, in the
absence of any malevolent forces seeking to
push bias or disinformation. However, trans-
mission of false information and bias from
these models to people has been prominently
absent from the discourse. Overhyped, unre-
alistic, and exaggerated capabilities perme-
ate how generative AI models are presented,
which contributes to the popular misconcep-
tion that these models exceed human-level
reasoning and exacerbates the risk of trans-
mission of false information and negative
stereotypes to people.
Generative AI models—including OpenAI’s
GPT variants, Google’s Bard, OpenAI’s
DALL·E, Stable Diffusion, and Midjourney—
have captured the minds of the public and
inspired widespread adoption. Yet, these
models contain known racial, gender, and
class stereotypes and biases from their train-
ing data and other structural factors, which
downstream into model outputs (1–3).
Marginalized groups are the most negatively
affected by these biases. Further, these mod-
els regularly fabricate information (4). Some
model developers have acknowledged these
problems but suggested that people must use
the systems to reveal trends in problematic
outputs to remedy them. This ignores that
distortions to human beliefs caused by gen-
erative AI models cannot be easily corrected
after problems are discovered. Further, the
reactive nature of this approach does not ac-
knowledge a key problem of current genera-
tive AI systems, the inability of their architec-
ture to distinguish fact from fiction (4).
Three core tenets of human psychol-
ogy can help build a bridge of understand-
ing about what is at stake when discussing
regulation and policy options. These ideas in
psychology can connect to machine learning
but also those in political science, education,
1
Department of Psychology, University of California Berkeley,
Berkeley, CA, USA. 2Mozilla Foundation, San Francisco, CA,
USA. 3Trinity College Dublin, School of Computer Science
and Statistics, Trinity College Dublin, Dublin, Ireland. Email:
celestekidd@gmail.com; adbirhane@gmail.com
communication, and the other fields that are
considering the impact of bias and misinfor-
mation on population-level beliefs.
People form stronger, longer-lasting beliefs
when they receive information from agents
that they judge to be confident and knowl-
edgeable, starting in early childhood. For
example, children learned better when they
learned from an agent who asserted their
knowledgeability in the domain as compared
with one who did not (5). That very young
children track agents’ knowledgeability and
use it to inform their beliefs and exploratory
p
behavior supports the theory that this abil-
ity reflects an evolved capacity central to our
species’ knowledge development.
Although humans sometimes communi-
cate false or biased information, the rate of
human errors would be an inappropriate
g
baseline for judging AI because of fundamen-
tal differences in the types of exchanges be-
tween generative AI and people versus peo-
ple and people. For example, people regularly
y
communicate uncertainty through phrases
such as “I think,” response delays, corrections,
and speech disfluencies. By contrast, genera-
tive models unilaterally generate confident,
fluent responses with no uncertainty repre-
sentations nor the ability to communicate
their absence. This lack of uncertainty signals
in generative models could cause greater dis-
tortion compared with human inputs.
Futher, people assign agency and inten-
y g
tionality readily. In a classic study, people
read intentionality into the movements
of simple animated geometric shapes (6).
Likewise, people commonly read intention-
ality—and humanlike intelligence or emer-
gent sentience—into generative models even
,
though these attributes are unsubstantiated
(7). This readiness to perceive generative
models as knowledgeable, intentional agents
implies a readiness to adopt the informa-
tion that they provide more rapidly and with
greater certainty. This tendency may be fur-
ther strengthened because models support
multimodal interactions that allow users
to ask models to perform actions like “see,”
“draw,” and “speak” that are associated with
intentional agents. The potential influence of
models’ problematic outputs on human be-
liefs thus exceeds what is typically observed
for the influence of other forms of algorith-
mic content suggestion such as search. These
issues are exacerbated by financial and liabil-
ity interests incentivizing companies to an-
science.org SCIENCE
I NS I GHTS | P E R S P E C T I V E S
an annual multi-ton level. This huge expan-
sion of production scale could soon reduce
conductor costs to ~$100/kA-m. HTS use
cost also depends strongly on the super-
conductor Jc and production yield. Today’s
best laboratory samples have Jc exceeding
that of commercial conductors by a factor
of 2 or more (15), thus providing a further
industrial improvement path. As produc-
tion technology matures, manufacturing
yield will also increase, further reducing
cost. This will allow HTS CCs to become
competitive for applications in which cop-
per and iron are replaced in electric utili-
ties and wind turbines, and perhaps even
enabling electric aircraft with hydrogen-
cooled superconducting motors.
Overall, the present outlook for HTS
materials and their industrial applications
is historic, because of the opportunity for
REBCO superconductor use to expand,
as happened 35 years ago for the produc-
tion of Nb47Ti for MRI electromagnets.
The development of compact nuclear fu-
sion power generation (which is still at the
prototype stage) is the immediate stimulus
that has driven exponential annual volume
increases. The applied superconductivity
community is anticipating the virtuous
cycle of price reduction and further de-
mand from other electrotechnology appli-
cations that are not yet economic at today’s
REBCO CC prices compared with the pres-
ent use of copper, iron, and LTSs. This pro-
spective sustainable market of HTS mate-
rials and applications promises numerous
public benefits for much human activity in
energy production, distribution, and use;
medicine; transportation; and research. j
R E F E R E N C ES A N D N OT ES
1. M. K. Wu et al., Phys. Rev. Lett. 58, 908 (1987).
2. P. Ball, Nature 599, 362 (2021).
3. D. Larbalestier, A. Gurevich, D. M. Feldmann, A.
Polyanskii, Nature 414, 368 (2001).
4. R. Scanlan, A. P. Malozemoff, D. C. Larbalestier, Proc.
IEEE 92, 1639 (2004).
5. S. Hahn et al., Nature 570, 496 (2019).
6. B. N. Sorbom et al., Fusion Eng. Des. 100, 378 (2015).
7. A. Sykes et al., Nucl. Fusion 58, 016039 (2018).
8. A. Molodyk et al., Sci. Rep. 11, 2084 (2021).
9. https://cfs.energy/news-and-media/
cfs-commercial-fusion-power-with-hts-magnet
10. P. Védrine et al., in European Strategy for Particle
Physics—Accelerator R&D Roadmap, N. Mounet, Ed.
(CERN Yellow Reports: Monographs, CERN-2022-001),
chap. 2, pp. 9–59.
11. D. J. Bishop, Nature 365, 394 (1993).
12. J. L. MacManus-Driscoll, S. Wimbush, Nat. Rev. Mater. 6,
587 (2021).
13. H. Hilgenkamp, J. Mannhart, Rev. Mod. Phys. 74, 485
(2002).
14. V. Matias, R. H. Hammond, Phys. Procedia 36, 1440
(2012).
15. G. Majkic et al., Supercond. Sci. Technol. 33, 07LT03
(2020).
AC K N OW L E D G M E N TS
A.M. is R&D director, shareholder, and board member at
Faraday Factory Japan.
10.1126/science.abq4137
1222 23 JUNE 2023 • VOL 380 ISSUE 6651
PSYCHOLOGY
How AI can distort human beliefs
Models can convey biases and false information to users
By Celeste Kidd1 and Abeba Birhane2,3
I
ndividual humans form their beliefs by
sampling a small subset of the available
data in the world. Once those beliefs are
formed with high certainty, they can be-
come stubborn to revise. Fabrication and
bias in generative artificial intelligence
(AI) models are established phenomena that
can occur as part of regular system use, in the
absence of any malevolent forces seeking to
push bias or disinformation. However, trans-
mission of false information and bias from
these models to people has been prominently
absent from the discourse. Overhyped, unre-
alistic, and exaggerated capabilities perme-
ate how generative AI models are presented,
which contributes to the popular misconcep-
tion that these models exceed human-level
reasoning and exacerbates the risk of trans-
mission of false information and negative
stereotypes to people.
Generative AI models—including OpenAI’s
GPT variants, Google’s Bard, OpenAI’s
DALL·E, Stable Diffusion, and Midjourney—
have captured the minds of the public and
inspired widespread adoption. Yet, these
models contain known racial, gender, and
class stereotypes and biases from their train-
ing data and other structural factors, which
downstream into model outputs (1–3).
Marginalized groups are the most negatively
affected by these biases. Further, these mod-
els regularly fabricate information (4). Some
model developers have acknowledged these
problems but suggested that people must use
the systems to reveal trends in problematic
outputs to remedy them. This ignores that
distortions to human beliefs caused by gen-
erative AI models cannot be easily corrected
after problems are discovered. Further, the
reactive nature of this approach does not ac-
knowledge a key problem of current genera-
tive AI systems, the inability of their architec-
ture to distinguish fact from fiction (4).
Three core tenets of human psychol-
ogy can help build a bridge of understand-
ing about what is at stake when discussing
regulation and policy options. These ideas in
psychology can connect to machine learning
but also those in political science, education,
1
Department of Psychology, University of California Berkeley,
Berkeley, CA, USA. 2Mozilla Foundation, San Francisco, CA,
USA. 3Trinity College Dublin, School of Computer Science
and Statistics, Trinity College Dublin, Dublin, Ireland. Email:
celestekidd@gmail.com; adbirhane@gmail.com
communication, and the other fields that are
considering the impact of bias and misinfor-
mation on population-level beliefs.
People form stronger, longer-lasting beliefs
when they receive information from agents
that they judge to be confident and knowl-
edgeable, starting in early childhood. For
example, children learned better when they
learned from an agent who asserted their
knowledgeability in the domain as compared
with one who did not (5). That very young
children track agents’ knowledgeability and
use it to inform their beliefs and exploratory
p
behavior supports the theory that this abil-
ity reflects an evolved capacity central to our
species’ knowledge development.
Although humans sometimes communi-
cate false or biased information, the rate of
human errors would be an inappropriate
g
baseline for judging AI because of fundamen-
tal differences in the types of exchanges be-
tween generative AI and people versus peo-
ple and people. For example, people regularly
y
communicate uncertainty through phrases
such as “I think,” response delays, corrections,
and speech disfluencies. By contrast, genera-
tive models unilaterally generate confident,
fluent responses with no uncertainty repre-
sentations nor the ability to communicate
their absence. This lack of uncertainty signals
in generative models could cause greater dis-
tortion compared with human inputs.
Futher, people assign agency and inten-
y g
tionality readily. In a classic study, people
read intentionality into the movements
of simple animated geometric shapes (6).
Likewise, people commonly read intention-
ality—and humanlike intelligence or emer-
gent sentience—into generative models even
,
though these attributes are unsubstantiated
(7). This readiness to perceive generative
models as knowledgeable, intentional agents
implies a readiness to adopt the informa-
tion that they provide more rapidly and with
greater certainty. This tendency may be fur-
ther strengthened because models support
multimodal interactions that allow users
to ask models to perform actions like “see,”
“draw,” and “speak” that are associated with
intentional agents. The potential influence of
models’ problematic outputs on human be-
liefs thus exceeds what is typically observed
for the influence of other forms of algorith-
mic content suggestion such as search. These
issues are exacerbated by financial and liabil-
ity interests incentivizing companies to an-
science.org SCIENCE
thropomorphize generative models as intel-
ligent, sentient, empathetic, or even childlike.
The number of exposures to fabricated in-
formation predicts how deeply ingrained the
belief in that information becomes. Greater
repetition predicted greater strength in a per-
son’s belief in a false statement—even when
the statement contradicts a person’s prior
knowledge (8). Trends that increase people’s
exposures to fabrications consequently in-
crease the strength of people’s beliefs in false
information. The trend of integrating genera-
tive AI models into existing technologies—
e.g., search engines and smartphones—will
almost certainly mean greater exposure to
the models’ fabrications and biases.
Similarly, repeated exposure to biases in
algorithmic systems transmits the biases to
human users over time. For example, when
a risk-assessment system, such as used by
court judges to determine how a defendant
should be sentenced (9), assigns Black indi-
viduals higher risk scores than white individ-
uals with the exact same criminal histories,
human judges learn these statistical regulari-
ties and may “change their sentencing prac-
tices in order to match the predictions of the
algorithms” through a process likened to an-
choring [(10), p. 287]. This mechanism of sta-
tistical learning could lead a judge to believe
Black individuals to be more likely to reof-
fend—even if use of the system is stopped by
regulations like those adopted in California.
Generative AI models have the potential to
further amplify the repeated exposure issues
for both fabrications and bias because of their
expected influence on contents of the World
Wide Web—a primary source of training data
for the models. For example, the rapid rise
and accessibility of generative models, such
as Stable Diffusion, have generated millions
of outputs each day (11). This output in turn
becomes part of the training data for the next
generation of models—thus amplifying the
impact of the systemic distortions and biases
into the future in a continuous feedback loop.
The more rapidly such systems are used
and adopted, and the more they are built
into the backend of systems used across all
sectors, the more influence the systems have
over human beliefs. For example, marketing
content can now be generated by generative
AI models, then targeted at users using psy-
chometric methods, then fine-tuned, looped,
and fed back to users in an automatic system
designed to induce engagement behaviors,
irrespective of and incapable of considering
how its content might distort human beliefs
in general or the inclusion of either fabrica-
tions or stereotyped biases in this material.
Users of conversational generative AI
models request information in particular
moments—when they are uncertain and thus
most open to learning something new. Once
SCIENCE science.org
a person has received an answer, their uncer-
tainty drops, their curiosity is diminished,
and they don’t consider or weigh subsequent
evidence in the same way as when they were
in the early stages of making up their minds
(12). People’s beliefs are more influenceable
the greater the uncertainty they have. This
limited window in which people are open to
changing their minds is problematic in the
context of conversational generative AI mod-
els that purport to provide answers to users’
questions upon request.
This aspect of human curiosity has long-
standing implications for how these systems
affect human beliefs. It means that informa-
tion transmitted from a large-scale language
model to an uncertain person will be difficult
to update after the fact—because the infor-
mation provided by the model will resolve
the person’s uncertainty even if it is incor-
“…a faulty belief…can pass
among people in the
population in perpetuity…”
rect (13). The problems also affect the use of
systems that generate images from users’ text
prompts because the act of asking a model
to translate text into visual imagery can be
driven by curiosity that resolves once the user
sees the visual output. Negative sterotyped
biases in such visual outputs run similar risks
of taking root in stubborn ways. Once a faulty
belief is fixed within a person—and especially
if the same fabrication or bias is passed and
then becomes fixed in many people who use
the same system—it can pass among people
in the population in perpetuity (14).
Thus, transmitted biases or fabricated
information are not easily correctable after
the fact either within individuals or at the
population level (15). This aspect of human
psychology interacts with how humans treat
agentive entities and, in particular, their ten-
dency to be more greatly swayed by agents
that they perceive as confident and knowl-
edgeable (6). The amount of information re-
quired to reach that threshold certainty will
be less in the context of it being delivered by
a seemingly confident and knowledgeable
agent—especially if it is presented in more-
humanlike ways, as in the context of a con-
versation. Thus, developers’ claims surround-
ing their generative AI system can affect how
much faulty outputs distort human beliefs.
The nascent stage of this technology offers
a transient opportunity to conduct interdis-
ciplinary studies that measure the impact
of generative models on human beliefs and
biases. This opportunity rapidly diminishes
once these systems are more widely adopted
and more deeply embedded into other every-
day technologies. Research on how genera-
tive AI models affect children’s beliefs is an
especially high priority. Children are more
vulnerable to belief distortion because of
their increased tendencies to anthropomor-
phize technology and their more nascent,
influenceable knowledge states.
Independent audits must include not only
assessments of fabrication and bias but also
measurements of how knowledgeable users
rate systems to be and how much they trust
the outputs. These data could be used to es-
timate both the rate of problematic model
outputs to users and how severely these out-
puts influence human beliefs in advance of
actual transmission. The fields of psychology
and machine learning could unite to turn
their attention, collaborative capacities, and
resources to doing this work.
Studies and subsequent interventions
would be most effectively focused on impacts
p
on marginalized populations who are dis-
proportionately affected by both fabrications
and negative stereotypes in model outputs.
Resources are needed for the education of the
public, policy-makers, and interdisciplinary
scientists to give realistically informed views of
g
how generative AI models work and to correct
existing misinformation and hype surround-
ing these new technologies. Collaborative
action requires teaching everyone how to dis-
y
criminate actual from imagined capabilities
of new technologies to focus on tackling real,
concrete challenges together. j
RE FE REN C ES AN D N OT ES
1. A. Birhane, Nature 610, 451 (2022).
2. A. Birhane, Patterns 2, 100205 (2021).
3. E. M. Bender, T. Gebru, A. McMillan-Major, S. Shmitchell,
in Proceedings of the 2021 ACM Conference on Fairness,
Accountability, and Transparency (Association for
Computing Machinery, 2021), pp. 610–623.
4. R. Azamfirei, S. R. Kudchadkar, J. Fackler, Crit. Care 27, 120
(2023).
y g
5. M. A. Sabbagh, D. A. Baldwin, Child Dev. 72, 1054 (2001).
6. F. Heider, M. Simmel, Am. J. Psychol. 57, 243 (1944).
7. A. Birhane, J. van Dijk, in Proceedings of the AAAI/ACM
Conference on AI, Ethics, and Society (Association for the
Advancement of Artificial Intelligence, 2020), pp. 207–213.
8. L. K. Fazio, R. M. Pillai, D. Patel, J. Exp. Psychol. 151, 2604
(2022).
9. B. Green, Y. Chen, in Proceedings of the 2019 ACM
,
Conference on Fairness, Accountability, and Transparency
(Association for Computing Machinery, 2019), pp. 90–99.
10. A. Christin, in The Decisionist Imagination: Sovereignty,
Social Science and Democracy in the 20th Century, D.
Besser, N. Guilhot, Eds. (Berghahn Books, 2018), pp.
272–294.
11. F. Bianchi et al., https://arxiv.org/abs/2211.03759 (2022).
12. L. Martí, F. Mollica, S. Piantadosi, C. Kidd, Open Mind 2, 47
(2018).
13. S. Wade, C. Kidd, Psychon. Bull. Rev. 26, 1377 (2019).
14. C. Kidd, B. Y. Hayden, Neuron 88, 449 (2015).
15. B. Thompson, T. L. Griffiths, Proc. R. Soc. B 288, 20202752
(2021).
AC KN OW LE DG M E N TS
The authors thank Ş. Wong, D. Raji, C. O’Neil, A. Smart, T. Ryan,
B. Thompson, A. Eason, S. T. Piantadosi, and two reviewers for
feedback. C.K. is supported by the Walton Family Foundation,
the Hellman Fellows Fund, DARPA Machine Common Sense
TA1 (BAA no. HR001119S0005), and the John Templeton
Foundation (Character Virtue Development, ID 61475). A.B. is
supported by Mozilla Senior Fellows, Trustworthy AI.
10.1126/science.adi0248
23 JUNE 2023 • VOL 380 ISSUE 6651 1223
I NS I GHTS
P OLICY FORUM
R ESEARCH ETHICS AND LAW
Ethical research when abortion
access is legally restricted
Risks and benefits of some clinical research may be altered
By Jeremy Sugarman1,2, Danielle M. Wenner3,
Annette Rid4, Leslie Meltzer Henry1,5,
Florencia Luna6,7, Robert Klitzman8, Kathleen
M. MacQueen9,10,11, Stuart Rennie12, Jerome
Amir Singh13,14, Lawrence O. Gostin15
T
he legal landscape surrounding abor-
tion in the United States has shifted
dramatically since the Supreme Court’s
June 2022 decision in Dobbs v. Jackson
Women’s Health Organization elimi-
nated a nationwide right to abortion
(1). In the year since, roughly half of US states
have expanded abortion restrictions. Some
consequences of heightened restrictions—in-
cluding increased maternal morbidity and
mortality and deepening socioeconomic and
racial inequities—have quickly come into
view. However, little attention has focused on
the ethical, legal, and practical implications
that such restrictions have for research in-
volving people who could become pregnant
during research and research staff. Notably,
limited access to abortion can pose risks to
clinical research participants and potentially
compromise the scientific and social value
of some research. As a result, assessments of
potential research risks and benefits may be
altered. We outline points for various stake-
holders [such as sponsors, investigators, re-
search sites, and institutional review boards
(IRBs)] to consider in addressing these issues.
To date, 13 US states have revived or adopt-
ed near-total bans on abortion, leaving ap-
proximately 22 million people who could
become pregnant without access to abortion
in their states. Other states have set early
gestational limits on abortion access, and
several penalize anyone who facilitates an
abortion. Although most US states currently
provide exceptions for the pregnant person’s
life (but not health), ambiguous laws and
fear of criminal prosecution raise profound
concerns among clinicians, those who might
become pregnant, and others (2–4). Given
the gender gap in evidence to inform clinical
care that has resulted from a combination of
traditional trial requirements for contracep-
tion and the active exclusion from research
of those who might become pregnant, it is
1224 23 JUNE 2023 • VOL 380 ISSUE 6651
imperative that decision-makers pay careful
attention to factors, such as limited access to
abortion, that could reflexively restrict clini-
cal research with that population.
IMPLICATIONS FOR CLINICAL RESEARCH
Abortion restrictions have special moral im-
plications in the context of some clinical re-
search with people who can become pregnant.
First, research interventions, by definition,
expose participants to unknown risks. When
research participants become pregnant, those
risks can include threats to the health or life
of the pregnant person as well as potential
harm to their fetus. A participant may decide
that a timely and safe abortion is the best way
to mitigate research-related harms.
Second, research studies can require par-
ticipants to undergo pregnancy tests that
they might not have taken outside the re-
search context. These tests might detect and
document early pregnancies that otherwise
would have gone unnoticed given high rates
of miscarriage in the first trimester, which
in turn might raise concerns in some juris-
dictions that the participant obtained an il-
legal abortion. That is, the simple fact that
a research participant is not pregnant nor
has given birth, but a test indicates that they
were pregnant during research, could put
them at risk of legal action.
Third, if risks to research participation
that result from legal restrictions on abortion
access are not sufficiently addressed, people
who can become pregnant might be deterred
from enrolling in clinical research. This could
compromise the scientific and social value of
research, reinforcing longstanding gender
disparities, which are due in part to long-
standing underrepresentation of people who
can become pregnant in research (5, 6).
Fourth, fear of legal risks associated with
facilitating an abortion, or uncertainty
about the rapidly evolving legal status of
abortion, might leave researchers reluctant
to obtain rigorous data on pregnancy, pos-
sibly including adverse pregnancy-related
outcomes. This could further reduce the
scientific and social value of clinical re-
search because of incomplete data on clini-
cal interventions for both pregnant people
and their fetuses.
Fifth, clinical research staff may face legal
risks in jurisdictions with laws that penalize
those who facilitate abortion (for example,
referring a participant to an abortion pro-
vider out of state). These risks, which are
difficult to assess in a rapidly changing legal
environment, might nevertheless have seri-
ous professional and personal implications.
POINTS TO CONSIDER
It is broadly recognized that research-
related risks be minimized and reason-
able in relation to the scientific and social
value of research. Here, we outline points
for stakeholders to consider in ensuring a
reasonable risk-benefit profile for research
that involves participants who may become
pregnant when the research is conducted in
locations with restrictive laws on abortion
p
access (see the box). This includes consid-
erations regarding research site selection
and management as well as study design
and implementation. There are additional
considerations for research focused specifi-
cally on pregnant people and their fetuses
g
or accruing data on the effects of particular
interventions among them, but these are
beyond our scope here.
y
Assess current laws
Laws that restrict abortion access are in flux
and vary markedly across jurisdictions. In-
complete or outdated knowledge about local
abortion laws and access to abortion services
(such as mifepristone) can pose risks to re-
search participants and staff. Therefore, an
accurate and current understanding of rele-
vant legal restrictions at each site is essential.
Individuals with local legal expertise should
y g
be engaged to gather and interpret relevant
local laws and policies not only as research
is being planned but also over time to ensure
understanding of the current landscape.
Evaluate site experience
,
The impact of laws often depends on the na-
ture and extent of their enforcement. Thus,
it is essential to have as clear an understand-
ing as possible of the likelihood that laws will
be enforced locally to more accurately assess
the true risks. Investigators should explore
local research and clinical staff experiences
and knowledge related to legal restrictions
on abortion at a site. For example, it would
be important to ascertain whether any prob-
lems have previously arisen related to legal
restrictions on abortion access, and if so,
what was done to address them. Such infor-
mation needs to be considered when making
difficult decisions about whether and how
the research might safely and appropriately
be conducted at each proposed site.
science.org SCIENCE
 ever, if it cannot be ensured that the risks
are reasonable in relation to the potential
Points to consider scientific and social value, researchers and
RESEARCH SITE SELECTION AND MANAGEMENT sponsors should consider not pursuing re-
Current laws. What are the current state laws regarding access to abortion services at the
search at that particular site. Of course,
research site? Include information concerning provisions for legal abortions, any allowance for
such decisions should not be taken lightly
citizen enforcement, penalties for referral, and resulting risks to participants and staff.
because such a choice obviates the opportu-
nity for people who can become pregnant to
Site experience. Have any problems arisen related to legal restrictions on abortion access participate in research and generate locally
in current studies? If so, what was done to address these problems? relevant data.
Stakeholder engagement. Have research staff, potential participants, or enrolled partici-
pants expressed concern about legal restrictions on abortion access at the site? Has there Make provisions for legal abortion access
been community engagement regarding legal restrictions on abortion access at the site in Before study commencement, stakeholders
relation to this study and/or research in general? Who was included in that engagement? should develop mechanisms to ensure timely
Site suitability. What, if any, legal restrictions on abortion access may affect the study at abortion access mechanisms for participants
the site (for example, do they affect recruitment or the ability to carry out the study)? who may become pregnant during the study
that minimize physical, social, legal, and
Provisions for abortion access. What plans are in place at the site to ensure that partici-
economic risks. In settings where those who
pants who may become pregnant can access timely abortion services without risks (physi-
facilitate abortion access face legal risk, in-
cal, social, legal, or economic) to themselves or to study staff?
vestigators should prepare research staff who
STUDY DESIGN AND IMPLEMENTATION may be interacting directly with participants

p
who become pregnant and desire abortion
Confidentiality. Are there any special provisions being made for data privacy regarding
services regarding appropriate ways to man-
pregnancy test results or access to abortion services? Has a Certificate of Confidentiality
age such requests. The need for these sorts
been obtained?
of provisions should be revisited in the event
Mitigating risk. Has the site, local research institution, or protocol team implemented or that the legal landscape changes.
considered any standard operating procedures or provisions to manage the risks (physical,

g
social, legal, and economic) associated with legal restrictions on abortion access? Ensure confidentiality
Informed consent. Are the risks (physical, social, legal, and economic) associated with The current legal environment underscores
legal restrictions on abortion access at the site clearly stated in the study’s consent form? Is the criticality of maintaining confidentiality
there a need for reconsent to address legal restrictions on abortion access given a change in of all information about pregnancy and use of

y
the law after enrollment? abortion services because such information
Social harm monitoring. Are participants explicitly asked about risks (physical, social,
may pose direct risks to participants and re-
legal, and economic) associated with legal restrictions on abortion access at regular study
search staff. As in other research settings that
visits? Is there a mechanism for participants and study staff to seek urgent assistance?
necessitate strict confidentiality protections,
it is essential to consider this issue during the
Independent oversight. Has the responsible institutional review board (IRB) made any design and implementation of data collection
determinations related to risks or issues associated with legal restrictions on abortion procedures and data management. For ex-
access for this study? For sites using a single IRB, was information about the local context ample, simple measures may include the use
requested by, or provided to, the single IRB? of participant identifiers with links to actual
participants maintained elsewhere and using

Engage local stakeholders
Potential or enrolled participants may
be particularly well situated to provide
relevant information regarding the risks
related to restrictive abortion laws and
Evaluate site suitability
Information about current abortion laws,
site experience, and stakeholder engage-
ment should be used as part of any evalu-
ation of whether existing legal restrictions
y g
appropriate encryption and password protec-
tion of data. Obtaining a Certificate of Con-
fidentiality, issued by the National Institutes
of Health to prevent compelled disclosure
of protected information, may provide addi-
tional protection in the case of civil or crimi-

how best to manage them. Consequently,
they should be explicitly engaged around
these issues. Community-based reproduc-
tive health providers and advocates are
also potential resources in this regard.
Consistent with good participatory re-
search practices, investigators should work
with local site staff to elicit such informa-
tion and explore options for navigating
optimal care (7).
could reasonably be expected to negatively
affect research at the site. For example, as
discussed above, legal restrictions might
pose risks to research staff or participants
and their fetuses, impede recruitment of a
sufficient number of participants, or com-
promise the research study’s scientific and
social value. In such circumstances, the
feasibility of modifying the research and its
implementation should be assessed. How-
,
nal prosecution, although such certificates
have yet to be tested in this context in court
(8). Furthermore, even if the research record
is protected by the Health Insurance Porta-
bility and Accountability Act Privacy Rule,
entities covered by this rule are currently
permitted to disclose this information in the
event of legal action. However, a revision has
been proposed to prohibit such disclosures,
so this should be monitored closely (9).

1
Berman Institute of Bioethics, Johns Hopkins University, Baltimore, MD, USA. 2Department of Medicine, Johns Hopkins University, Baltimore, MD, USA. 3Department of Philosophy and Center
for Ethics and Policy, Carnegie Mellon University, Pittsburgh, PA, USA. 4Department of Bioethics, The Clinical Center, National Institutes of Health, Bethesda, MD, USA. 5University of Maryland
Carey School of Law, Baltimore, MD, USA. 6Latin American School of Social Sciences (FLACSO) Bioethics Program, Institute for Social Research of Latin America (IICSAL), Buenos Aires,
Argentina. 7National Scientific and Technical Research Council (CONICET), Buenos Aires, Argentina. 8Vagelos College of Physicians and Surgeons and Joseph Mailman School of Public Health,
Columbia University, New York, NY, USA. 9FHI 360, Durham, NC, USA. 10UNC Center for AIDS Research, Chapel Hill, NC, USA. 11Gillings School of Global Public Health and School of Medicine,
University of North Carolina, Chapel Hill, NC, USA. 12UNC Center for Bioethics, University of North Carolina, Chapel Hill, NC, USA. 13School of Law, Howard College, University of KwaZulu-Natal,
Durban, South Africa. 14University of Toronto, Toronto, Canada. 15O’Neill Institute for National and Global Health Law, Georgetown University, Washington, DC, USA. Email: jsugarman@jhu.edu

SCIENCE science.org 23 JUNE 2023 • VOL 380 ISSUE 6651 1225

I NS I GHTS | P O L I C Y F O RU M
Mitigate risks to participants
Researchers and sites should develop and
implement locally informed approaches to
managing the physical, social, legal, and
economic risks associated with laws re-
stricting abortion. This may involve modi-
fying study designs or standard operating
procedures. Of note, this may involve the
need to reconsider some aspects of research
implementation that may typically seem
mundane. For example, although it is com-
monplace to regularly assess pregnancy
during the course of some clinical research,
researchers should consider risks to partici-
pants when planning both the frequency, if
any, of pregnancy testing as well as the in-
clusion of these test results in research rec-
ords. Frequent testing may result in docu-
menting pregnancies that otherwise would
have gone undetected because of spontane-
ous early abortion. When it is necessary to
have longitudinal information about the
possibility of pregnancy over the course of
a research study, stakeholders should con-
sider whether pregnancy self-testing might
be an appropriate substitute for testing
done at research sites. Study staff should
be appropriately educated and regularly
reminded about the importance of gather-
ing accurate data about pregnancy-related
outcomes in a manner that minimizes po-
tential risks to participants. On the basis of
experience in other research settings that
pose heightened legal and social risks to
participants, consideration should be given
to developing and implementing partici-
pant safety plans to minimize potential so-
cial and legal harms to participants (10, 11).
The obligation to mitigate risk also under-
scores the need to ensure that participants
have access to effective contraception.
Obtain informed consent
Although researchers and study teams may
feel confident in their ability to navigate re-
strictions on abortions, ultimately partici-
pants will be those most affected. Therefore,
potential participants should be informed
about possible clinical risks to themselves,
or their fetuses, should they become preg-
nant during research, along with possible
options for continuing with study interven-
tions while pregnant as well as for accessing
effective contraception if they wish to avoid
pregnancy while in the study. Critically,
potential participants may be unaware of
the status of abortion restrictions and how
these may relate to the research (12). This
information, along with associated risks
(social, legal, and economic) and measures
taken to minimize them, should be explic-
itly included in the consent process. Given
the potentially rapid evolution of legal re-
strictions, research teams and IRBs should
1226 23 JUNE 2023 • VOL 380 ISSUE 6651
monitor for relevant legal changes, deter-
mine whether any changes affect the ethical
acceptability of continuing research, and
obtain IRB-approved reconsent if risk pro-
files and the potential for social harms to
participants dramatically change (13).
Monitor for social harms
Despite best and well-intended efforts to
ensure safety in research that at the out-
set seems to pose particular risks related
to undesired pregnancies (such as test-
ing contraceptive modalities) or other re-
search that simply includes people who
can become pregnant, some participants
may experience physical, social, legal, and
economic harms related not only to re-
search but also to restrictive abortion laws.
Consequently, researchers should use ex-
plicit mechanisms to detect and help man-
age both expected or unexpected harms
(14). For example, this could include a
safe mechanism for participants and study
staff to seek urgent assistance as well as
explicitly inquiring about these issues at
regular study visits. Research participants’
partners may also face harms that warrant
monitoring and attention. Although issues
that arise in relation to restrictions on
abortion access may be outside the narrow
scope of some research studies, research-
ers and sites arguably have ancillary care
obligations that extend to this issue and
therefore should anticipate this possibility.
Conduct independent oversight
In meeting their oversight responsibilities,
IRBs should explicitly assess risks related
to restrictions on abortion access to par-
ticipants who may become pregnant along
with proposed approaches to minimiz-
ing these risks. This can be complicated
given variability in laws both across states
or countries and within any one state or
country over time. Additional concerns
arise for research overseen by single IRBs
when research is conducted in multiple
states or countries. Single IRBs must be
especially vigilant to ensure that they have
adequate and updated information when
conducting local context reviews (15).
Where necessary, IRBs should consult with
those with appropriate expertise to help
guide their deliberations. IRBs should also
require information about changing laws
and local context as these arise as well as
during the process of continuing review.
In addition, data safety and monitoring
boards should include considerations of
participant safety related to restrictions
on abortion access during their initial and
interval review. In especially restrictive en-
vironments, stopping rules regarding these
issues may be indicated.
CLOSING COMMENTS
Stakeholders involved in research with
participants who could become pregnant
should explicitly consider the points out-
lined here, both to minimize the risks to
participants and staff and to help safeguard
the scientific and social value of research. If
on careful examination it seems implausible
to safely conduct the proposed research at
a particular site, consideration should be
given to conducting the research elsewhere.
Nevertheless, lessons learned about efficient
processes for the safe design and imple-
mentation of research with people who can
become pregnant in the face of restricted
abortion access should be described and
disseminated widely as a means of help-
ing generate best practices. Doing so would
be facilitated by collecting and analyzing
systematic data regarding how often chal-
lenges due to abortion restrictions are en-
p
countered in research as well as how they
are managed. In the meantime, education of
researchers, IRBs, institutional officials and
state and local policy-makers is crucial. Last,
those contemplating the development and
implementation of policies pertaining to
g
abortion should also consider the potential
negative impact on the ability of research-
ers to advance science that can improve the
health and well-being of those who are or
y
may become pregnant and their fetuses. j
R E F E R E N C ES A N D N OT ES
1. Dobbs v. Jackson Women’s Health Organization, 142 S.
Ct. 2228 (2022).
2. R. B. Reingold, L. O. Gostin, JAMA 329, 877 (2023).
3. L. O. Gostin, R. B. Reingold, BMJ 378, o1897 (2022).
4. R. B. Reingold et al., JAMA 328, 1695 (2022).
5. A. C. Mastroianni et al., Hastings Cent. Rep. 47, 38 (2017).
6. C. A. Sewell et al., Am. J. Obstet. Gynecol. 227, 805 (2022).
7. AVAC, Joint United Nations Programme on HIV/AIDS
(UNAIDS), Good Participatory Practice: Guidelines for
Biomedical HIV Prevention Trials (UNAIDS, ed. 2, 2011).
8. L. E. Wolf, L. M. Beskow, Am. J. Law Med. 44, 343 (2018).
y g
9. Department of Health and Human Services, Fed. Regist.
88, 23506 (2023).
10. T. G. M. Sandfort et al., J. Int. AIDS Soc. 23 (suppl. 6),
e25600 (2020).
11. J. Sugarman et al., Lancet HIV 5, e468 (2018).
12. Council for the International Organization of Medical
Sciences (CIOMS) and the World Health Organization,
(WHO), “International Ethical Guidelines for Health
,
Related Research Involving Humans” (CIOMS, ed. 4,
2016), Guideline 18.
13. D. Wendler, J. Rackoff, IRB 24, 1 (2002).
14. J. Sugarman, S. M. Rose, D. Metzger, Clin. Trials 11, 239
(2014).
15. A. R. Johnson et al., J. Clin. Transl. Sci. 6, e53 (2022).
AC K N OW L E D G M E N TS
The authors thank W. Chege for helpful input. All of the
authors are or have been members of the HIV Prevention
Trials Network’s (HPTN) Ethics Working Group. Overall
support for the HPTN is provided by the National Institute
of Allergy and Infectious Diseases, Office of the Director,
National Institutes of Health (NIH), National Institute on Drug
Abuse, and the National Institute of Mental Health under
awards UM1AI068619, UM1AI068617, and UM1AI068613.
This work was supported in part by the NIH Clinical Center
Department of Bioethics. The views expressed here are those
of the authors and do not necessarily reflect the policies of the
NIH or the US Department of Health and Human Services.
10.1126/science.adh3104
science.org SCIENCE

B O OKS et al .
PUBLIC HEALTH
Performing science
A biologist captures the humanity of scientific research
in a play about the AIDS crisis
By H. Holden Thorp
D
epictions of science and scientists in
television, theater, and movies usu-
ally miss the mark. The Big Bang
Theory, for example, is a spoof that
is sometimes degrading to scientists.
Michael Frayn’s highly stylized Co-
penhagen ultimately presents a fictional-
ized account of the lives and work of Niels
Bohr and Werner Heisenberg. Neither of
these projects, nor various others like them,
show the human side of research. In his new
play, Love + Science, biologist David Glass
succeeds in capturing this vital part of the
scientific enterprise.
In the play’s opening scenes, two fic-
tional first-year medical students—Matt
(played by Matt Walker, an accomplished
actor and genetics PhD student at Columbia
University) and Jeff (Jonathan Burke)—fall
in love and end up working in the same lab
at Columbia in the early 1980s. The pair are
supervised by another fictional figure, prin-
PHOTO: EMILIO MADRID
cipal investigator Diane Gold (Thursday
Farrar), a demanding but caring mentor ob-
The reviewer is editor-in-chief of the Science family of jour-
nals, Washington, DC 20005, USA. Email: hthorp@aaas.org
SCIENCE science.org
sessed with the proper execution of positive
and negative controls whose work focuses
on mouse retroviruses that can induce
cancer. (Matt, we learn, sought out Gold to
serve as his mentor after reading a flatter-
ing profile of her in Science.)
Matt and Jeff attend a lecture about
an emerging public health concern, “gay-
related immune deficiency,” and decide that
the medical community’s initial ex-
planation for the condition—that
repeated exposure to sperm from Love + Science
large numbers of same-sex part- David J. Glass
In Vitro Productions,
ners somehow suppresses immune New York City Center, one of someone with AIDS. Matt
function—has to be wrong. When New York, NY, USA, 26 observes dispassionately that the
the pair approach Gold about their May to 6 July 2023. patient is certain to die. Jeff, the
concerns, she agrees that the expla-
nation cannot be right. “So the big news,”
she says, “is we have idiots teaching medical
students.”
But when Matt and Jeff speculate that
the cause must be a virus, Gold cautions
them not to let their own work on viruses
bias their thinking. “When you have a ham-
mer,” she says, “the whole world looks like
a nail.” She agrees, however, that a virus
seems likely.
Eventually, they huddle around a 1983 is-
sue of Science that contains a paper confirm-
The personal and professional are intertwined for
medical students Matt (left) and Jeff.
ing their hunch about the viral origins of the
condition that had come to be known as ac-
quired immunodeficiency syndrome (AIDS)
(1). While the Gold lab does not work on
AIDS directly, the mouse retroviruses they
study provide insight into the mechanism
exploited by human immunodeficiency virus
(HIV), the virus that causes AIDS.
Over the rest of the play, Matt and Jeff
make their way through their respective
careers. Matt decides to forgo residency to
focus on science and eventually becomes a
principal investigator at Columbia. Jeff goes
the clinical route, caring for AIDS patients
in high-need areas. Their romance ends,
but they remain friends. By watching them
decide on and execute their life’s work, we
see how doing science and building a career
p
intersect with identity, trauma, history, and
politics. The play ends with a time jump to
2021, drawing parallels between the AIDS
crisis and the COVID-19 pandemic.
What makes Love + Science special is its
unapologetic commitment to presenting a
g
real story about science. Glass—who is gay
and came of age as a scientist during the
AIDS crisis—brings all his experiences to
bear on the crucial decisions and interper-
y
sonal dynamics that play out on stage. He
makes a point, for example, of showing the
audience that the study of cancer-causing
retroviruses set the stage for the understand-
ing of HIV. And as a strong advocate for re-
producibility and scientific integrity, he cre-
ated a principal investigator dedicated to the
beauty of excellent controls.
Glass also wanted the main character to
be played by a practicing scientist and identi-
y g
fied Matt Walker—who is currently juggling
nightly performances and daytime research
projects with the support of his advisers—for
the role early on. Walker’s perfor-
mance is vital to the play. In one
revealing sequence, Matt and Jeff
,
clash over how to talk to a loved
caring clinician, scolds Matt for not
providing any hope. “This is about how you
speak to patients,” Jeff says, “not about sci-
ence. This is a different thing.”
With Love + Science, Glass has brought
the scientific community something to trea-
sure: a play in which scientists do realistic
research while embracing their humanity.
It’s about time. j
RE FE REN C ES AN D N OT ES
1. F. Barré-Sinoussi et al., Science 220, 868 (1983).
10.1126/science.adj0762
23 JUNE 2023 • VOL 380 ISSUE 6651 1227
I N SI G HT S | B O O K S
ANIMAL BEHAVIOR
An ode to owls
New tools and techniques offer insight into
the lives of these secretive birds
By Alan B. Franklin
O
wls are mysterious creatures, mainly
because we humans hardly ever en-
counter them. They are mostly noc-
turnal, are silent in their flight, and
tend to unnerve us with their ghostly
and sometimes quavering vocaliza-
tions. There have been numerous books on
the natural history of owls written for nonex-
perts, but Jennifer Ackerman’s What an Owl
Knows takes a novel tack, blending natural
history and scientific discovery with a discus-
sion of recent technological innovations. The
result is a fascinating read on how scientists
are beginning to better understand
the lives and ecology of these secre-
tive and rarely visible birds.
Ackerman’s highly readable sci-
ence of owls is made all the more
interesting by her decision to dive
into the backgrounds of the sci-
entists who study them, illustrat-
ing how and where they do their
research. She describes, for exam-
ple, how Japanese neurobiologist
Masakazu Konishi set out to un-
derstand how owls can track their
prey in total darkness and how he
discovered that owls build auditory
maps of their surroundings. Such
stories are often sprinkled with hu-
morous elements. One of Konishi’s
most frequent test subjects, for ex-
ample—an owl he named Roger—
surprised the researcher one day
when “he” laid an egg.
The book’s nine chapters cover
broad areas of knowledge about owls, much
of it recently discovered. In the opening
pages, Ackerman describes how research-
ers are using new approaches to study owls
and how our understanding of their abili-
ties has increased exponentially with tech-
nological advances. In some areas, such as
how owls communicate with each other, we
are now able to explore nuances that were
imperceivable before the advent of advanced
computing systems and artificial intelligence.
Another particularly interesting topic consid-
ered in this section is the trade-offs inherent
The reviewer is at the USDA/APHIS/WS National
Wildlife Research Center, Fort Collins, CO 80521, USA.
Email: alan.b.franklin@usda.gov
1228 23 JUNE 2023 • VOL 380 ISSUE 6651
What an Owl Knows:
The New Science of the
World’s Most Enigmatic
Birds
Jennifer Ackerman
Penguin, 2023. 352 pp.
in an owl’s feather coloration and the role
that energetics and melanin play in balanc-
ing plumage coloring. The reason many owls
have barred feathers—feathers with distinct
bars or lines—is that barring provides an
economy of nature; the dark bars are forti-
fied with melanin, which provides strength
but requires considerable energy to produce,
while the intervening areas of light colors
are less strong but are also energetically less
costly to produce and weigh less.
The chapters that follow describe the
breeding behaviors of different species of
owls, their roosting behaviors, how owls
sleep, and their varying migratory strategies.
A pair of barred owl chicks rests momentarily on a branch.
Here again, new technologies, such as the
Motus Wildlife Tracking System and satellite
transmitters, are helping researchers answer
key questions about owl migration, nomad-
ism, and irruptive behavior.
Ackerman next focuses on what captive
owls have taught us. Here, we learn that
Teddy Roosevelt, Florence Nightingale, and
Pablo Picasso all had pet owls. Although this
chapter is short on science, it provides in-
sights into the distinctive and individual per-
sonalities of owls and the role that rehabili-
tation centers play in helping us understand
their psychology.
The book then delves into the place that
owls hold in human culture, from appear-
ing in cave paintings in France to the rever-
ence with which they have been regarded by
many cultures, who have considered them
both gods and messengers from the gods.
Here, Ackerman describes a massive survey
undertaken by a team led by David Johnson,
founder of the Global Owl Project, to illumi-
nate cultural understandings of owls. More
than 6000 interviews were conducted with
participants in 26 countries, who were asked
questions that varied from the scientific
(“How much do you know about owls, their
ecology, their habitats, nesting, and diet?”)
p
to the spiritual (“What does your culture
believe about owls?”). Attitudes toward owls
varied, but the birds were often regarded as
inauspicious and, in some cases,
even harbingers of death.
The final chapter focuses on how
g
the brains of owls work and delves
into theories that owls use hippo-
campal-based spatial memory to
navigate at night. Here, Ackerman
y
describes the differences between
avian and mammalian brains and
then how owls differ from other
avian species—dedicating as much
as 75% of the cortex-like portion of
their brains to vision and hearing—
primarily because of their noctur-
nal lifestyle.
In the book’s afterword about
conservation of owls, Ackerman
y g
devotes a single sentence to the
plight of the northern spotted owl,
a threatened North American sub-
species, but offers little context
about the factors undermining its
survival, such as habitat loss and
,
displacement by invasive barred owls. This
felt like a missed opportunity to dig into
the tensions between owl conservation and
human development and the unsettling no-
tion that barred owls may have to be lethally
removed to save spotted owls—both conun-
drums that warrant serious discussion.
Ultimately, however, this book provides
a fascinating glimpse into how dedicated
researchers have deployed new technology
and approaches to discover how owls fulfill
their distinctive roles as nocturnal predators.
Although written for nonexperts, this book
will be of interest to scientists from a broad
array of disciplines, from neurobiology to
general ecology. j
10.1126/science.adi4560
science.org SCIENCE
 LET TERS
An ecoacoustic recording station documents the soundscape in West Cork, Ireland.
Edited by Jennifer Sills
Sound stewardship for
a noisy planet
Human-caused noise pollution pervades
many types of ecosystems (1, 2) but
receives lower prioritization than other
forms of pollution in policy deliberations
about global environmental change. To
protect species that depend on sound
to communicate and derive information
about their environment, noise pollu-
tion mitigation must be integrated into
the global change research and policy
agenda. The new UN Environmental
Programme Science-Policy Panel
on Chemicals, Waste, and Pollution
Prevention is currently developing its
scope, and noise pollution should be
included.
In aquatic ecosystems (3), the noise
pollution caused by ship traffic and
global economic growth have had an
increasing impact on marine life (4).
Aquatic biota use sound to navigate, to
avoid predators, and to locate habitat,
prey, mates, and offspring. Sound pollu-
tion can mask these environmental cues
PHOTO: MARCUS MAEDER/CRESPO FOUNDATION
and reduce survival (4). Noise affects
terrestrial species, such as birds (5), by
degrading habitat quality and behavior.
Belowground soundscapes influence how
species, especially animals, behave in
the soil (6). Noise pollution also affects
human health and life quality by influ-
encing sleep, cardiovascular health, cog-
nition, and mental health (7).
To reduce noise, policy measures
SCIENCE science.org
should favor the development and imple-
mentation of less noisy alternatives in
the transportation, construction, and
industrial sectors (8). Sound sanctuaries
(9)—areas surrounded by sound barri-
ers, within which anthropogenic noise
is limited—could shield ecosystems
that are particularly sensitive to human
noise. Soundscape stewardship must
also involve active interventions, such as
sound generation and exposure methods
that revitalize degraded soundscapes in
ecosystem restoration (10). For example,
recordings of native bird calls could be
played in a restored area where those
birds no longer live. Beneficial interven-
tions also include recovery efforts and
urban soundscape improvements (11).
Efforts to raise public awareness about
the harmful effects of noise pollution will
be crucial. Finally, multidisciplinary col-
laborations could work toward a proac-
tive and sustainable soundscape composi-
tion. Scientists, policymakers, and artists,
for example, could create natural as well
as artistically designed artificial sound
sources that help to reinstall or enhance
local community compositions and habi-
tats (12).
To address the global noise pollution
challenge, scientists should investigate
the ecosystem-level effects of noise and
the interactions between noise pollution
and other global change factors such as
chemical pollution and climate change.
A noise analysis should be added to the
study of planetary boundaries. A guid-
ing principle of planetary soundscape
stewardship must harness the potential
of precautionary approaches, recognizing
p
that natural sounds are an integral part of
planetary biodiversity and ecosystems. We
can and must make our planet a quieter
g
and healthier place for ourselves, for eco-
systems, and for future generations.
Matthias C. Rillig1,2*, Michael S. Bank3,4, Stefanie
Maaß1,2, Mélia Roger5, Marcus Maeder5,6
y
1
Freie Universität Berlin, Institute of Biology, 14195
Berlin, Germany. 2Berlin-Brandenburg Institute
of Advanced Biodiversity Research (BBIB),
14195 Berlin, Germany. 3Institute of Marine
Research, 5005 Bergen, Norway. 4Department
of Environmental Conservation, University of
Massachusetts Amherst, Amherst, MA 01003,
USA. 5Zurich University of the Arts, Institute
for Computer Music and Sound Technology,
8031 Zürich, Switzerland. 6Department of
Environmental Systems Science, Institute
for Environmental Decisions, Eidgenössische
Technische Hochschule Zürich, 8092 Zürich,
y g
Switzerland.
*Corresponding author.
Email: rillig@zedat.fu-berlin.de
RE FE REN C ES AN D N OT ES
1. R. M. Schafer, The Soundscape: Our Sonic Environment
and the Tuning of the World (Destiny Books, 1993).
,
2. E. Murphy, E. A. King, Environmental Noise Pollution:
Noise Mapping, Public Health, and Policy (Elsevier, ed. 2,
2022).
3. C. M. Duarte et al., Science 371, eaba4658 (2021).
4. A. N. Popper, A. Hawkins, Eds., The Effects of Noise on
Aquatic Life II (Springer New York, 2016), vol. 875 of
Advances in Experimental Medicine and Biology.
5. H. E. Ware et al., Proc. Natl. Acad. Sci. U.S.A. 112, 12105
(2015).
6. M. Maeder, X. Guo, F. Neff, D. Schneider Mathis, M. M.
Gossner, PLOS ONE 17, e0263618 (2022).
7. C. Clark, C. Crumpler, H. Notley, Int. J. Environ. Res.
Public Health 17, 393 (2020).
8. J. W. Smith, B. C. Pijanowski, Glob. Environ. Change 8, 63
(2014).
9. S. L. Dumyahn, B. C. Pijanowski, Landsc. Ecol. 26, 1327
(2011).
10. T. A. C. Gordon et al., Nat. Commun. 10, 5414 (2019).
11. P. Jennings, R. Cain, Appl. Acoust. 74, 293 (2013).
12. B. Martin, Org. Sound 23, 20 (2018).
10.1126/science.adi3600
23 JUNE 2023 • VOL 380 ISSUE 6651 1229
I NS I GHTS | L E T T E R S
China’s water diversion
carries invasive species
The Chinese government has constructed
the world’s largest water conservancy
initiative, the South-to-North Water
Diversion Project, to address drought and
water scarcity in northern China (1, 2).
The project encompasses an area with 438
million inhabitants and spans a total of
4350 km across its eastern, middle, and
western routes. As of March, the cumula-
tive water diversion volume of the eastern
and middle routes has surpassed 61.2 bil-
lion cubic meters (3). Although the project
has generated substantial economic bene-
fits, it has also provided a path for biologi-
cal invasions (4). The Chinese government
must take urgent action to stop invasive
species from infiltrating water diversion
canals and mitigate the harm they cause if
they are transported to northern regions.
The water diversion project has trans-
ported a variety of invasive species to
northern China. The bearded worm
goby (Taenioides cirratus), native to the
Yangtze River mouth (5), typically spawns
during spring and summer (6), coincid-
ing with the water diversion period. As
a result, the canals carry numerous fish
eggs northward (7), infiltrating inland
lakes, with Chaohu Lake suffering par-
ticularly severe consequences (8). The
bearded worm goby is aggressive, repro-
duces frequently (5, 6), and has wolflike
teeth that ruthlessly prey on local fish
populations.
Aquatic plants such as alligator weed
(Alternanthera philoxeroides), water
hyacinth (Eichhornia crassipes), and
water lettuce (Pistia stratiotes) have also
invaded the northern water systems (9).
The water bodies are rich in nutrients
like nitrogen and phosphorus, which
further promote plant growth (10, 11).
Those plants deplete substantial amounts
of nutrients and oxygen from the riv-
ers, resulting in the death of numerous
local fish due to oxygen deficiency (12).
These invasive species may alter local
agriculture and ecology, posing a threat to
the objectives of the water transfer.
To safeguard the northern ecosystem
from the costly negative impacts of inva-
sive species, the Chinese government
must implement screening and continu-
ous monitoring of aquatic organisms in
susceptible areas, including the headwa-
ters, routes, and destinations of water
diversion projects. Such efforts should
aim to either bar invasive organisms from
entry into the canal or eradicate them
before they propagate within the canal.
1230 23 JUNE 2023 • VOL 380 ISSUE 6651
Furthermore, the government and water
diversion project managers should investi-
gate the overlap between water diversion
periods and invasive species breeding sea-
sons and adjust the water diversion sched-
ule to avoid that window of time.
Kai Zhu*, Yufeng Cheng, Quan Zhou
Faculty of Resources and Environmental Science,
Hubei University, Wuhan 430062, China.
*Corresponding author. Email: hizhukai@163.com
R EFER ENC ES AN D N OT ES
1. Y. Li, W. Xiong, W. Zhang, C. Wang, P. Wang, Water Res. 89,
9 (2016).
2. Z.-Y. Zhao, J. Zuo, G. Zillante, J. Clean Prod. 163, 136
(2017).
3. China South-to-North Water Diversion Corporation
Limited, “The South-to-North Water Diversion
Project’s eastern and central routes transferred
18.57 billion cubic meters of water in the first quarter,
with a cumulative water transfer of over 61.2 billion
cubic meters,” State-owned Assets Supervision and
Administration Commission of the State Council
(2023); http://www.sasac.gov.cn/n2588025/
n2588124/c27728909/content.html [in Chinese].
4. “The ‘South-to-North Water Diversion’ project has
become a ‘highway’ for biological invasion. What can we
do?” Wangyi News (2023); https://www.163.com/dy/
article/HTKFM4G20517Q7DN.html [in Chinese].
5. J. Qin et al., Manag. Biol. Invasion. 10, 139 (2019).
6. Y. Liang et al., J. Appl. Ichthyol. 36, 219 (2020).
7. C. Guo et al., Sci. Tot. Environ. 713, 136515 (2020).
8. J. Qin, S. Bjorn Victor, L. Zhang, F. Cheng, S. Xie, Freshw.
Biol. 67, 2078 (2022).
9. D. Liu et al., Environ. Sci. Technol. 51, 1450 (2017).
10. X. Hu et al., J. Environ. Manag. 318, 115561 (2022).
11. S. Zhang et al., Environ. Res. 227, 115805 (2023).
12. W. Xia et al., J. Environ. Manag. 319, 115726 (2022).
10.1126/science.adi6022
New farmlands threaten
the North China leopard
To ensure food security, China released
a high-standard farmland construction
plan in 2021 (1). According to the plan,
China aims to build 26.67 million hectares
Planned farms will further fragment the already
compromised habitat of the North China leopard.
of new high-standard farmland by 2030.
Because urbanization has encroached
upon lands previously used for agriculture
(2), the plan designated hilly and moun-
tainous regions for farmland develop-
ment (3). These regions are not suitable
for farming, so crop yield is likely to be
low (4). Moreover, the areas targeted for
conversion to farmland include the habi-
tat of the North China leopard (Panthera
pardus japonensis), a rare leopard sub-
species endemic to China. To protect the
North China leopard from extinction,
the farmland construction taking place
in its distribution area should be halted
immediately.
The North China leopard has lost about
98% of its historical habitat as a result of
anthropogenic activities such as hunting
and poaching, agricultural and industrial
development, and infrastructure construc-
p
tion (5). Only a few hundred individuals
remain (6). Because of its fragmented dis-
tribution and small population, the North
China leopard was categorized on China’s
list of National Class I key protected wild-
life (6). If assessed independently instead
g
of as part of the worldwide Panthera
pardus population, the Panthera pardus
japonensis subspecies would likely meet
the criteria for classification as Critically
y
Endangered by the International Union
for Conservation of Nature (6).
The North China leopard needs
increased protection. In addition to end-
ing farm construction in leopard habitat,
China should establish more natural
reserves for the surviving leopards, many
of which have been found outside pro-
tected areas (7, 8). More ecological cor-
ridors are also needed to foster gene flow
y g
between otherwise isolated subpopula-
tions. To safeguard human food security,
China should work to conserve food and
restore farmland in flatlands, where fewer
extant wild species will be threatened.
Xianghong Dong1 and Tao Ju2*
,
1
College of Animal Science, Guizhou University,
Guiyang 550025, China. 2Guangxi Academy of
Sciences, Nanning 530007, China.
*Corresponding author.
Email: lanqiuzhejutao@sina.com
RE FE REN C ES AN D N OT ES
1. “Notice on issuance of national high standard farm-
land construction plan (2021–2030) by the Ministry
of Agriculture and Rural Affairs of the People’s
Republic of China” (2021). http://www.moa.gov.cn/
hd/zbft_news/qggbzntjsgh/xgxw_28866/202109/
t20210916_6376566.htm [in Chinese].
2. C. He et al., Sci. Tot. Environ. 576, 660 (2017).
3. H. Chen et al., Sci. Tot. Environ. 831, 154895 (2022).
4. X. Cui et al., Land 11, 488 (2022).
5. K. Vitekere et al., J. Anim. Plant Sci. 31, 1 (2021).
6. A. Laguardia et al., Oryx 51, 153 (2015).
7. D. Song et al., Biodivers. Sci. 22, 733 (2014) [in Chinese].
8. D. Song, For. Humankind 1, 22 (2019) [in Chinese].
10.1126/science.adi7209
science.org SCIENCE
 RESEARCH
IN S CIENCE JOU R NA L S
Edited by Michael Funk

p
ARCTIC ECOLOGY

Herbivore diversity reduces climate effects

A
rctic tundra is experiencing rapid climate change, including was mainly explained by losses of sea ice with ambient warming.

warming temperatures and loss of sea ice. Plants and
herbivores are both affected by these abiotic changes. Post
et al. examined the effects of climate change and herbivory
on the diversity of tundra plants, fungi, and lichens using a
g
However, herbivores tempered this decline in plots where they
were not excluded, particularly under experimental warming.
The two herbivores studied, caribou and muskoxen, had different
effects on the understory, with a positive effect of increasing

y
15-year warming and herbivore exclusion experiment. They found muskoxen (and thus herbivore diversity) on tundra diversity. —BEL
that diversity decreased over time across all treatments, which Science, add2679, this issue p. 1282

Muskoxen, pictured above near Kangerlussuaq, Greenland, contribute to preserving biodiversity in tundra ecosystems.

POLYMERS physical cross-links that further also found that a binding protein proposal experimentally in a two-
toughen the polymer. —MSL called tropomodulin stabilizes species system of neutral atoms
Enhancing toughness Science, adg3229, this issue p. 1248 the pointed end and a protein placed in optical tweezers. The
by being weak

Tough materials can sustain
substantial subcritical damage
without complete failure of the
material, but trying to improve
toughness can often lead to the
y g
called cap-Z binds at the barbed researchers demonstrated the
end. —MAF success of the procedure by
STRUCTURAL BIOLOGY
Science, adg6812, this issue p. 1287 mitigating the effects of injected
Actin reveals its ends magnetic field noise. —JS
Actin is a key protein in cellular Science, ade5337, this issue p. 1265
QUANTUM INFORMATION
motility, forming filaments that

degradation of other mechani-
cal properties. Wang et al.
designed elastomeric polymer
networks that use a mecha-
nophore as a cross-linker. The
mechanophore cross-links are
much easier to mechanically
break than the ordinary car-
bon–carbon bonds in polymer
strands or the second, stronger
cross-linker used as a control.
The mechanophore cross-links
function as sacrificial bonds to
elongate rather than break the
bridge chains. When the cross-
link density is low, the primary
backbones can form tortuous
give dynamic structure to cellular
appendages and serve as scaf-
fold for motor proteins. The ends
of these filaments are constantly
extending and dissociating,
and their structure has been
difficult to capture. Carman et al.
optimized a mixture of binding
proteins so that they could visual-
ize the ends in different forms.
The structures revealed the inter-
nal conformations of the actin
subunits at the ends, and the
authors discuss how differences
from the main filament structure
facilitate differential addition or
loss of subunits at the ends. They
Spectators to the rescue
Qubits, the quantum informa-
tion version of bits, are prone to
decoherence as a consequence
of their interaction with the
environment. One way to man-
age decoherence is to correct for
its effects. Recently, a scheme
was proposed in which so-called
“spectator” qubits are embed-
ded in the system and used to
read out the accumulated noise.
This information is then used in
real time to counter the decoher-
ence of the “data” qubits that
are performing the computation.
Singh et al. implemented this
,
DENDRITIC CELLS
Lung development and
immunity intertwined
Dendritic cells (DCs) can
participate in processes under-
pinning immune tolerance.
Silva-Sanchez et al. found that
a discrete population of mature
DCs accumulated transiently
in the lungs and lung-draining
lymph nodes of neonatal mice
during a period of elevated
apoptosis associated with the
postnatal development of alveoli.
These DCs, distinguished by

1232 23 JUNE 2023 • VOL 380 ISSUE 6651 science.org SCIENCE

 intermediate expression levels of
the integrin CD103, could restrain
CD8+ T cell responses, had a
transcriptome that was similar to
that of regulatory and tolerogenic
DCs, and underwent a distinct
maturation process that was
dependent on the receptor Mertk
and the engulfment of apoptotic
cells. The results of this study
indicate that tissue remodeling
may stimulate the maturation
of tolerogenic DCs to restrain
immune responses and contrib-
ute to suppressed pulmonary
immunity in neonates. —SHR
Sci. Immunol. (2023)
10.1126/sciimmunol.adc9081
NEUROSCIENCE
movies of spike propagation
IN OTHER JOURNALS Edited by Caroline Ash
and subthreshold dynamics in
and Jesse Smith
Drosophila neurons at kilohertz
frame rates and picosecond
lifetime resolution. —PRS
Science, adf9725, this issue p. 1270
BIOENGINEERING
A point-of-care HPV
test for all
DNA-based screening methods
for high-risk human papilloma-
virus (HPV) have proven to be
highly effective for early cervical
cancer detection but remain
too expensive and difficult to
use in resource-limited set-
tings. Kundrod et al. developed
a prototype, sample-to-answer,

Recording neuron activity
in lifetimes
Fluorescence lifetime measure-
ment is a promising technique
for voltage imaging. However,
p
point-of-care test for detecting
HPV16 and HPV18. This test uses
isothermal DNA amplification
and lateral flow detection with
low-cost components that can
detect HPV DNA at a clinically

presently available applications
fail to provide the throughput
and sensitivity required for
functional imaging. Bowman et
g
relevant limit of detection. This
test was successfully demon-
strated in a high-resource setting
in the United States and in a low-

al. developed an electro-optic
fluorescence lifetime microscopy
(EO-FLIM) method with a mas-
sive improvement in throughput
y
resource setting in Mozambique, NEUROSCIENCE
and it could be performed by
minimally trained personnel at Stellar timekeeping
an estimated cost of less than $5

over existing techniques.
Lifetime readout decisively
improved spike detection fidelity
and threshold voltage stability
while significantly suppressing
motion artifacts. This approach
US per test. These results sug-
gest that a sample-to-answer,
point-of-care HPV DNA test is
viable for low-resource settings
and could improve access to
cervical cancer screening. —CNF
D
isruption of physiological daily rhythms (circadian
clocks) is associated with an increased risk of developing
Alzheimer’s disease and other neurological disorders. In
the brain, the suprachiasmatic nucleus (SCN) regulates
circadian rhythms, but it is not known how it does so.
Patton et al. show that astrocytes in the SCN control circadian
rhythms by modulating neuronal activity. During the circa-

enabled the detection of action
potentials in vivo, capturing
Neurons visualized within the
brain of a fruit fly using composite
Sci. Transl. Med. (2023)
10.1126/scitranslmed.abn4768
y g
dian day, astrocytes remove the inhibitory neurotransmitter
GABA, which is released by SCN neurons into the extracellular
space to facilitate neuronal firing. During the circadian night,
GABA uptake is reduced, resulting in increased inhibition and
decreased firing. —MMa
Proc. Natl. Acad. Sci. U. S. A. (2023) 10.1073/pnas.2301330120

,
CREDITS (LEFT TO RIGHT): CHRISTOPH MELCHER; SELVANEGRA/ISTOCKPHOTO

fluorescence microscopy
Neuronal dysfunction of astrocytes (shown above) in the brain alters
circadian rhythms and is associated with neurodegenerative diseases.

MANUFACTURING poor adhesion at the inter-

Overcoming interlayer
adhesion
Stereolithographic additive
manufacturing enables free-
form printing of complex shapes
from a pool of resin. This works
well for amorphous polymers
that form cross-linked networks,
but for semicrystalline ones,
faces between layers ruins the
mechanical strength. However,
if successful, semicrystalline
polymers could lead to parts
with greater toughness and
solvent resistance. Commissio
et al. printed seven- and eight-
membered cyclic allylic sulfides,
which rapidly polymerized and
crystallized upon irradiation. The

SCIENCE science.org 23 JUNE 2023 • VOL 380 ISSUE 6651 1233


ALSO IN SCIENCE JOURNALS
TOPOLOGICAL MATTER
On the trail of elusive
quasiparticles
A definitive discovery of
Majorana quasiparticles would
bring the potential of topological
quantum computing closer to
reality. In the original propos-
als, the “recipe” for observing
Majoranas experimentally
appeared deceptively simple.
In the intervening years, it has
become clear that the real
world is more complicated
than the models predicted,
and Majoranas remain elusive.
Yazdani et al. review our growing
understanding of a very complex
topic and speculate on the most
promising directions for the
future. —JS
Science, ade0850, this issue p. 1235
CELL BIOLOGY
Mitofusin functions
beyond mitochondria
Inside eukaryotic cells, different
types of organelles interact at
membrane contact sites that
are stabilized by protein bridges
called tethers. The mitochon-
drial fusion protein mitofusin 2
(MFN2) links mitochondria to
an unknown partner on the
endoplasmic reticulum (ER).
Naón et al. found two splice
variants of mitofusin 2, ERMIT2
and ERMIN2. ERMIN2 affected
ER morphology, and ERMIT2
was localized to the ER and
enriched at the contact sites
with mitochondria. Treatment
with adenovirus encoding for
ERMIT2 allowed for the transfer
of calcium and lipids between
the two organelles and reduced
liver damage in models of
reduced MFN2 expression.
—SMH
Science, adh9351, this issue p. 1237
CELL BIOLOGY
There’s a NAC for that
All proteins are initially synthe-
sized with methionine as their
first amino acid. Membrane
SCIENCE science.org
Edited by Michael Funk
proteins and proteins translo-
cated across the endoplasmic
reticulum retain this methio-
nine, but most cytosolic
proteins have it removed by
methionine aminopeptidase 1.
The mechanism through which
METAP1 discerns its target
proteins has remained elusive.
Gamerdinger et al. revealed
the crucial role of the nascent
polypeptide–associated com-
plex (NAC) in this process. NAC
binds to the ribosome to guide
the aminopeptidase to the
correct proteins, ensuring the
removal of methionine residues
from cytosolic proteins while
sparing those targeted to the
endoplasmic reticulum. —SMH
Science, adg3297, this issue p. 1238
SUPERCONDUCTIVITY
Vortices go fractional
Superconductors can expel
weak external magnetic
fields. However, when the field
becomes sufficiently strong, it
can penetrate the supercon-
ductor, creating vortices that
each correspond to a quantized
magnetic flux. For a multiband
superconductor, the picture is
more complicated because the
vortices can also carry a frac-
tion of the unit flux. Iguchi et al.
used superconducting quantum
interference device magne-
tometry to observe signatures
of such fractional flux quan-
tum vortices in the material
Ba1−xKxFe2As2. The results may
find practical use in developing
superconducting devices. —JS
Science, abp9979, this issue p. 1244
BIOMATERIALS
An open and shut case
The calcareous hinge in bivalve
mussels must deform repeat-
edly without sustaining damage
while also providing elastic
energy storage to enable the
return of the valve to its open
configuration. Meng et al. used a
combination of mechanical test-
ing and microscopy to analyze
the fatigue-resistant properties
of mussel hinge tissue (see
the Perspective by Crane and
Denny). The authors showed
that the resistance arises from
radially aligned aragonite nano-
wires embedded in an organic
matrix. This geometry allows for
the translation of external loads
to the circumference. The hard-
soft composite also suppresses
stress localizations during valve
motion. —MSL
Science, ade2038, this issue p. 1252;
see also adi5939, p. 1217
MOLECULAR BIOLOGY
WAPL regulates
neuronal diversity
Neural self-nonself discrimina-
tion requires the generation of
cell surface “barcoding” diver-
sity. Neurons expressing the
same barcode avoid each other
to maximize interactions with
other neurons. In mammals,
clustered Protocadherin (Pcdh)
proteins function as barcodes.
By studying the expression
regulation of Pcdh genes in
mouse serotonergic and olfac-
tory sensory neurons, Kiefer et
al. uncovered that the combi-
natorial space of Pcdh isoform
diversity in individual cells rests
on the DNA translocation activ-
ity of cohesin and its unloader,
WAPL. These findings expand
the repertoire of molecular logic
that generates the protein iso-
form diversity required for cell
fate and function. —DJ
Science, adf8440, this issue p. 1236
CELLULAR IMMUNOLOGY
Endosomal escape in
cross presentation
Endocytic compartments
in dendritic cells (DCs) are
unusually leaky, which allows
exogenous proteins to be fed
to the ubiquitin-proteasome
degradation system. The result-
ing peptides are then passed
to the major histocompatibility
complex
class I presentation pathway,
23 JUNE 2023 • VOL 380 ISSUE 6651
R E S E A RC H
resulting in antigen cross-pre-
sentation and priming of CD8+
T cells specific to exogenous
antigens not expressed in DCs.
Rodríguez-Silvestre et al. found
that type 1 conventional DCs
express a dedicated effector of
escape, a pore-forming protein
called perforin-2. Perforin-2
pores in DC endosomes deliver
antigens into the cytosol for
cross-presentation (see the
Perspective by Rawat and
Jakubzick). The restricted
expression of perforin-2 thus
provides a mechanistic explana-
tion for the high efficiency of
p
endocytic escape in cross-pre-
senting DCs. —SMH
Science, adg8802, this issue p. 1258;
see also adi5711, p. 1219
g
PLANT SCIENCE
Cell layers build tissue
shapes
y
Cell layers can slip, slide, and
fold in among one another dur-
ing development to generate the
complex forms of the biological
world. Studying a small aquatic
carnivorous plant as well as
the land-locked mustard plant
Arabidopsis, Kelly-Bellow et al.
built a connection from genes
driving a hormone’s biosyn-
y g
thesis to the curved shapes of
the mature plants. The plant
hormone brassinosteroid
reduces mechanical constraints
imposed by the plant’s epider-
mis, thus allowing internal cell
,
layers to drive the formation of
new shapes with development.
—PJH
Science, adf0752, this issue p. 1275
CALCIUM SIGNALING
Sticking sensitizes T cell
activation
During the activation of T cells,
intracellular calcium (Ca2+)
globally increases and leads to
changes in gene expression by
promoting the nuclear transloca-
tion of the transcription factor
NFAT1. Weib et al. found that
1234-B
R ES E ARCH | I N S C I E N C E J O U R NA L S

T cell activation was facili-

tated by the formation of Ca2+
microdomains that occurred
when mouse T cells adhered to
extracellular matrix components
that might be typically encoun-
tered during migration to a site of
inflammation. These Ca2+ micro-
domains were critical for the
global increase in intracellular
Ca2+ and the nuclear transloca-
tion of NFAT1 after T cell receptor
stimulation. —WW
Sci. Signal. (2023)
10.1126/sciscignal.abn9405

p
g
y
y g
,

1234-C 23 JUNE 2023 • VOL 380 ISSUE 6651 science.org SCIENCE

 intermediate expression levels of
the integrin CD103, could restrain
CD8+ T cell responses, had a
transcriptome that was similar to
that of regulatory and tolerogenic
DCs, and underwent a distinct
maturation process that was
dependent on the receptor Mertk
and the engulfment of apoptotic
cells. The results of this study
indicate that tissue remodeling
may stimulate the maturation
of tolerogenic DCs to restrain
immune responses and contrib-
ute to suppressed pulmonary
immunity in neonates. —SHR
Sci. Immunol. (2023)
10.1126/sciimmunol.adc9081
NEUROSCIENCE
movies of spike propagation
IN OTHER JOURNALS Edited by Caroline Ash
and subthreshold dynamics in
and Jesse Smith
Drosophila neurons at kilohertz
frame rates and picosecond
lifetime resolution. —PRS
Science, adf9725, this issue p. 1270
BIOENGINEERING
A point-of-care HPV
test for all
DNA-based screening methods
for high-risk human papilloma-
virus (HPV) have proven to be
highly effective for early cervical
cancer detection but remain
too expensive and difficult to
use in resource-limited set-
tings. Kundrod et al. developed
a prototype, sample-to-answer,

Recording neuron activity
in lifetimes
Fluorescence lifetime measure-
ment is a promising technique
for voltage imaging. However,
p
point-of-care test for detecting
HPV16 and HPV18. This test uses
isothermal DNA amplification
and lateral flow detection with
low-cost components that can
detect HPV DNA at a clinically

presently available applications
fail to provide the throughput
and sensitivity required for
functional imaging. Bowman et
g
relevant limit of detection. This
test was successfully demon-
strated in a high-resource setting
in the United States and in a low-

al. developed an electro-optic
fluorescence lifetime microscopy
(EO-FLIM) method with a mas-
sive improvement in throughput
y
resource setting in Mozambique, NEUROSCIENCE
and it could be performed by
minimally trained personnel at Stellar timekeeping
an estimated cost of less than $5

over existing techniques.
Lifetime readout decisively
improved spike detection fidelity
and threshold voltage stability
while significantly suppressing
motion artifacts. This approach
US per test. These results sug-
gest that a sample-to-answer,
point-of-care HPV DNA test is
viable for low-resource settings
and could improve access to
cervical cancer screening. —CNF
D
isruption of physiological daily rhythms (circadian
clocks) is associated with an increased risk of developing
Alzheimer’s disease and other neurological disorders. In
the brain, the suprachiasmatic nucleus (SCN) regulates
circadian rhythms, but it is not known how it does so.
Patton et al. show that astrocytes in the SCN control circadian
rhythms by modulating neuronal activity. During the circa-

enabled the detection of action
potentials in vivo, capturing
Neurons visualized within the
brain of a fruit fly using composite
Sci. Transl. Med. (2023)
10.1126/scitranslmed.abn4768
y g
dian day, astrocytes remove the inhibitory neurotransmitter
GABA, which is released by SCN neurons into the extracellular
space to facilitate neuronal firing. During the circadian night,
GABA uptake is reduced, resulting in increased inhibition and
decreased firing. —MMa
Proc. Natl. Acad. Sci. U. S. A. (2023) 10.1073/pnas.2301330120

,
CREDITS (LEFT TO RIGHT): CHRISTOPH MELCHER; SELVANEGRA/ISTOCKPHOTO

fluorescence microscopy
Neuronal dysfunction of astrocytes (shown above) in the brain alters
circadian rhythms and is associated with neurodegenerative diseases.

MANUFACTURING poor adhesion at the inter-

Overcoming interlayer
adhesion
Stereolithographic additive
manufacturing enables free-
form printing of complex shapes
from a pool of resin. This works
well for amorphous polymers
that form cross-linked networks,
but for semicrystalline ones,
faces between layers ruins the
mechanical strength. However,
if successful, semicrystalline
polymers could lead to parts
with greater toughness and
solvent resistance. Commissio
et al. printed seven- and eight-
membered cyclic allylic sulfides,
which rapidly polymerized and
crystallized upon irradiation. The

SCIENCE science.org 23 JUNE 2023 • VOL 380 ISSUE 6651 1233

R ES E A RC H | I N O T H E R J O U R NA L S

NAVIGATION

Finding the way home

T
he Saharan desert ant (Cataglyphis
fortis) lives in harsh salt pan environ-
ments in Tunisia. They are impressive
navigators, traveling long distances
over barren surfaces in search of
scattered shrubs, where they forage for
food, then relying on path integration to
return home. Freire et al. discovered that
in featureless flat salt pan habitats, these
ants build taller nest hills than in more
structured habitats. The ants use the nest
hills as a visual guide to find their way
home. Remarkably, the ants rebuilt the
nest hills if the old ones were flattened, but
did not do so when the scientists replaced
the anthills with artificial pillars. —DJ

p
Curr. Biol. (2023) 10.1016/j.cub.2023.05.019

In flat, uniform territory, the desert ant Cataglyphis

g
fortis builds taller nest mounds as a navigational aid.

printed parts showed mechani-
cal properties that approached
those of bulk copolymers.
Further, because the materials
are thermoplastics, the parts can
be melted down and the polymer
recycled. —MSL
Chem. Mater. (2023)
10.1021/acs.chemmater.2c03352
y
not genetic. However, weakened metabolism, but the shift actu- entry points. Perivascular cells
kinship ties also drove increases ally improved cardiac function. identified as pericytes appeared to
in living in institutional settings —YN mark hotspots of invasion. —SMH
among the elderly, destitute, and Circulation (2023) Nat. Commun. (2023)
infirm. —BW 10.1161/CIRCULATIONAHA.122.062166 10.1038/s41467-023-38706-z
Q. J. Econ. (2023)
10.1093/qje/qjad018
CELL BIOLOGY VOLCANOLOGY
HEART FAILURE Hotspots for blood vessel Spectacular
Flex fuel for the heart invasion sedimentation

ADDITIVE ECONOMICS
Benefits of breaking
family ties
Bans on marriage among first
To keep up with the demands of
continuously pumping blood, the
heart consumes large amounts
of energy. This energy is typically
provided by fatty acids moreso
After delivery by a mosquito
bite, sporozoites of the malaria-
causing parasite Plasmodium
actively migrate within the skin.
Subsequently, they invade blood
y g
Caldera-forming explosive erup-
tions produce huge amounts of
ash that fall across large areas, in
addition to dramatic pyroclastic
flows. Although their dynamics

cousins break down tightly knit
family structures that constrain
mobility, leading to increased
migration off of farms and
into urban areas and higher-
paying occupations. Ghosh et
al. integrated US census data
with records of millions of US
marriages from the 18th to 20th
centuries to show that when
states instituted bans, people
from families with higher pre-ban
rates of cousin marriage spread
across a broader range of places
and pursued a broader range of
occupations, indicating that the
drivers were social and cultural,
than glucose, but the healthy
heart is metabolically flexible
and can adapt to the nutrients
available. By contrast, the
failing heart is thought to lack
metabolic flexibility and to shift
toward glucose consumption.
Surprisingly, though, Watson et
al. discovered that even failing
hearts can be metabolically
flexible. The authors tracked the
metabolism of glucose or fatty
acids after the infusion of the
corresponding substrate in 20
patients with heart failure and
discovered that not only was
there a shift toward fatty acid
vessels to infect the liver, but
they are successful only 20% of
the time. Formaglio et al. used in
vivo imaging in a rodent malaria
model to assess the strategies
used by Plasmodium sporozoites
to find and invade blood vessels.
Modeling and observation of
parasite behavior indicated that
sporozoites in the skin used an
intermittent search strategy,
switching between high- and
low-motility states. Sporozoites
that successfully invaded blood
vessels initially displayed reduced
motility suggestive of local scan-
ning of the vasculature to locate
,
are not well understood, terraced
ash deposits are left behind from
previous aerial and submarine
eruptions. Gilchrist et al. used
analog experiments to put better
constraints on the sedimentation
of ash from submarine caldera-
forming eruptions. The structure
of the terraced ash deposits
constrains the source parameters
of the caldera-forming eruption
column and may help to outline
better the impact of large subma-
rine eruptions on the ocean and
seafloor. —BG
Nat. Geosci. (2023)
10.1038/s41561-023-01160-z

1234 23 JUNE 2023 • VOL 380 ISSUE 6651 science.org SCIENCE

RES EARCH
◥ iment and theory, this has already led to an
REVIEW SUMMARY
TOPOLOGICAL MATTER
Hunting for Majoranas
Ali Yazdani*, Felix von Oppen, Bertrand I. Halperin, Amir Yacoby
BACKGROUND: The past decade has witnessed
considerable progress toward the creation of
new quantum technologies. Substantial advances
in present leading qubit technologies, which
are based on superconductors, semiconductors,
trapped ions, or neutral atoms, will undoubt-
edly be made in the years ahead. Beyond these
present technologies, there exist blueprints for
topological qubits, which leverage fundamen-
tally different physics for improved qubit
performance. These qubits exploit the fact that
quasiparticles of topological quantum states
allow quantum information to be encoded and
processed in a nonlocal manner, providing
inherent protection against decoherence and
potentially overcoming a major challenge of
the present generation of qubits. Although still
far from being experimentally realized, the
potential benefits of this approach are evident.
The inherent protection against decoherence
implies better scalability, promising a con-
siderable reduction in the number of qubits
needed for error correction. Transcending pos-
sible technological applications, the underly-
ing physics is rife with exciting concepts and
challenges, including topological supercon-
ductors, non-abelian anyons such as Majorana
zero modes (MZMs), and non-abelian quan-
tum statistics.
ADVANCES: In a wide-ranging and ongoing
effort, numerous potential material platforms
Majorana and e/4 quasiparticles
Proposed topological platforms. (Left) Proposed state of electrons in a high magnetic field (even-denominator fractional quantum Hall states) are predicted to
host Majorana quasiparticles. (Right) Hybrid structures of superconductors and other materials have also been proposed to host such quasiparticles and can
be tailored to create topological quantum bits based on Majoranas.
Yazdani et al., Science 380, 1235 (2023) 23 June 2023
are being explored that may realize the re-
quired topological quantum states. Non-abelian
anyons were first predicted as quasiparticles of
topological states known as fractional quan-
tum Hall states, which are formed when elec-
trons move in a plane subject to a strong
perpendicular magnetic field. The prediction
that hybrid materials that combine topolog-
ical insulators and conventional supercon-
ductors can support localized MZMs, the
simplest type of non-abelian anyon, brought
entirely new material platforms into view.
These include, among others, semiconductor-
superconductor hybrids, magnetic adatoms
on superconducting substrates, and Fe-based
superconductors. One-dimensional systems
are playing a particularly prominent role,
with blueprints for quantum information ap-
plications being most developed for hybrid
semiconductor-superconductor systems. There
have been numerous attempts to observe non-
abelian anyons in the laboratory. Several ex-
perimental efforts observed signatures that
are consistent with some of the theoretical
predictions for MZMs. A few extensively studied
platforms were subjected to intense scrutiny
and in-depth analyses of alternative inter-
pretations, revealing a more complex reality
than anticipated, with multiple possible in-
terpretations of the data. Because advances
in our understanding of a physical system
often rely on discrepancies between exper-
Majorana zero modes
Superconductor
Semiconductor
Gate
improved understanding of Majorana sig-
natures; however, our ability to detect and
manipulate non-abelian anyons such as MZMs
remains in its infancy. Future work can build
on improved materials in some of the existing
platforms but may also exploit new mate-
rials such as van der Waals heterostructures,
including twisted layers, which promise many
new options for engineering topological phases
of matter.
OUTLOOK: Experimentally establishing the
existence of non-abelian anyons constitutes
an outstandingly worthwhile goal, not only
from the point of view of fundamental phys-
ics but also because of their potential appli-
cations. Future progress will be accelerated
if claims of Majorana discoveries are based
on experimental tests that go substantially
beyond indicators such as zero-bias peaks that,
p
at best, suggest consistency with a Majorana
interpretation. It will be equally important
that these discoveries build on an excellent
understanding of the underlying material
systems. Most likely, further material im-
provements of existing platforms and the ex-
g
ploration of new material platforms will both
be important avenues for progress toward
obtaining solid evidence for MZMs. Once that
has been achieved, we can hope to explore—
and harness—the fascinatin!g physics of non-
y
abelian anyons such as the topologically pro-
tected ground state manifold and non-abelian
statistics.
▪
The list of author affiliations is available in the full article online.
*Corresponding author. Email: yazdani@princeton.edu
Cite this article as A. Yazdani et al., Science 380, eade0850
(2023). DOI: 10.1126/science.ade0850
READ THE FULL ARTICLE AT
https://doi.org/10.1126/science.ade0850
y g
,
1 of 1
RES EARCH

◥ engineered in hybrid structures that involve

REVIEW conventional superconductors, an idea that
has stimulated a plethora of further proposals.
TOPOLOGICAL MATTER As we outline here, several experimental ef-

Hunting for Majoranas

forts observed signatures that at first appeared
consistent with theoretical predictions for
MZMs. Unfortunately, many of the experi-
Ali Yazdani1*, Felix von Oppen2, Bertrand I. Halperin3, Amir Yacoby3 ments revealed a more complex reality than
anticipated, allowing for multiple interpre-
Over the past decade, there have been considerable efforts to observe non-abelian quasiparticles in tations of the data. Although this can be seen
novel quantum materials and devices. These efforts are motivated by the goals of demonstrating as a severe drawback or limitation of the ex-
quantum statistics of quasiparticles beyond those of fermions and bosons and of establishing the perimental status, we believe that this is
underlying science for the creation of topologically protected quantum bits. In this Review, we focus actually a desired and welcome outcome. Ad-
on efforts to create topological superconducting phases that host Majorana zero modes. We consider vances in our understanding of a physical
the lessons learned from existing experimental efforts, which are motivating both improvements to system largely rely on discrepancies between
present platforms and the exploration of new approaches. Although the experimental detection of non- experiment and theory. Such discrepancies
abelian quasiparticles remains challenging, the knowledge gained thus far and the opportunities ahead suggest that the theoretical description is
offer high potential for discovery and advances in this exciting area of quantum physics. too simplistic, calling for modifications or
extensions. Moreover, multiple possible in-

T
terpretations stimulate new and sharpened
he past decade has witnessed consider- tion. Despite numerous attempts, experimen- experiments, which can distinguish between

able progress toward the creation of new
quantum technologies. Recent efforts (1)
to demonstrate computational quantum
advantage (2) using processors based on
superconducting quantum bits (qubits) and
efforts to simulate complex quantum phases
tal evidence for non-abelian anyons is still
limited, and their non-abelian statistics re-
main largely a theoretical prediction. Addi-
tionally, the material synthesis, experimental
advances, and theoretical modeling that are
required to realize, explore, and understand
p
competing theoretical interpretations. Over-
all, although our ability to detect and ma-
nipulate non-abelian anyons such as MZMs
remains in its infancy, the landscape of mate-
rials, tools, and ideas in this area is notably
rich. Here, we provide a perspective on the

of matter with such technologies exemplify the
notable advances of a rapidly evolving field.
Substantial advances in present qubit technol-
ogies, which are based on conventional super-
conductors (3), semiconductors (4), trapped ions
the underlying topological quantum states
provide an extraordinarily rich setting for the
discovery of new quantum phenomena.
Quasiparticles with properties distinct from
those of free electrons are a hallmark of con-
g
progress made over the past decade, briefly
reviewing the key concepts and the lessons
learned from the experimental efforts thus far
and also looking ahead to potential areas for
future progress and discovery in this field.

(5), or Rydberg atoms (6), will undoubtedly be
made in the years ahead.
Beyond these present technologies, there
exist blueprints for topological qubits that
leverage conceptually different physics for im-
proved qubit performance (7–12). These qubits
exploit the fact that quasiparticles of topolog-
ical quantum states allow quantum informa-
tion to be encoded and processed in a nonlocal
manner, providing inherent protection against
densed matter systems. A notable manifes-
tation is the fractional quantum Hall (FQH)
effect. Strong interactions among electrons
moving in two dimensions subject to a mag-
netic field create topological quantum states
that host anyonic quasiparticles (14, 15). For
many FQH states, the anyons are abelian, but
some states are expected to support non-abelian
anyons (16). Read and Green (17) proposed that
the most prominent of these states, correspond-
y
Fundamentals
Non-abelian statistics are a property of quasi-
particles or defects in special 2D electron sys-
tems such as FQH systems (16, 22–24) or
topological superconductors (17, 25) (Fig. 1). In
the absence of defects, these systems have an
energy gap for excitations far from any bound-
aries. But when there are defects or quasi-
particles at fixed locations, the ground state

decoherence (7, 13). Although still far from
being experimentally realized, the potential
benefits of this approach are evident. The in-
herent protection against decoherence implies
better scalability while also promising a sub-
ing to the Landau level filling factor n = 5/2
(16, 18), can be understood as a superconduc-
tor of composite fermions with chiral p-wave
symmetry. Quasiparticles of this n = 5/2 state
realize the simplest type of non-abelian anyons
y g
becomes a manifold of nearly degenerate low-
energy eigenstates, whose dimension grows
exponentially with the number of defects. The
energy differences between these states are
predicted to fall off exponentially with the sepa-

stantial reduction in the number of qubits
needed for error correction.
Apart from possible applications, funda-
mental physics furnishes ample motivation for
pursuing this line of research. Topological
qubits rely on quantum states that are pre-
dicted to host a new class of quasiparticles,
termed non-abelian anyons; their quantum
properties offer fundamentally new ways of
encoding and processing quantum informa-
1 dimensional (2D) p-wave superconductors.
Joseph Henry Laboratories and Department of Physics,
Princeton University, Princeton, NJ 08540, USA. 2Dahlem
Center for Complex Quantum Systems and Fachbereich
Physik, Freie Universität Berlin, 14195 Berlin, Germany.
3
Department of Physics, Harvard University, Cambridge, MA
02138, USA.
*Corresponding author. Email: yazdani@princeton.edu
known as Ising anyons. The presence of Ising
anyons implies a topological degeneracy of the
quantum state. The degeneracy can be asso-
ciated with Majorana zero modes (MZMs) in the
corresponding p-wave superconductor and is
the foundation for building a topological qubit.
A deeper understanding of topological quan-
tum states has led to the realization that a far
wider range of materials can host non-abelian
anyons. Read and Green (17) already under-
stood that MZMs exist in vortex cores of two-
Kitaev (19) recognized that MZMs occur in
1D p-wave superconductors of spinless fer-
mions. Whereas intrinsic p-wave supercon-
ductors are, at best, rare in nature, Fu and Kane
(20, 21) showed that they can be effectively
,
ration between the quasiparticles, so, in princi-
ple, they can be exceedingly small. In the case
of the fractional quantized Hall state at filling
factor n = 5/2, the proposed non-abelian quasi-
particles are objects with electric charge ±e/4
(16), whereas in the model of a px + ipy super-
conductor, they are associated with vortex cores
(in two dimensions) (17) or domain walls (in
one dimension) (19). (Here, e denotes the ele-
mentary charge.)
If well-separated quasiparticles are slowly
moved around each other or interchanged, in
such a way that the sets of initial and final
positions for each quasiparticle type are iden-
tical, the final state of the system will be related
to the initial state by a unitary transformation
in the Hilbert space of low-energy eigenstates.

Yazdani et al., Science 380, eade0850 (2023) 23 June 2023 1 of 12

RES EARCH | R E V I E W

If braiding can be performed on the appropriate

timescale, the unitary transformations could
be used to perform topologically protected
quantum computations (7).
Ising anyons, which are the simplest non-
abelian quasiparticles, have a ground-state
degeneracy of 2N/2, where N is the number
of quasiparticles. This can be described by
associating with each localized quasiparticle a
MZM, characterized by a Majorana operator gj
(17). Majorana operators for different quasi-
particles anticommute, like ordinary fermion
operators, but a single Majorana operator obeys
the relations gj = gj† and gj2 = 1 (i.e., it is its own
antiparticle). If state jYi belongs to the mani-
fold of degenerate ground states, then gj jYi
will also belong to this manifold but will have
opposite fermion number parity. If quasipar-
ticles can be moved around, their MZMs will
move with them, leading to the unitary trans-
formations within the degenerate ground-state

p
manifold underlying non-abelian statistics.
Equivalent unitary transformations can be
produced without physical motion, if it is
possible to effectively turn on and off the in-
teractions between nearby MZMs (26–29).
Alternative schemes induce or replace braiding

g
operations by measurement protocols (30–33).

Model systems
FQH systems
The first systems that were predicted (16) to host

y
non-abelian anyons are the even-denominator
FQH states observed in GaAs heterostructures
at Landau-level filling fractions n = 5/2 and
7/2 (18). There are, as of now, three compet-
ing theoretical models for these states (see
section on experimental searches and lesson
learned below). In all three cases, it is pre-
dicted that the elementary quasiparticles should
be Ising anyons with charge ±e/4. Each of
these models can be mapped mathematically

y g
onto a topological superconductor interacting
with an emergent gauge field, with the e/4
quasiparticles appearing as vortices (Fig. 2A).
Fig. 1. Fundamentals. (A) The ground state of a topological phase with localized anyons is highly
degenerate, providing a protected subspace for storing quantum information. The ground states are Kitaev’s 1D model
separated from excited states by an energy gap. (B) The quantum exchange of quasiparticles in two

,
Whereas in two dimensions, MZMs are asso-
dimensions is fundamentally different from that in three dimensions. As the figure shows, the path of one
ciated with vortex cores, Kitaev (19) showed
particle looping the other can keep shrinking, whereas in two dimensions, the loop runs into the other particle
that MZMs also emerge at the ends of 1D to-
as it shrinks. (C) This difference manifests itself in topological quantum phases, which host a class of
pological superconductors. He discusses a sim-
quasiparticles termed non-abelian anyons. The low-energy quantum states of non-abelian anyons depend
ple and elegant model of spinless fermions
on the order in which exchange (or braiding) processes are performed. The state of the system after
hopping on a linear chain in the presence of
braiding is related to the initial state by a unitary transformation, which depends only on the topology of
nearest-neighbor p-wave superconducting pair-
the braiding operation and can be exploited to process quantum information. The illustration shows two
ing D (19) (see Box 1 and Fig. 2B) Although the
braiding sequences of anyons A, B, C, and D (such as Ising anyons or MZMs), which implement different
braiding of quasiparticles is impossible in a
unitary transformations and generally result in different quantum states.
strictly 1D setting, the chains can be connected
into a 2D network, which allows for braiding
Furthermore, if this process is fast compared world lines. It will be independent of other and non-abelian statistics (34).
with the exponentially small energy splittings details and will be unaffected by any local per- At first sight, there are severe obstacles to a
of the Hilbert space, but slow compared with turbations. Because the resulting unitary trans- physical realization of the Kitaev chain. Elec-
the energy gap for other excitations, the uni- formation will generally depend on the order in trons have spin, conventional superconductors
tary transformation will depend only on the which the interchanges have been performed, form s-wave rather than p-wave Cooper pairs, and
topology of the braiding of the quasiparticle the system is said to obey non-abelian statistics. thermal fluctuations suppress superconducting

Yazdani et al., Science 380, eade0850 (2023) 23 June 2023 2 of 12

RES EARCH | R E V I E W

Fig. 2. Model systems. There is a

diverse set of possible platforms
for realizing non-abelian anyons
such as MZMs. (A) In some FQH
states, such as n = 5/2, strong
electronic correlations create
topological phases that are
predicted to support fractionally
charged quasiparticles (QPs)
with associated MZMs. B, applied
magnetic field. (B) Several
platforms such as semiconductor
quantum wires and chains
of magnetic adatoms on super-
conducting substrates are
closely related to the much-
studied Kitaev chain (see
also Box 1). Hopping and p-wave
pairing of spinless electrons in
one dimension can create the
conditions for localizing isolated

p
MZMs at the ends of the chain.
The blue and red spheres repre-
sent the p-wave pairing amplitude
(D) that is on the nearest-
neighbor site on a chain.
(C) Effective p-wave super-

g
conductivity can also be realized by proximity coupling the helical edge mode (red and blue arrows at the edges) of a 2D TI to a magnetic insulator and a
superconductor, both of which will gap the edge mode. A MZM is localized at the boundary between these gapped regions. (D) Combining a strong spin-orbit material
and superconducting electrodes into a Josephson junction can also emulate the Kitaev chain. The junction hosts MZMs at the two ends, when tuned to the topological
phase by adjusting the phase difference across the junction and applying a magnetic field.

y
states by possibly very small energies, making
Box 1. Kitaev chain. detection of MZM in vortex cores challenging.

Kitaev’s model of a spinless p-wave superconductor exemplifies the emergence of MZMs at the
boundaries of a topological superconductor (19). The spinless fermions (creation operator cj† at site j)
hop along a chain with amplitude t subject to nearest-neighbor p-wave pairing D
Xn o X
H¼ j
tc†jþ1 cj þ Dc†jþ1 c†j þ h:c:  m j c†j cj
where H is the Hamiltonian, m is chemical potential, and h.c. is the hermitian conjugate.
Quantum wires and atomic chains
MZMs and 1D topological superconductivity
can also be engineered in more conventional
materials by using a by-now-familiar recipe
(35–38). One induces superconductivity into
a (quasi) 1D system by proximity from a bulk

y g
The Kitaev chain enters a topological superconducting phase when the chemical potential m falls superconductor with conventional s-wave pair-
within the normal-state band Dk = −2tcosk (with k denoting the fermion wave vector). Reflecting ing and suppresses the doubling of the Fermi
the bulk-boundary correspondence, this phase supports gapless boundary excitations at the ends surface by applying a magnetic field or mag-
of the chain, which are isolated MZMs. Apart from exponentially small corrections, a finite chain with netism without quenching the bulk supercon-
a pair of MZMs g1 and g2 has two degenerate ground states that differ in fermion parity and can ductor. In the presence of spin-orbit coupling,
be characterized by the occupation d†d of the conventional fermion mode d ¼ 21 ðgi  ig2 Þ.

order in a 1D setting. Several approaches have
been proposed to overcome these challenges.
Topological boundary modes and vortex cores
With the advent of topological band theory, it
was predicted that boundary modes of 2D or
3D topological insulators (TIs) can be used for
creating topological superconductors (20, 21).
The edge mode of a 2D TI or the 2D surface
state of a 3D TI exhibits spin-momentum lock-
ing, a property that can be viewed as creating
effectively spinless quasiparticles, as in the
Kitaev model. The edge mode of a 2D TI can be
gapped out by proximity coupling to a super-
conductor with s-wave pairing or to a ferro-
magnetic insulator. Isolated MZMs appear
at domain walls between these two types of
gapped regions (Fig. 2C). A 2D analog can be
realized by inducing superconductivity in the
surface state of a 3D TI by proximity; MZMs
are then expected to be bound to individual
vortices induced by a perpendicular magnetic
field. Individual vortices also host trivial sub-
gap Caroli–de Gennes–Matricon states, the
spectrum of which would be shifted in energy
when a MZM appears at the core of the vor-
tices. However, the MZM may be separated
from finite-energy Caroli–de Gennes–Matricon
,
either in the proximity-providing supercon-
ductor or in the 1D system, the spin-singlet
Cooper pairs of the bulk superconductor can
still proximitize the 1D system, where they
effectively induce p-wave rather than s-wave
pairing. Because pairing in the 1D system is
inherited from a bulk superconductor, the
superconducting correlations are not destroyed
by thermal fluctuations. Following this hybrid
approach, experimental searches for MZMs have
thus far focused on two classes of systems—
semiconductor quantum wires and chains
of magnetic adatoms on superconducting
substrates.
Semiconductor quantum wires made of InAs
or InSb are known to have strong spin-orbit
coupling and are readily proximitized by a

Yazdani et al., Science 380, eade0850 (2023) 23 June 2023 3 of 12

RES EARCH | R E V I E W

superconductor (Fig. 3B). Material science ad-
vances motivated by the search for MZMs
have led to epitaxial growth of superconduc-
tors (Al) on semiconductor quantum wires with
exquisite interface quality and excellent prox-
imity coupling (39). A variant to create 1D
channels is to use gating of a 2D electron gas
in InAs proximitized by an epitaxially grown
superconductor (40). A useful feature of epi-
taxial semiconductor-superconductor materials
is that one can realize structures with sub-
stantial charging energies. The associated
Coulomb blockade physics admits additional
experimental tests of MZMs (41, 42) and is
predicted to lead to Majorana teleportation
(43) as well as a topological Kondo effect (44).
Coulomb blockade also suppresses quasipar-
ticle poisoning by stabilizing specific charge
states (45). This is an important ingredient in
some of the most promising designs of a topo-
logical qubit (see Fig. 4) (11, 31, 46).
Chains of magnetic adatoms on conven-
tional superconducting substrates can also ef-
fectively emulate the Kitaev model (47). This
platform lends itself to in situ characterization
by high-resolution scanning tunneling micros-
copy (STM) and spectroscopy (STS). Densely
packed chains of adatoms form spin-polarized
1D bands by hybridizing their d-orbitals, for
instance, in ferromagnetically ordered chains
(Fig. 3B). When an odd number of these bands
crosses the Fermi energy of the substrate
superconductor and spin-orbit coupling en-
ables proximitization of the spin-polarized
bands, a p-wave gap forms, effectively result-
ing in a 1D topological superconductor with
MZMs localized at the ends of the chain
(48, 49). Alternatively, the chain can order
into a spin spiral, which might self-tune the
system into the topological phase (50–52).
A characteristic feature of this platform is
the strong hybridization of the adatoms with

A E

p
B D

g
y
C

Fig. 3. Experimental platforms for MZM searches. (A) Interferometry of FQH
edge states can probe the phase acquired in exchange processes of FQH
quasiparticles. Such experiments have confirmed that the quasiparticles of
the n = 1/3 state are abelian anyons (60). Similar experiments on candidate non-
abelian states, such as the n = 5/2 state, may provide evidence for Majoranas
associated with non-abelian quasiparticles. [Adapted by permission from
Springer Nature Customer Service Center GmbH, Springer Nature (60), copyright
(2020)] (B) Experiments on semiconductors with strong spin-orbit coupling
proximitized by a superconductor have been the focus of numerous works in
search of MZMs. The zero-bias peak routinely observed in these experiments
(67) appears for field orientations along the wire and has been interpreted
as evidence for MZMs. However, other experiments show that such features may
originate from Andreev bound states. [Adapted with permission from (67)]
(C) Chains of magnetic atoms on the surface of a superconductor provide
another platform for realizing MZMs. Experiments on a ferromagnetic Fe chain
on a Pb surface show spatially localized zero-energy states (101) with characteristic
signatures of MZMs in spin-resolved measurements (102, 103). The large
y g
,
number of low-energy in-gap states in such chains has made the identification
of the expected topological gap difficult. [Left image adapted by permission
from Springer Nature Customer Service Center GmbH, Springer Nature
(101), copyright (2017); right image adapted with permission from (102)]
(D) Topological surface states in Fe-based superconductors are another possible
platform for MZMs. Spectroscopy of vortices on such surfaces reveal zero-bias
peaks with some reported behavior that is distinct from that of trivial vortex
bound states (114). The interplay between states on the surface and in the
bulk of such vortices makes stabilizing MZMs sensitive to bulk properties.
[Adapted with permission from (114)] (E) Phase-biased Josephson junctions
that use strongly spin-orbit–coupled semiconductors in the presence of an
in-plane magnetic field are predicted to exhibit a topological phase with
MZMs; experiments on HgTe junctions show zero-bias peaks in tunneling
spectroscopy (123). The difficulty in distinguishing zero-bias peaks due to MZMs
from other nontopological in-gap states across the various platforms calls for
more sophisticated experiments. [Adapted by permission from Springer Nature
Customer Service Center GmbH, Springer Nature (123), copyright (2019)]

Yazdani et al., Science 380, eade0850 (2023) 23 June 2023 4 of 12

RES EARCH | R E V I E W
Fig. 4. Majorana-based topological qubit. Ideas
for realizing topological qubits are most developed
for Majorana wires (31, 46). The two degenerate
ground states of a single Majorana wire have
different fermion parities and cannot be brought
into a coherent superposition because of fermion-
parity superselection. Coherent qubit dynamics
requires at least two quantum wires (green) with
four MZMs gj (j = 1, …, 4). Their ground-state
manifold is spanned by four states—two with
even and two with odd fermion parity. Both parity
sectors can act as the two-level system of the
topological qubit. After connecting the two wires
by a conventional superconductor (S, orange), the
total electron number and hence the fermion
parity can be controlled by a gate voltage Vg.
Coulomb blockade can therefore help protect
the qubit against leakage errors into the non-
computational subspace. The Pauli operators of
the qubit are bilinears in the Majoranas, for example,
X = ig2g3 and Z = ig1g2, which can be read out
by appropriately coupling to the respective pair of
MZMs (31, 46, 184). This setting is known as a
Majorana box qubit or tetron. A minimal test of a
topological qubit would be the implementation of
sequential Stern-Gerlach–type experiments.
the superconducting substrate, which has
been shown to result in strongly localized
MZMs (53).
Topological superconductivity has also been
predicted in dilute chains, in which coupling
between adatoms is entirely mediated by means
of the superconducting substrate (47, 54, 55).
In this limit, each adatom induces one (or
more) pairs of spin-polarized Yu-Shiba-Rusinov
(YSR) states symmetric in energy about the
center of the superconducting gap of the sub-
strate. If the adatoms order magnetically, the
subgap states hybridize along the chain and
form YSR bands. Topological superconductiv-
ity ensues when the bands that emerge from
the positive- and negative-energy YSR states
overlap and develop a gap because of induced
p-wave superconductivity. For ferromagnetic
order, such a gap requires the presence of spin-
orbit coupling.
Josephson junctions
Josephson junctions that couple two conven-
tional superconductors via a semiconductor
with strong spin-orbit interactions can also
Yazdani et al., Science 380, eade0850 (2023) 23 June 2023
realize a 1D topological superconducting phase.
In this setting, the subgap states propagating
along the junction are tuned into a topological
superconducting phase by controlling the
phase bias across the junction and applying
an in-plane magnetic field (Figs. 2D and 3D)
(56, 57). The phase bias greatly enlarges the
topological region of the phase diagram, where
MZMs are confined at the ends of the junc-
tion. Crucially, this makes the topological
phase quite robust to changes in geometry
and chemical potential, although the super-
current induced by the phase bias tends to
diminish the gap.
Apart from these model platforms, numer-
ous other ideas, based on a wide variety of
materials and hybrid structures, have been
proposed for creating topological supercon-
ductivity. Ideas for realizing topological qubits
and Majorana-based quantum computations
are most developed for the nanowire plat-
form (see Fig. 4). After briefly describing the
present experimental status of detecting MZMs
in the model platforms, we sketch the po-
tential of other proposed approaches for ad-
vancing the field. Ultimately, besides creating
topological qubits, the field aims to provide
a platform for realizing and discovering new
topological phases and phenomena.
Experimental searches and lessons learned
FQH states
The search for non-abelian quasiparticles in
FQH systems has focused on detecting the
quantum mechanical phase associated with
quasiparticle exchanges in electronic interfer-
ometers (58, 59). Recently, such experiments
on a GaAs-based 2D electron gas measured
the phase shift associated with the exchange of
the abelian quasiparticles in the n = 1/3 FQH
state (60). Similar experiments on the n = 5/2
state of a GaAs device found results consistent
with expectations for Ising anyons (61) (Fig. 3A).
However, so far, the evidence remains statisti-
cal, and various questions remain about the
microscopic geometry of the interference re-
gion. Several experiments confirmed that the
quasiparticles have charge e/4 (62, 63), but
such charges could also occur, in principle, in a
system with abelian statistics.
Three topologically distinct states, all con-
taining Ising anyons, are competing candidates
to explain the Hall plateau at n = 5/2. Numer-
ical studies of finite systems have strongly
favored a state originally proposed by Moore
and Read (16), known as the Pfaffian state,
or, perhaps preferably, its particle-hole (PH)
conjugate, the anti-Pfaffian state. However, mea-
surements of the quantized thermal conduc-
tance of the edge modes at n = 5/2 (64) have
strongly favored a third state, known as the
PH-Pfaffian, which has not appeared as a plau-
sible candidate in numerical studies. Further
evidence for the PH-Pfaffian has come from
measurements of thermal transport in geome-
tries where the 5/2 state is bordered by states
with n = 2 and n = 3 (65). The resulting dis-
crepancy between theory and experiment re-
mains to be understood [see (66) and references
therein].
1D systems
In the search for experimental signatures of
MZMs, much attention has focused on 1D
systems. The putative MZMs are commonly
probed by tunneling from a normal-metal elec-
trode (67). At biases within the supercon-
ducting gap, tunneling is dominated by the
hybridization of the electrode states with
the MZM at zero energy, predicting a sub-
gap tunneling resonance at zero voltage. Zero-
bias peaks in the differential conductance that
are consistent with this expectation are
routinely observed in studies of proximity-
coupled semiconductor nanowires with strong
p
spin-orbit coupling and subject to a parallel
field (Fig. 3B). Numerous studies have exam-
ined whether this finding is consistent with a
MZM interpretation (68). Possible alterna-
tives with similar phenomenology assume an
Andreev bound state fine-tuned to zero energy
g
by the suppression of pairing, the formation
of multiple quasi-Majoranas, or the forma-
tion of a quantum dot near the end of the wire
(69–74). In a related experimental platform,
the semiconductor nanowire is fully encap-
y
sulated by a superconducting shell (at the ex-
pense of a lack of gate control). In the presence
of magnetic flux threading the superconduct-
ing shell, zero-bias peaks appear when tun-
neling into the end of such a wire (75) [see also
(76)]. However, comparable phenomenology
in other experiments on similar devices that
explored a larger number of experimental con-
figurations has been interpreted in terms of
Andreev bound states. These can be tuned to
y g
zero energy by the magnetic flux (77), provid-
ing an alternative explanation for the zero-
bias peaks.
Experiments on semiconductor nanowires
in the Coulomb-blockade regime also provide
an interesting probe of MZMs (41, 78). Such
,
experiments show a transition from 2e charg-
ing peaks at weak magnetic fields to 1e charg-
ing peaks at higher magnetic fields, consistent
with the emergence of MZMs at the topolog-
ical phase transition. In addition, the spacing
of the Coulomb blockade peaks can be used to
extract the residual splitting of the MZMs as a
function of the length of the wire. An expo-
nential length dependence of this splitting,
reported in (41), was interpreted as consistent
with the behavior of MZMs. A more recent
study found a more complex picture (78). For
instance, the transition from 2e to 1e periodic
Coulomb-blockade peaks was not accompa-
nied by the emergence of zero-bias peaks in
tunneling into the ends of the nanowire.
5 of 12
RES EARCH | R E V I E W
Under optimal conditions, a Majorana zero-
bias peak is predicted to have a quantized
height of 2e2/h (where h is Planck’s constant)
(79–82). Experimental tests of the quantized
conductance have been subject to consider-
able recent discussion. One report of its obser-
vation (83) was subsequently retracted (84, 85);
a second study exhibited scaling of the peak
conductance that was interpreted as consis-
tent with quantization (86). A pertinent compli-
cation is that experiments consistently observe
a substantial softening of the superconduct-
ing gap once the magnetic field is sufficiently
strong to induce the putative topological super-
conducting phase. Furthermore, theoretical
work suggests that peak quantization is not
a definite indicator of Majorana end states but
can also occur outside of the topological super-
conducting phase (74, 87).
Present efforts on the nanowire platform
focus on cross-correlating the existence of zero-
bias peaks with other signatures such as the
gap closing and reopening at the topological
phase transition (88, 89). A recent experiment
implemented a previously defined gap proto-
col (88) that was designed to identify param-
eter regions, in which the nanowires exhibit a
topological superconducting gap with high like-
lihood. The experiment on gate-defined 1D
channels in hybrid InAs-Al devices provided
evidence of the expected cross-correlations for
a few devices and narrow ranges of chemical
potential (90).
An important issue for the nanowire plat-
form, but also more generally, is the role of
disorder. Unlike the case of s-wave supercon-
ductivity, which is largely immune to non-
magnetic disorder, in odd-parity superconductors,
even weak disorder can have detrimental ef-
fects. In semiconducting wires proximitized by
an s-wave superconductor, potential disorder
introduces subgap states that may considera-
bly soften the gap, lead to partitioning of the
wire into segments that each have localized
MZMs, and drive the system out of the topo-
logical superconducting phase beyond a crit-
ical strength of disorder (90, 91). Disorder
also induces large fluctuations of the residual
Majorana splitting (92) and can result in ex-
perimental signatures such as zero-bias peaks
in conductance that are not indicative of MZMs
(93). Over the years, considerable effort has
gone into simulating the effects of disorder
in more realistic models of the nanowire plat-
form [see, for example, (89, 94–96)]. It is also
interesting to comment that although disorder
in the nanowire is potentially detrimental,
disorder in the superconducting shell can be
crucial to establishing effective proximity
coupling (97, 98).
In chains of magnetic atoms, spectroscopy
with a STM has detected signatures of highly
localized zero-energy modes at the ends of
ferromagnetic chains (Fe) assembled on the
Yazdani et al., Science 380, eade0850 (2023) 23 June 2023
surface of a strongly spin-orbit–coupled super-
conductor (Pb) (48, 99, 100). Just as for the
nanowire platform, zero-bias peaks by them-
selves do not distinguish between MZMs and
conventional Andreev bound states or quasi-
Majoranas. However, unlike conductance mea-
surements on device-like structures, STM
experiments can spatially resolve MZMs
and other low-energy states, allowing one
to distinguish between end and bulk states.
The zero-bias peaks detected in these experi-
ments at the ends of the chain are robust to
being buried with an additional superconduct-
ing layer (101) and display spin-polarization
signatures that are consistent with a MZM
(Fig. 3C) (102, 103). The spin-polarized mea-
surements on the Fe chains also provide evi-
dence against impurity-induced end states.
However, they cannot eliminate the possibil-
ity of multiple MZMs originating from dif-
ferent channels if their splitting is below the
experimental resolution (presently at 80 meV).
STM spectroscopy of chains that exhibit sig-
natures of zero-energy end states also show
residual in-gap states along the chain, obs-
tructing the identification of a topological
bulk gap (101). Theoretical studies have ad-
dressed mechanisms for the appearance of
trivial zero-energy end states (104, 105).
Several experiments have explored possible
signatures of MZMs in magnetic atom chains
with various combinations of magnetic adatoms
and substrates, including chains assembled
atom by atom (106–111). These experiments
highlight that the emergence of zero-energy
modes localized near the chain ends is not a
generic feature of atomic chains. Some experi-
ments do not observe zero-energy features at
all, whereas they are not localized at the ends of
the chains in others. This is consistent with the
theoretical expectation that the emergence of
topological superconductivity is contingent on
conditions such as the number of 1D bands
(related to the occupation of the adatom d-
shell), the strength of spin-orbit coupling,
or the magnetic ordering. In dilute chains,
topological superconductivity may also be
suppressed by the quantum nature of the
adatom spins (55, 110).
Topological boundary modes and vortex cores
Efforts to realize MZMs in topological bound-
ary modes initially focused on heterostructures
of TIs and conventional superconductors. STM
spectroscopy was used to probe properties of
vortex cores in the resulting superconducting
2D surface states. We note that in these set-
tings, the presence or absence of MZMs depends
sensitively on doping because the vortex core
must realize a 1D topological superconducting
state to support MZMs at its ends (112, 113).
More recent efforts investigate Fe-based super-
conductors, which combine a 2D topological
surface state with intrinsic bulk supercon-
ductivity. STM experiments on the surface of
FeTe0.55Se0.45 resulted in the observation of
a sharp zero-bias peak inside the vortex core
(114, 115). Although some of the reported prop-
erties in these measurements are consistent
with MZMs, distinguishing them from trivial
Caroli–de Gennes–Matricon states remains chal-
lenging (116). For example, only a fraction of
vortices were actually found to host a zero-
energy state, with others hosting Caroli–
de Gennes–Matricon states at finite energy.
The role of surface disorder in these experi-
ments has also added challenges. Experiments
have reported that disorder can induce trivial
surface states in FeTe0.55Se0.45 and that vortex-
core zero-energy states are even entirely ab-
sent (115, 117). Reports of plateau-like behavior
in STS measurements of zero-energy vortex-
core states as a function of tip-sample distance
have been interpreted as a possible observa-
tion of the quantized conductance for tunneling
p
into a MZM (118). However, similar experiments
show larger conductance than the predicted
quantized value (119), and the plateau-like be-
havior may be caused by suppressed quasipar-
ticle relaxation at higher tunneling current
into a trivial vortex-core state, rather than by
g
quantization (116).
Interesting high-resolution STM experiments
of vortex cores in FeSe0.5Te0.5 show a zero-bias
resonance that is separated from finite-energy
resonances and occurs only at low fields, pre-
y
sumably owing to core-core overlap of the zero
modes at higher magnetic fields (Fig. 3D) (120).
These most recent experiments highlight the
key advantage of Fe-based superconductors, in
which the short coherence length can potentially
allow for distinguishing Caroli–de Gennes–
Matricon states from MZMs. Theoretical work
has discussed the possible emergence of trivial
zero-energy states due to impurities (121).
Experiments have also explored inducing
y g
superconductivity and magnetism in the topo-
logical edge modes of Bi bilayers (122). The Bi
bilayers are grown on a superconductor, and
their edges are covered by magnetic clusters.
By probing the edge mode by local STM spec-
,
troscopy, experiments detect zero-bias peaks
between regions influenced by superconductiv-
ity and magnetism, respectively. Similar to
the experiments on atomic chains (102), spin-
polarized STM was used to show that the zero-
energy states have spin properties distinct
from those induced by magnetic clusters and
consistent with those expected for a MZM.
Josephson junctions
Signatures of topological superconductivity
have also been observed in Josephson junc-
tions with the strongly spin-orbit–coupled
semiconductors HgTe and InAs (Fig. 3E)
(123, 124). The application of a phase bias
across the junction enlarges the topological
region. At the same time, the gap is reduced by
6 of 12
RES EARCH | R E V I E W
the in-plane magnetic field of ~1 T (weakening
superconductivity in the electrodes) and by
the supercurrent associated with the phase
bias. Moreover, the gap can become soft owing
to electron and hole trajectories, which prop-
agate nearly in parallel to the superconducting
electrodes and are thus weakly affected by
the superconductor. In experiment, this reduc-
tion results in a large localization length (on
the scale of micrometers) of the MZMs, ob-
scuring their experimental signatures and
complicating implementations of braiding.
Other efforts
Metallic surface states can have particularly
strong Rashba spin-orbit coupling and thus
provide a possible alternative to spin-orbit–
coupled semiconductors for realizing MZMs
(125). This has stimulated the proposal to en-
gineer a (quasi-)1D topological superconductor
using Au nanowires on a superconducting sub-
strate, despite their much smaller Fermi wave-
length. Recent STM experiments on Au wires
proximity-coupled to the ferromagnetic insu-
lator EuS show preliminary signatures con-
sistent with MZMs (126). However, structural
variations of the samples have thus far ham-
pered consistent observations of MZM signa-
tures, and high-resolution experiments would
be desirable.
Future directions
The search for non-abelian anyons continues
to provide ample opportunities for experi-
ments across a wide range of material plat-
forms and for the exploration of a broad
spectrum of quantum phenomena. The im-
mediate challenge in the field remains to
firmly establish the experimental existence of
MZMs and other non-abelian anyons. It is
now clear that, by themselves, zero-bias peaks
do not reliably indicate MZMs. Even the recent
minimal approach of cross-correlating several
features of the topological phases (88, 89, 96)
remains largely insensitive to quasiparticle poi-
soning and therefore oblivious to the topolog-
ical ground-state degeneracy associated with
the presence of MZMs. Both in superconduct-
ing platforms and in FQH systems, experi-
mental tests of the unique fusion properties
would be highly desirable to establish the non-
abelian behavior of these quasiparticles. More
stringent, next-generation tests of MZMs would
also probe the coherent dynamics within their
degenerate subspace (see Fig. 4). This consti-
tutes an essential precursor to implementing
topological qubits and to full-scale tests of non-
abelian statistics.
Quantum Hall systems
A major goal for studies of FQH systems is to
find more-convincing evidence for non-abelian
quasiparticles. It would be desirable to improve
measuring techniques for interference experi-
Yazdani et al., Science 380, eade0850 (2023) 23 June 2023
ments on quasiparticles traveling along edge
states. It would also be helpful to obtain better
control of the electron gas in GaAs devices, for
instance, to control more precisely when a
localized quasiparticle enters or leaves the
area enclosed by the interferometer. It is also
important to pin down the actual nature of
the states at filling factors n = 5/2 and 7/2 and
to resolve the present disagreements between
theory and experiment.
An exciting new direction for exploring non-
abelian anyons is to examine the even-denom-
inator FQH states observed in systems other
than GaAs (127–131). Bilayer graphene consti-
tutes a particularly promising system (127, 128).
When encapsulated by hexagonal boron ni-
tride, it has a smaller dielectric constant (~4.4)
than GaAs systems (~12.8), implying stronger
Coulomb interactions. Consequently, experi-
ments on bilayer graphene detected even-
denominator states with gaps that are an
order of magnitude larger (5 K at 14 T) than in
GaAs. Recent experiments indicate that these
even-denominator states are dominantly spin
and valley polarized, ruling out competing
abelian ground states (128) (Fig. 5A). By com-
bining 2D layered materials and their stacks
with new nanofabrication techniques, interfer-
ence as well as fusion experiments with trapped
non-abelian anyons may become possible. Re-
cent advances in STM spectroscopy of FQH
states (132) may also be used to detect non-
abelian anyons trapped near defects (133).
Quasiparticles with richer non-abelian sta-
tistics than that of Ising anyons would ul-
timately be needed for universal quantum
computation with full topological protection.
Suitable excitations have been predicted for
certain quantized Hall states, such as the n =
12/5 state in GaAs structures (134–140). How-
ever, corresponding experimental evidence is
not strong at the present time. Theoretical
work has also shown that states supporting
various types of non-abelian quasiparticles
can be produced in hybrid structures of abe-
lian quantum Hall systems and supercon-
ductors (141–143). These proposals have, so
far, not been realized in experiment because
superconductivity and quantum Hall systems
have conflicting magnetic field requirements.
Moreover, the anyons produced in the struc-
tures proposed in (141) and (142) would still
fall short of what is needed for universal
topologically protected quantum computing.
The proposal of (143) realizes anyons that
allow universal topological quantum compu-
tation but is rather challenging to implement
experimentally.
Nevertheless, preliminary experiments,
in which a narrow superconducting finger
penetrates the edge of a Hall bar, have been
successfully realized, and crossed Andreev
processes have been detected (144). In this
geometry, the finger can be thought of as bend-
ing the chiral edge modes along its boundary,
thereby forming two counter-propagating
modes proximity-coupled to a superconductor.
If the edge channels are spin polarized and
the superconductor is intrinsically spin-orbit
coupled to facilitate pairing, this setup might
support topological superconductivity. The
problem of large magnetic fields can be cir-
cumvented by using thin films of magneti-
cally doped TIs, which exhibit the quantum
anomalous Hall effect even in the absence
of a magnetic field.
Perhaps even more intriguing is the possi-
bility of proximity coupling fractional Chern
insulating states and superconductors (145).
Theory predicts that such hybrid systems can
also host more complex excitations, known as
parafermions, with richer braiding statistics
than Ising anyons (146). Realizing fractional
Chern insulators (FCIs) in conditions com-
patible with superconductivity is most likely
p
to occur in new materials systems, such as the
moiré materials considered in the section
titled New material platforms.
Josephson junctions
Several approaches have been proposed to im-
g
prove the existing experiments that search for
topological superconductivity in Josephson
junctions. This platform exploits the fact that
applied supercurrents reduce the threshold
magnetic field required to enter the topolog-
y
ical phase (147). As shown recently, supercur-
rents can obviate the need for a magnetic field
entirely, with the applied phase difference
across the junction providing the required
breaking of time-reversal symmetry (148–151).
One approach inserts a chain of additional
superconducting islands into the original
Josephson junction geometry. For appropriate
phase biasing of the islands relative to the
superconducting banks of the Josephson junc-
y g
tion, each unit cell that contains a single island
harbors two MZMs (Fig. 5B). Neighboring
MZMs that belong to different unit cells gap
out, leaving only the isolated MZMs at the
ends of the junction intact. In this setting, the
,
additional phase knob introduced by the inter-
mediate islands enables the realization of a
topological superconducting phase in the ab-
sence of a magnetic field. The localization
length of the MZMs is set by the size of the
unit cell, so that the MZMs can be effectively
separated further by adding more unit cells
along the wire. Counterintuitively, it has been
pointed out that disorder can stabilize the topo-
logical phase in this platform (152, 153).
1D topological superconductors
Work to date has focused on MZMs in systems
with broken time-reversal symmetry. Time-
reversal symmetric topological superconduc-
tors support stable Kramers pairs of MZMs
with fractional boundary spin (154). Advances
7 of 12
RES EARCH | R E V I E W

A C

B D

Fig. 5. Future directions. A number of recent developments are expanding the
material platforms and device concepts for realizing MZMs and other non-abelian
quasiparticles. (A) A graphene bilayer encapsulated by boron nitride offers a
p
g
same material (160) and thereby creating the conditions for MZMs. Similarly,
inducing spin-orbit coupling into graphene-based superconducting systems by
proximity (165) can be used to create new platforms for MZMs. QSHI, quantum

promising platform for pursuing experiments on even denominator FQH states,
detected (right) by capacitance measurements as a function of electron density
and perpendicular displacement field (127, 128). BLG, bilayer graphene; hBN,
hexagonal boron nitride. [Adapted by permission from Springer Nature Customer
Service Center GmbH, Springer Nature (128), copyright (2017)] (B) Chain of
Josephson junctions with multiple superconducting islands. Controlling their
superconducting phases is predicted to allow for realizing more localized MZMs.
[Reproduced from (148)] (C) The rapidly expanding field of 2D van der Waals
materials is creating opportunities for future research on MZMs. The discovery of
both a topological insulating phase and superconductivity in gated WTe2 devices
y
spin Hall insulator. [Left image adapted with permission from (160); right image
adapted by permission from Springer Nature Customer Service Center GmbH,
Springer Nature (165), copyright (2019)] (D) Moiré materials (left) have
substantially expanded the discovery of superconducting and topological phases,
including interaction-driven Chern and fractional Chern insulating phases
(166, 167). Compressibility measurements on magic-angle graphene as a function
of carrier density and magnetic field (right) provide a highly sensitive probe of
such phases. [Left image adapted by permission from Springer Nature Customer
Service Center GmbH, Springer Nature (185), copyright (2020); right image
adapted by permission from Springer Nature Customer Service Center GmbH,

y g
has opened up the possibility of laterally combining these phases within the Springer Nature (167), copyright (2021)]

in creating more robust 1D topological super- in the context of topological quantum comput- promises to play a central role in advancing
conductors in hybrid structures would open ing (11). Islands with four MZMs, known as the field of topological quantum phenomena
exciting possibilities for engineering more com- tetrons (31) or Majorana box qubits (46), ef- and non-abelian anyons. The potential of new

plex phases with previously uncharacterized
topological properties and for exploring ex-
citing quantum phenomena. Josephson junc-
tions of 1D topological superconductors should
exhibit the celebrated 4p-periodic Josephson
effect. Coupling several 1D topological super-
conductors to a disordered quantum dot has
been predicted to provide a realization of the
Sachdev-Ye-Kitaev model (155, 156). Signatures
of this phase can be observed by tunneling
into the quantum dot. Arrays of 1D systems
with controllable coupling between them pro-
vide a pathway to realizing a 2D topological
superconducting phase with multiple chiral
and nonchiral modes at its boundary (157).
Coulomb blockaded islands with multiple
MZMs are particularly powerful building blocks
fectively implement a topological qubit or,
equivalently, local spin degrees of freedom
(see Fig. 4). Hybridization of MZMs between
islands realizes anisotropic spin-spin interac-
tions or even higher spin-cluster interactions.
This can be exploited to faithfully realize the
Hamiltonians of topological error-correcting
codes (158, 159). Although such networks of
coupled islands of topological superconduc-
tors realize 2D topological superconductors
in the absence of charging effects, charging
can effectively drive them into topologically
ordered insulating states.
New material platforms
Beyond improvements to existing material
platforms, the discovery of new materials
,
materials is exemplified by the discovery of
tunable superconductivity and topological
phases in monolayers of WTe2 (Fig. 5C) (160).
WTe2 combines superconductivity with strong
intrinsic spin-orbit coupling as well as gate
tunability into a 2D topological-insulator phase.
This material is thus a promising platform for
realizing phase-controlled topological super-
conductivity without the need to induce super-
conductivity externally.
Robust and tunable superconductivity has
also been observed in twisted matter such as
magic-angle twisted bilayer, trilayer, and multi-
layer graphene (161–164). Although graphene
has very weak intrinsic spin-orbit coupling,
spin-orbit coupling can be induced by includ-
ing, for example, additional WSe2 layers into

Yazdani et al., Science 380, eade0850 (2023) 23 June 2023 8 of 12

RES EARCH | R E V I E W
the device (165) (Fig. 5C). These systems have
been shown to support FCI states down to
low magnetic fields (166). Most recently, FCIs
have been reported at magnetic fields as low
as 5 T (167) (Fig. 5D). In present samples, other
phases are stabilized at lower magnetic fields.
Understanding the competition between these
phases and what physical parameters favor the
FCI state will help reduce the magnetic field in
which FCIs can form such that superconducti-
vity and FCI phases can coexist in the same sample.
These systems could thus provide an inter-
esting platform to search for parafermions,
combining low magnetic fields with intrinsic
superconductivity. Proximity coupling various
van der Waals materials with magnetic and
superconducting ground states may also be
used to create hybrid structures for topological
superconductivity. It would be particularly ex-
citing if non-abelian FQH states could be real-
ized in ferromagnetic structures without an
applied magnetic field. So far, there is evidence
for zero-field integer anomalous quantum Hall
states in several materials (168, 169), but frac-
tional states have not yet been observed.
There have also been efforts to combine
various magnetic materials, including doped
TIs, with superconductors to realize chiral
Majorana edge modes. These works are moti-
vated by proposals to use chiral edge modes
as a platform for topological quantum com-
puting (170). However, this area is, at present,
far less advanced experimentally, and initial
results (171) have been retracted (172) and
shown to lack reproducibility (173). Chiral
Majorana edge modes also appear in the exactly
solvable Kitaev honeycomb model, a spin model
with anisotropic exchange couplings (174). Ex-
perimental realizations have been proposed
based on judiciously engineering the hybrid-
ization of Coulomb-blockaded islands hosting
multiple MZMs (11, 175, 176). Realizations in
Kitaev materials (177, 178), possibly evidenced
by thermal transport experiments (179–181),
have also been discussed in the context of
Majorana-based quantum computation (182).
During the exploitation of steady advances
of quantum simulations with existing quantum
computing platforms, there have also been
interesting efforts (183) to simulate Floquet
incarnations of spin models such as the quan-
tum Ising model, which is closely related to
the Kitaev chain. These models also support
zero (as well as p) modes and provide a setting
for studying their stability to perturbations in
chains of superconducting qubits. This arena
may help explore the utility of topological con-
cepts such as MZMs in advancing noisy inter-
mediate quantum technologies beyond their
present coherence limitations.
Conclusion
Experimentally establishing the existence of
non-abelian anyons such as MZMs constitutes
Yazdani et al., Science 380, eade0850 (2023) 23 June 2023
an outstandingly worthwhile goal, not only
from the point of view of fundamental physics
but also because of their potential applications.
Triggered by the theoretical understanding of
non-abelian FQH states and superconductor-TI
hybrids, there have been numerous attempts to
observe non-abelian anyons in the laboratory.
Many claims are based on rather circumstan-
tial evidence and, apart from a few extensively
studied platforms, were not subjected to intense
scrutiny or in-depth analysis of alternative
interpretations. Future progress will be possi-
ble when claims of Majorana discoveries are
based on experimental tests that go greatly
beyond indicators such as zero-bias peaks, which,
at best, suggest consistency with a Majorana
interpretation. It will be equally essential that
these discoveries build on an excellent under-
standing of the underlying materials system.
It seems likely that further material improve-
ments of existing platforms and the explora-
tion of new material platforms will both be
important avenues to make progress toward
solid evidence for Majoranas. Then we can hope
to explore—and harness—the fascinating physics
of the topologically protected ground-state
manifold and non-abelian statistics.
RE FERENCES AND NOTES
1. F. Arute et al., Quantum supremacy using a programmable
superconducting processor. Nature 574, 505–510 (2019).
doi: 10.1038/s41586-019-1666-5; pmid: 31645734
2. J. Preskill, Quantum computing and the entanglement
frontier. arXiv:1203.5813 [quant-ph] (2012).
3. M. Kjaergaard et al., Superconducting qubits: Current state of
play. Annu. Rev. Condens. Matter Phys. 11, 369–395 (2020).
doi: 10.1146/annurev-conmatphys-031119-050605
4. L. M. K. Vandersypen, M. A. Eriksson, Quantum computing
with semiconductor spins. Phys. Today 72, 38–45 (2019).
doi: 10.1063/PT.3.4270
5. C. Monroe et al., Programmable quantum simulations of spin
systems with trapped ions. Rev. Mod. Phys. 93, 025001
(2021). doi: 10.1103/RevModPhys.93.025001
6. A. Browaeys, T. Lahaye, Many-body physics with individually
controlled Rydberg atoms. Nat. Phys. 16, 132–142 (2020).
doi: 10.1038/s41567-019-0733-z
7. A. Y. Kitaev, Fault-tolerant quantum computation by
anyons. Ann. Phys. 303, 2–30 (2003). doi: 10.1016/
S0003-4916(02)00018-0
8. C. Nayak, S. H. Simon, A. Stern, M. Freedman, S. Das Sarma,
Non-Abelian anyons and topological quantum computation.
Rev. Mod. Phys. 80, 1083–1159 (2008). doi: 10.1103/
RevModPhys.80.1083
9. C. W. J. Beenakker, Search for Majorana fermions in
superconductors. Annu. Rev. Condens. Matter Phys. 4, 113–136
(2013). doi: 10.1146/annurev-conmatphys-030212-184337
10. J. Alicea, New directions in the pursuit of Majorana
fermions in solid state systems. Rep. Prog. Phys. 75,
076501 (2012). doi: 10.1088/0034-4885/75/7/076501;
pmid: 22790778
11. Y. Oreg, F. von Oppen, Majorana zero modes in networks of
Cooper-pair boxes: Topologically ordered states and
topological quantum computation. Annu. Rev. Condens.
Matter Phys. 11, 397–420 (2020). doi: 10.1146/annurev-
conmatphys-031218-013618
12. R. Aguado, L. Kouwenhoven, Majorana qubits for topological
quantum computing. Phys. Today 73, 44–50 (2020).
doi: 10.1063/PT.3.4499
13. M. H. Freedman, P/NP, and the quantum field computer.
Proc. Natl. Acad. Sci. U.S.A. 95, 98–101 (1998). doi: 10.1073/
pnas.95.1.98; pmid: 9419335
14. D. P. Arovas, J. R. Schrieffer, F. Wilczek, Fractional statistics
and the quantum Hall effect. Phys. Rev. Lett. 53, 722–723
(1984). doi: 10.1103/PhysRevLett.53.722
15. B. I. Halperin, Statistics of quasiparticles and the hierarchy of
fractional quantized Hall states. Phys. Rev. Lett. 52, 2390
(1984). doi: 10.1103/PhysRevLett.52.2390.4
16. G. Moore, N. Read, Nonabelions in the fractional quantum
Hall effect. Nucl. Phys. B 360, 362–396 (1991). doi: 10.1016/
0550-3213(91)90407-O
17. N. Read, D. Green, Paired states of fermions in two
dimensions with breaking of parity and time-reversal
symmetries and the fractional quantum Hall effect. Phys. Rev.
B 61, 10267–10297 (2000). doi: 10.1103/PhysRevB.61.10267
18. R. Willett et al., Observation of an even-denominator
quantum number in the fractional quantum Hall effect. Phys.
Rev. Lett. 59, 1776–1779 (1987). doi: 10.1103/
PhysRevLett.59.1776; pmid: 10035326
19. A. Y. Kitaev, Unpaired Majorana fermions in quantum wires.
Phys. Uspekhi 44, 131–136 (2001). doi: 10.1070/1063-7869/
44/10S/S29
20. L. Fu, C. L. Kane, Superconducting proximity effect and
Majorana fermions at the surface of a topological insulator.
Phys. Rev. Lett. 100, 096407 (2008). doi: 10.1103/
PhysRevLett.100.096407; pmid: 18352737
21. L. Fu, C. L. Kane, Josephson current and noise at a
superconductor/quantum-spin-Hall-insulator/superconductor
junction. Phys. Rev. B 79, 161408 (2009). doi: 10.1103/
PhysRevB.79.161408
22. X. G. Wen, Non-Abelian statistics in the fractional quantum
Hall states. Phys. Rev. Lett. 66, 802–805 (1991).
p
doi: 10.1103/PhysRevLett.66.802
23. N. Read, E. Rezayi, Beyond paired quantum Hall states:
Parafermions and incompressible states in the first excited
Landau level. Phys. Rev. B 59, 8084–8092 (1999).
doi: 10.1103/PhysRevB.59.8084
24. C. Nayak, F. Wilczek, 2n-quasihole states realize 2n−1-
dimensional spinor braiding statistics in paired quantum Hall
states. Nucl. Phys. B 479, 529–553 (1996). doi: 10.1016/
0550-3213(96)00430-0
g
25. D. A. Ivanov, Non-Abelian statistics of half-quantum
vortices in p-wave superconductors. Phys. Rev. Lett. 86,
268–271 (2001). doi: 10.1103/PhysRevLett.86.268;
pmid: 11177808
26. J. Sau, D. Clarke, S. Tewari, Controlling non-Abelian statistics
y
of Majorana fermions in semiconductor nanowires. Phys. Rev.
B 84, 094505 (2011). doi: 10.1103/PhysRevB.84.094505
27. B. van Heck, A. R. Akhmerov, F. Hassler, M. Burrello,
C. W. J. Beenakker, Coulomb-assisted braiding of Majorana
fermions in a Josephson junction array. New J. Phys. 14,
035019 (2012). doi: 10.1088/1367-2630/14/3/035019
28. T. Hyart et al., Flux-controlled quantum computation with
Majorana fermions. Phys. Rev. B 88, 035121 (2013).
doi: 10.1103/PhysRevB.88.035121
29. D. Aasen et al., Milestones Toward Majorana-based quantum
computing. Phys. Rev. X 6, 031016 (2016). doi: 10.1103/
PhysRevX.6.031016
30. P. Bonderson, M. Freedman, C. Nayak, Measurement-only
y g
topological quantum computation. Phys. Rev. Lett. 101,
010501 (2008). doi: 10.1103/PhysRevLett.101.010501;
pmid: 18764095
31. T. Karzig et al., Scalable designs for quasiparticle-poisoning-
protected topological quantum computation with
Majorana zero modes. Phys. Rev. B 95, 235305 (2017).
doi: 10.1103/PhysRevB.95.235305
32. D. Litinski, F. von Oppen, Quantum computing with Majorana
,
fermion codes. Phys. Rev. B 97, 205404 (2018). doi: 10.1103/
PhysRevB.97.205404
33. C. Knapp, J. I. Väyrynen, R. M. Lutchyn, Number-conserving
analysis of measurement-based braiding with Majorana zero
modes. Phys. Rev. B 101, 125108 (2020). doi: 10.1103/
PhysRevB.101.125108
34. J. Alicea, Y. Oreg, G. Refael, F. von Oppen, M. P. A. Fisher,
Non-Abelian statistics and topological quantum information
processing in 1D wire networks. Nat. Phys. 7, 412–417
(2011). doi: 10.1038/nphys1915
35. J. D. Sau, R. M. Lutchyn, S. Tewari, S. Das Sarma, Generic
new platform for topological quantum computation using
semiconductor heterostructures. Phys. Rev. Lett. 104,
040502 (2010). doi: 10.1103/PhysRevLett.104.040502;
pmid: 20366693
36. J. Alicea, Majorana fermions in a tunable semiconductor
device. Phys. Rev. B 81, 125318 (2010). doi: 10.1103/
PhysRevB.81.125318
37. R. M. Lutchyn, J. D. Sau, S. Das Sarma, Majorana fermions
and a topological phase transition in semiconductor-
superconductor heterostructures. Phys. Rev. Lett. 105,
9 of 12
RES EARCH | R E V I E W
077001 (2010). doi: 10.1103/PhysRevLett.105.077001;
pmid: 20868069
38. Y. Oreg, G. Refael, F. von Oppen, Helical liquids and
Majorana bound states in quantum wires. Phys. Rev. Lett.
105, 177002 (2010). doi: 10.1103/PhysRevLett.105.177002;
pmid: 21231073
39. P. Krogstrup et al., Epitaxy of semiconductor-superconductor
nanowires. Nat. Mater. 14, 400–406 (2015). doi: 10.1038/
nmat4176; pmid: 25581626
40. J. Shabani et al., Two-dimensional epitaxial superconductor-
semiconductor heterostructures: A platform for topological
superconducting networks. Phys. Rev. B 93, 155402
(2016). doi: 10.1103/PhysRevB.93.155402
41. S. M. Albrecht et al., Exponential protection of zero modes in
Majorana islands. Nature 531, 206–209 (2016). doi: 10.1038/
nature17162; pmid: 26961654
42. B. van Heck, R. M. Lutchyn, L. I. Glazman, Conductance
of a proximitized nanowire in the Coulomb blockade
regime. Phys. Rev. B 93, 235431 (2016). doi: 10.1103/
PhysRevB.93.235431
43. L. Fu, Electron teleportation via Majorana bound states
in a mesoscopic superconductor. Phys. Rev. Lett. 104,
056402 (2010). doi: 10.1103/PhysRevLett.104.056402;
pmid: 20366777
44. B. Béri, N. R. Cooper, Topological Kondo effect with Majorana
fermions. Phys. Rev. Lett. 109, 156803 (2012). doi: 10.1103/
PhysRevLett.109.156803; pmid: 23102351
45. S. M. Albrecht et al., Transport signatures of quasiparticle
poisoning in a Majorana island. Phys. Rev. Lett. 118,
137701 (2017). doi: 10.1103/PhysRevLett.118.137701;
pmid: 28409973
46. S. Plugge, A. Rasmussen, R. Egger, K. Flensberg, Majorana
box qubits. New J. Phys. 19, 012001 (2017). doi: 10.1088/
1367-2630/aa54e1
47. S. Nadj-Perge, I. K. Drozdov, B. A. Bernevig, A. Yazdani,
Proposal for realizing Majorana fermions in chains of
magnetic atoms on a superconductor. Phys. Rev. B 88,
020407 (2013). doi: 10.1103/PhysRevB.88.020407
48. S. Nadj-Perge et al., Observation of Majorana fermions in
ferromagnetic atomic chains on a superconductor. Science
346, 602–607 (2014). doi: 10.1126/science.1259327;
pmid: 25278507
49. J. Li et al., Topological superconductivity induced by
ferromagnetic metal chains. Phys. Rev. B 90, 235433 (2014).
doi: 10.1103/PhysRevB.90.235433
50. B. Braunecker, P. Simon, Interplay between classical
magnetic moments and superconductivity in quantum one-
dimensional conductors: Toward a self-sustained topological
Majorana phase. Phys. Rev. Lett. 111, 147202 (2013).
doi: 10.1103/PhysRevLett.111.147202; pmid: 24138267
51. J. Klinovaja, P. Stano, A. Yazdani, D. Loss, Topological
superconductivity and Majorana fermions in RKKY systems.
Phys. Rev. Lett. 111, 186805 (2013). doi: 10.1103/
PhysRevLett.111.186805; pmid: 24237550
52. M. M. Vazifeh, M. Franz, Self-organized topological state with
Majorana fermions. Phys. Rev. Lett. 111, 206802 (2013).
doi: 10.1103/PhysRevLett.111.206802; pmid: 24289700
53. Y. Peng, F. Pientka, L. I. Glazman, F. von Oppen, Strong
localization of Majorana end States in chains of magnetic
adatoms. Phys. Rev. Lett. 114, 106801 (2015). doi: 10.1103/
PhysRevLett.114.106801; pmid: 25815952
54. F. Pientka, L. I. Glazman, F. von Oppen, Topological
superconducting phase in helical Shiba chains. Phys. Rev. B
88, 155420 (2013). doi: 10.1103/PhysRevB.88.155420
55. J. F. Steiner, C. Mora, K. J. Franke, F. von Oppen, Quantum
magnetism and topological superconductivity in Yu-Shiba-
Rusinov chains. Phys. Rev. Lett. 128, 036801 (2022).
doi: 10.1103/PhysRevLett.128.036801; pmid: 35119905
56. F. Pientka et al., Topological superconductivity in a planar
Josephson junction. Phys. Rev. X 7, 021032 (2017).
doi: 10.1103/PhysRevX.7.021032
57. M. Hell, K. Flensberg, M. Leijnse, Distinguishing Majorana
bound states from localized Andreev bound states by
interferometry. Phys. Rev. B 97, 161401 (2018). doi: 10.1103/
PhysRevB.97.161401
58. S. Das Sarma, M. Freedman, C. Nayak, Topologically
protected qubits from a possible non-Abelian fractional
quantum Hall state. Phys. Rev. Lett. 94, 166802 (2005).
doi: 10.1103/PhysRevLett.94.166802; pmid: 15904258
59. A. Stern, B. I. Halperin, Proposed experiments to probe
the non-Abelian n = 5/2 quantum hall state. Phys. Rev. Lett.
96, 016802 (2006). doi: 10.1103/PhysRevLett.96.016802;
pmid: 16486495
Yazdani et al., Science 380, eade0850 (2023) 23 June 2023
60. J. Nakamura, S. Liang, G. C. Gardner, M. J. Manfra, Direct
observation of anyonic braiding statistics. Nat. Phys. 16,
931–936 (2020). doi: 10.1038/s41567-020-1019-1
61. R. L. Willett et al., Interference measurements of non-Abelian
e/4 & Abelian e/2 quasiparticle braiding. Phys. Rev. X 13,
011028 (2023). doi: 10.1103/PhysRevX.13.011028
62. A. Bid, N. Ofek, M. Heiblum, V. Umansky, D. Mahalu,
Observation of a quarter of an electron charge at the n = 5/2
quantum Hall state. Nature 452, 829–834 (2008).
doi: 10.1038/nature06855; pmid: 18421345
63. V. Venkatachalam, A. Yacoby, L. Pfeiffer, K. West, Local
charge of the n = 5/2 fractional quantum Hall state. Nature
469, 185–188 (2011). doi: 10.1038/nature09680;
pmid: 21228871
64. M. Banerjee et al., Observation of half-integer thermal Hall
conductance. Nature 559, 205–210 (2018). doi: 10.1038/
s41586-018-0184-1; pmid: 29867160
65. B. Dutta et al., Distinguishing between non-abelian
topological orders in a quantum Hall system. Science 375,
193–197 (2022). doi: 10.1126/science.abg6116;
pmid: 34941364
66. D. E. Feldman, B. I. Halperin, Fractional charge and
fractional statistics in the quantum Hall effects. Rep. Prog.
Phys. 84, 076501 (2021). doi: 10.1088/1361-6633/ac03aa;
pmid: 34015771
67. V. Mourik et al., Signatures of Majorana fermions in hybrid
superconductor-semiconductor nanowire devices. Science
336, 1003–1007 (2012). doi: 10.1126/science.1222360;
pmid: 22499805
68. R. M. Lutchyn et al., Majorana zero modes in superconductor–
semiconductor heterostructures. Nat. Rev. Mater. 3, 52–68
(2018). doi: 10.1038/s41578-018-0003-1
69. G. Kells, D. Meidan, P. W. Brouwer, Near-zero-energy end
states in topologically trivial spin-orbit coupled
superconducting nanowires with a smooth confinement.
Phys. Rev. B 86, 100503 (2012). doi: 10.1103/
PhysRevB.86.100503
70. E. Prada, P. San-Jose, R. Aguado, Transport spectroscopy of
N S nanowire junctions with Majorana fermions. Phys. Rev. B
86, 180503 (2012). doi: 10.1103/PhysRevB.86.180503
71. E. J. Lee et al., Zero-bias anomaly in a nanowire quantum
dot coupled to superconductors. Phys. Rev. Lett. 109,
186802 (2012). doi: 10.1103/PhysRevLett.109.186802;
pmid: 23215310
72. E. J. H. Lee et al., Spin-resolved Andreev levels and parity
crossings in hybrid superconductor-semiconductor
nanostructures. Nat. Nanotechnol. 9, 79–84 (2014).
doi: 10.1038/nnano.2013.267; pmid: 24336403
73. C. X. Liu, J. D. Sau, T. D. Stanescu, S. Das Sarma, Andreev
bound states versus Majorana bound states in quantum
dot-nanowire-superconductor hybrid structures: Trivial
versus topological zero-bias conductance peaks. Phys. Rev. B
96, 075161 (2017). doi: 10.1103/PhysRevB.96.075161
74. A. Vuik, B. Nijholt, A. R. Akhmerov, M. Wimmer, Reproducing
topological properties with quasi-Majorana states. SciPost
Phys. 7, 61 (2019). doi: 10.21468/SciPostPhys.7.5.061
75. S. Vaitiekėnas et al., Flux-induced topological
superconductivity in full-shell nanowires. Science 367,
eaav3392 (2020). doi: 10.1126/science.aav3392;
pmid: 32217701
76. H. H. Thorp, Editorial expression of concern. Science 373,
500 (2021). doi: 10.1126/science.abl5286; pmid: 34326226
77. M. Valentini et al., Nontopological zero-bias peaks in
full-shell nanowires induced by flux-tunable Andreev states.
Science 373, 82–88 (2021). doi: 10.1126/science.abf1513;
pmid: 34210881
78. M. Valentini et al., Majorana-like Coulomb spectroscopy in the
absence of zero-bias peaks. Nature 612, 442–447 (2022).
doi: 10.1038/s41586-022-05382-w; pmid: 36517713
79. K. T. Law, P. A. Lee, T. K. Ng, Majorana fermion induced
resonant Andreev reflection. Phys. Rev. Lett. 103, 237001
(2009). doi: 10.1103/PhysRevLett.103.237001;
pmid: 20366164
80. K. Flensberg, Tunneling characteristics of a chain of Majorana
bound states. Phys. Rev. B 82, 180516 (2010). doi: 10.1103/
PhysRevB.82.180516
81. M. Wimmer, A. R. Akhmerov, J. P. Dahlhaus, C. W. J. Beenakker,
Quantum point contact as a probe of a topological
superconductor. New J. Phys. 13, 053016 (2011).
doi: 10.1088/1367-2630/13/5/053016
82. F. Pientka, G. Kells, A. Romito, P. W. Brouwer, F. von Oppen,
Enhanced zero-bias Majorana peak in the differential
tunneling conductance of disordered multisubband quantum-
wire/superconductor junctions. Phys. Rev. Lett. 109,
227006 (2012). doi: 10.1103/PhysRevLett.109.227006;
pmid: 23368154
83. H. Zhang et al., RETRACTED ARTICLE: Quantized Majorana
conductance. Nature 556, 74–79 (2018). doi: 10.1038/
nature26142; pmid: 29590094
84. H. Zhang et al., Retraction note: Quantized Majorana
conductance. Nature 591, E30 (2021). doi: 10.1038/s41586-
021-03373-x; pmid: 33686283
85. S. Frolov, Quantum computing’s reproducibility crisis:
Majorana fermions. Nature 592, 350–352 (2021).
doi: 10.1038/d41586-021-00954-8; pmid: 33846620
86. F. Nichele et al., Scaling of Majorana zero-bias conductance
peaks. Phys. Rev. Lett. 119, 136803 (2017). doi: 10.1103/
PhysRevLett.119.136803; pmid: 29341695
87. H. Pan, W. S. Cole, J. D. Sau, S. Das Sarma, Generic
quantized zero-bias conductance peaks in superconductor–
semiconductor hybrid structures. Phys. Rev. B 101, 024506
(2020). doi: 10.1103/PhysRevB.101.024506
88. A. Banerjee et al., Local and nonlocal transport spectroscopy
in planar Josephson junctions. Phys. Rev. Lett. 130,
096202 (2023). doi: 10.1103/PhysRevLett.130.096202;
pmid: 36930915
89. Microsoft Quantum, InAs-Al hybrid devices passing the
topological gap protocol. arXiv.2207.02472 [cond-mat.mes-
hall] (2022).
90. O. Motrunich, K. Damle, D. A. Huse, Griffiths effects and
p
quantum critical points in dirty superconductors without spin-
rotation invariance: One-dimensional examples. Phys. Rev. B 63,
224204 (2001). doi: 10.1103/PhysRevB.63.224204
91. P. W. Brouwer, M. Duckheim, A. Romito, F. von Oppen,
Topological superconducting phases in disordered quantum
wires with strong spin-orbit coupling. Phys. Rev. B 84,
144526 (2011). doi: 10.1103/PhysRevB.84.144526
92. P. W. Brouwer, M. Duckheim, A. Romito, F. von Oppen,
Probability distribution of Majorana end-state energies in
g
disordered wires. Phys. Rev. Lett. 107, 196804 (2011).
doi: 10.1103/PhysRevLett.107.196804; pmid: 22181633
93. D. Bagrets, A. Altland, Class D spectral peak in Majorana
quantum wires. Phys. Rev. Lett. 109, 227005 (2012).
doi: 10.1103/PhysRevLett.109.227005; pmid: 23368153
y
94. J. D. Sau, S. Tewari, S. Das Sarma, Experimental and
materials considerations for the topological superconducting
state in electron- and hole-doped semiconductors: Searching
for non-Abelian Majorana modes in 1D nanowires and 2D
heterostructures. Phys. Rev. B 85, 064512 (2012).
doi: 10.1103/PhysRevB.85.064512
95. S. Ahn, H. Pan, B. Woods, T. D. Stanescu, S. Das Sarma,
Estimating disorder and its adverse effects in semiconductor
Majorana nanowires. Phys. Rev. Mater. 5, 124602 (2021).
doi: 10.1103/PhysRevMaterials.5.124602
96. D. I. Pikulin et al., Protocol to identify a topological
superconducting phase in a three-terminal device.
arXiv:2103.12217 [cond-mat.mes-hall] (2021).
y g
97. C. Reeg, D. Loss, J. Klinovaja, Metallization of a Rashba wire
by a superconducting layer in the strong-proximity regime.
Phys. Rev. B 97, 165425 (2018). doi: 10.1103/
PhysRevB.97.165425
98. T. Kiendl, F. von Oppen, P. W. Brouwer, Renormalization
effects in spin-polarized metallic wires proximitized by a
superconductor: A scattering approach. Phys. Rev. B 99,
104510 (2019). doi: 10.1103/PhysRevB.99.104510
,
99. M. Ruby et al., End states and subgap structure in proximity-
coupled chains of magnetic adatoms. Phys. Rev. Lett. 115,
197204 (2015). doi: 10.1103/PhysRevLett.115.197204;
pmid: 26588411
100. R. Pawlak et al., Probing atomic structure and Majorana
wavefunctions in mono-atomic Fe chains on superconducting
Pb surface. NPJ Quantum Inf. 2, 16035 (2016). doi: 10.1038/
npjqi.2016.35
101. B. E. Feldman et al., High-resolution studies of the Majorana
atomic chain platform. Nat. Phys. 13, 286–291 (2017).
doi: 10.1038/nphys3947
102. S. Jeon et al., Distinguishing a Majorana zero mode using
spin-resolved measurements. Science 358, 772–776 (2017).
doi: 10.1126/science.aan3670; pmid: 29025997
103. J. J. He, T. K. Ng, P. A. Lee, K. T. Law, Selective equal-spin
Andreev reflections induced by Majorana fermions. Phys. Rev.
Lett. 112, 037001 (2014). doi: 10.1103/PhysRevLett.112.037001;
pmid: 24484161
104. J. D. Sau, P. M. R. Brydon, Bound states of a ferromagnetic
wire in a superconductor. Phys. Rev. Lett. 115, 127003 (2015).
doi: 10.1103/PhysRevLett.115.127003; pmid: 26431011
10 of 12
RES EARCH | R E V I E W
105. R. Hess, H. F. Legg, D. Loss, J. Klinovaja, Prevalence of trivial
zero-energy subgap states in nonuniform helical spin
chains on the surface of superconductors. Phys. Rev. B 106,
104503 (2022). doi: 10.1103/PhysRevB.106.104503
106. M. Ruby, B. W. Heinrich, Y. Peng, F. von Oppen, K. J. Franke,
Exploring a proximity-coupled Co chain on Pb(110) as a
possible Majorana platform. Nano Lett. 17, 4473–4477
(2017). doi: 10.1021/acs.nanolett.7b01728; pmid: 28640633
107. H. Kim et al., Toward tailoring Majorana bound states in
artificially constructed magnetic atom chains on elemental
superconductors. Sci. Adv. 4, eaar5251 (2018). doi: 10.1126/
sciadv.aar5251; pmid: 29756034
108. A. Kamlapure, L. Cornils, J. Wiebe, R. Wiesendanger,
Engineering the spin couplings in atomically crafted spin
chains on an elemental superconductor. Nat. Commun. 9,
3253 (2018). doi: 10.1038/s41467-018-05701-8;
pmid: 30108221
109. L. Schneider et al., Topological Shiba bands in artificial spin
chains on superconductors. Nat. Phys. 17, 943–948 (2021).
doi: 10.1038/s41567-021-01234-y
110. E. Liebhaber et al., Quantum spins and hybridization in
artificially-constructed chains of magnetic adatoms on a
superconductor. Nat. Commun. 13, 2160 (2022).
doi: 10.1038/s41467-022-29879-0; pmid: 35443753
111. L. Schneider et al., Precursors of Majorana modes and their
length-dependent energy oscillations probed at both ends of
atomic Shiba chains. Nat. Nanotechnol. 17, 384–389 (2022).
doi: 10.1038/s41565-022-01078-4; pmid: 35256768
112. P. Hosur, P. Ghaemi, R. S. K. Mong, A. Vishwanath, Majorana
modes at the ends of superconductor vortices in doped
topological insulators. Phys. Rev. Lett. 107, 097001 (2011).
doi: 10.1103/PhysRevLett.107.097001; pmid: 21929261
113. C.-K. Chiu, M. J. Gilbert, T. L. Hughes, Vortex lines in topological
insulator-superconductor heterostructures. Phys. Rev. B 84,
144507 (2011). doi: 10.1103/PhysRevB.84.144507
114. D. Wang et al., Evidence for Majorana bound states in an iron-
based superconductor. Science 362, 333–335 (2018).
doi: 10.1126/science.aao1797; pmid: 30115743
115. L. Kong et al., Half-integer level shift of vortex bound states in
an iron-based superconductor. Nat. Phys. 15, 1181–1187
(2019). doi: 10.1038/s41567-019-0630-5
116. B. Jäck, Y. Xie, A. Yazdani, Detecting and distinguishing
Majorana zero modes with the scanning tunnelling microscope.
Nat. Rev. Phys. 3, 541–554 (2021). doi: 10.1038/s42254-021-00328-z
117. M. Chen et al., Discrete energy levels of Caroli-de Gennes-
Matricon states in quantum limit in FeTe0.55Se0.45.
Nat. Commun. 9, 970 (2018). doi: 10.1038/s41467-018-
03404-8; pmid: 29511191
118. S. Zhu et al., Nearly quantized conductance plateau of
vortex zero mode in an iron-based superconductor. Science
367, 189–192 (2020). doi: 10.1126/science.aax0274;
pmid: 31831637
119. C. Chen et al., Quantized conductance of Majorana zero
mode in the vortex of the topological superconductor
(Li0.84Fe0.16)OHFeSe. Chin. Phys. Lett. 36, 057403 (2019).
doi: 10.1088/0256-307X/36/5/057403
120. T. Machida et al., Zero-energy vortex bound state in the
superconducting topological surface state of Fe(Se,Te).
Nat. Mater. 18, 811–815 (2019). doi: 10.1038/s41563-019-
0397-1; pmid: 31209388
121. B. S. de Mendonça, A. L. R. Manesco, N. Sandler,
L. G. G. V. D. da Silva, Near zero-energy Caroli-de Gennes-
Matricon vortex states in the presence of impurities.
arXiv:2204.05078 [cond.mat-supr-con] (2022).
122. B. Jäck et al., Observation of a Majorana zero mode in a
topologically protected edge channel. Science 364, 1255–1259
(2019). doi: 10.1126/science.aax1444; pmid: 31196882
123. H. Ren et al., Topological superconductivity in a phase-
controlled Josephson junction. Nature 569, 93–98 (2019).
doi: 10.1038/s41586-019-1148-9; pmid: 31019296
124. A. Fornieri et al., Evidence of topological superconductivity in
planar Josephson junctions. Nature 569, 89–92 (2019).
doi: 10.1038/s41586-019-1068-8; pmid: 31019303
125. A. C. Potter, P. A. Lee, Topological superconductivity and
Majorana fermions in metallic surface states. Phys. Rev. B 85,
094516 (2012). doi: 10.1103/PhysRevB.85.094516
126. S. Manna et al., Signature of a pair of Majorana zero modes in
superconducting gold surface states. Proc. Natl. Acad.
Sci. U.S.A. 117, 8775–8782 (2020). doi: 10.1073/
pnas.1919753117; pmid: 32253317
127. J. I. A. Li et al., Even-denominator fractional quantum Hall
states in bilayer graphene. Science 358, 648–652 (2017).
doi: 10.1126/science.aao2521; pmid: 28982799
Yazdani et al., Science 380, eade0850 (2023) 23 June 2023
128. A. A. Zibrov et al., Tunable interacting composite fermion
phases in a half-filled bilayer-graphene Landau level.
Nature 549, 360–364 (2017). doi: 10.1038/nature23893;
pmid: 28933427
129. J. Falson et al., A cascade of phase transitions in an orbitally
mixed half-filled Landau level. Sci. Adv. 4, eaat8742 (2018).
doi: 10.1126/sciadv.aat8742; pmid: 30225370
130. D.-K. Ki, V. I. Fal’ko, D. A. Abanin, A. F. Morpurgo, Observation
of even denominator fractional quantum Hall effect in
suspended bilayer graphene. Nano Lett. 14, 2135–2139
(2014). doi: 10.1021/nl5003922; pmid: 24611523
131. Y. Kim et al., Fractional quantum Hall states in bilayer graphene
probed by transconductance fluctuations. Nano Lett. 15,
7445–7451 (2015). doi: 10.1021/acs.nanolett.5b02876;
pmid: 26479836
132. X. Liu et al., Visualizing broken symmetry and topological
defects in a quantum Hall ferromagnet. Science 375,
321–326 (2022). doi: 10.1126/science.abm3770;
pmid: 34855512
133. Z. Papić, R. S. K. Mong, A. Yazdani, M. P. Zaletel, Imaging
anyons with scanning tunneling microscopy. Phys. Rev. X 8,
011037 (2018). doi: 10.1103/PhysRevX.8.011037
134. K. T. Law, Probing non-abelian statistics in n = 12/5 quantum
Hall state. Phys. Rev. B 77, 205310 (2008). doi: 10.1103/
PhysRevB.77.205310
135. A. Wójs, Transition from Abelian to non-Abelian quantum
liquids in the second Landau level. Phys. Rev. B 80, 041104
(2009). doi: 10.1103/PhysRevB.80.041104
136. E. H. Rezayi, R. Read, Non-Abelian quantized Hall states of
electrons at filling factors 12/5 and 13/5 in the first
excited Landau level. Phys. Rev. B 79, 075306 (2009).
doi: 10.1103/PhysRevB.79.075306
137. P. Bonderson, A. E. Feiguin, G. Möller, J. K. Slingerland,
Competing topological orders in the n = 12/5 quantum Hall
state. Phys. Rev. Lett. 108, 036806 (2012). doi: 10.1103/
PhysRevLett.108.036806; pmid: 22400774
138. R. S. K. Mong, M. P. Zaletel, F. Pollmann, Z. Papić, Fibonacci
anyons and charge density order in the 12/5 and 13/5
quantum Hall plateaus. Phys. Rev. B 95, 115136 (2017).
doi: 10.1103/PhysRevB.95.115136
139. K. Pakrouski, M. Troyer, Y.-L. Wu, S. Das Sarma,
M. R. Peterson, Enigmatic 12/5 fractional quantum Hall
effect. Phys. Rev. B 94, 075108 (2016). doi: 10.1103/
PhysRevB.94.075108
140. W. Zhu, S. S. Gong, F. D. M. Haldane, D. N. Sheng, Fractional
quantum Hall states at n = 13/5 and 12/5 and their non-
Abelian nature. Phys. Rev. Lett. 115, 126805 (2015).
doi: 10.1103/PhysRevLett.115.126805; pmid: 26431006
141. N. H. Lindner, E. Berg, G. Refael, A. Stern, Fractionalizing
Majorana fermions: Non-Abelian statistics on the edges of
Abelian quantum Hall states. Phys. Rev. X 2, 041002 (2012).
doi: 10.1103/PhysRevX.2.041002
142. D. J. Clarke, J. Alicea, K. Shtengel, Exotic non-Abelian
anyons from conventional fractional quantum Hall states.
Nat. Commun. 4, 1348 (2013). doi: 10.1038/ncomms2340;
pmid: 23299898
143. R. S. K. Mong et al., Universal topological quantum
computation from a superconductor-Abelian quantum Hall
heterostructure. Phys. Rev. X 4, 011036 (2014). doi: 10.1103/
PhysRevX.4.011036
144. G.-H. Lee et al., Inducing superconducting correlation in
quantum Hall edge states. Nat. Phys. 13, 693–698 (2017).
doi: 10.1038/nphys4084
145. O. Gül et al., Andreev reflection in the fractional quantum Hall
state. Phys. Rev. X 12, 021057 (2022). doi: 10.1103/
PhysRevX.12.021057
146. J. Alicea, P. Fendley, Topological phases with parafermions:
Theory and blueprints. Annu. Rev. Condens. Matter Phys. 7,
119–139 (2016). doi: 10.1146/annurev-conmatphys-031115-011336
147. A. Romito, J. Alicea, G. Refael, F. von Oppen, Manipulating
Majorana fermions using supercurrents. Phys. Rev. B 85,
020502 (2012). doi: 10.1103/PhysRevB.85.020502
148. O. Lesser, A. Saydjari, M. Wesson, A. Yacoby, Y. Oreg,
Phase-induced topological superconductivity in a planar
heterostructure. Proc. Natl. Acad. Sci. U.S.A. 118,
e2107377118 (2021). doi: 10.1073/pnas.2107377118;
pmid: 34183403
149. A. Melo, S. Rubbert, A. R. Akhmerov, Supercurrent-induced
Majorana bound states in a planar geometry. SciPost Phys. 7,
39 (2019). doi: 10.21468/SciPostPhys.7.3.039
150. O. Lesser, K. Flensberg, F. von Oppen, Y. Oreg, Three-phase
Majorana zero modes at tiny magnetic fields. Phys. Rev. B
103, L121116 (2021). doi: 10.1103/PhysRevB.103.L121116
151. O. Lesser, Y. Oreg, A. Stern, One-dimensional topological
superconductivity based entirely on phase control. Phys. Rev.
B 106, L241405 (2022). doi: 10.1103/PhysRevB.106.L241405
152. A. Haim, A. Stern, Benefits of weak disorder in one-dimensional
topological superconductors. Phys. Rev. Lett. 122, 126801
(2019). doi: 10.1103/PhysRevLett.122.126801; pmid: 30978105
153. T. Laeven, B. Nijholt, M. Wimmer, A. R. Akhmerov, Enhanced
proximity effect in zigzag-shaped Majorana Josephson
junctions. Phys. Rev. Lett. 125, 086802 (2020). doi: 10.1103/
PhysRevLett.125.086802; pmid: 32909794
154. A. Haim, Y. Oreg, Time-reversal-invariant topological
superconductivity in one and two dimensions. Phys. Rep.
825, 1–48 (2019). doi: 10.1016/j.physrep.2019.08.002
155. D. I. Pikulin, M. Franz, Black hole on a chip: Proposal for a
physical realization of the Sachdev-Ye-Kitaev model in a
solid-state system. Phys. Rev. X 7, 031006 (2017).
doi: 10.1103/PhysRevX.7.031006
156. A. Chew, A. Essin, J. Alicea, Approximating the Sachdev-Ye-Kitaev
model with Majorana wires. Phys. Rev. B 96, 121119 (2017).
doi: 10.1103/PhysRevB.96.121119
157. I. Seroussi, E. Berg, Y. Oreg, Topological superconducting
phases of weakly coupled quantum wires. Phys. Rev. B 89,
104523 (2014). doi: 10.1103/PhysRevB.89.104523
158. S. Vijay, T. H. Hsieh, L. Fu, Majorana fermion surface code for
universal quantum computation. Phys. Rev. X 5, 041038
(2015). doi: 10.1103/PhysRevX.5.041038
159. L. A. Landau et al., Towards realistic implementations of a
p
Majorana surface code. Phys. Rev. Lett. 116, 050501 (2016).
doi: 10.1103/PhysRevLett.116.050501; pmid: 26894694
160. V. Fatemi et al., Electrically tunable low-density
superconductivity in a monolayer topological insulator.
Science 362, 926–929 (2018). doi: 10.1126/science.aar4642;
pmid: 30361384
161. Y. Cao et al., Unconventional superconductivity in magic-
angle graphene superlattices. Nature 556, 43–50 (2018).
doi: 10.1038/nature26160; pmid: 29512651
g
162. Z. Hao et al., Electric field-tunable superconductivity in
alternating-twist magic-angle trilayer graphene. Science 371,
1133–1138 (2021). doi: 10.1126/science.abg0399;
pmid: 33542148
163. Y. Zhang et al., Promotion of superconductivity in magic-
y
angle graphene multilayers. Science 377, 1538–1543 (2022).
doi: 10.1126/science.abn8585; pmid: 36173835
164. J. M. Park et al., Robust superconductivity in magic-angle
multilayer graphene family. Nat. Mater. 21, 877–883 (2022).
doi: 10.1038/s41563-022-01287-1; pmid: 35798945
165. J. O. Island et al., Spin-orbit-driven band inversion in bilayer
graphene by the van der Waals proximity effect. Nature 571,
85–89 (2019). doi: 10.1038/s41586-019-1304-2; pmid: 31189959
166. K. P. Nuckolls et al., Strongly correlated Chern insulators in
magic-angle twisted bilayer graphene. Nature 588, 610–615
(2020). doi: 10.1038/s41586-020-3028-8; pmid: 33318688
167. Y. Xie et al., Fractional Chern insulators in magic-angle
twisted bilayer graphene. Nature 600, 439–443 (2021).
y g
doi: 10.1038/s41586-021-04002-3; pmid: 34912084
168. C.-Z. Chang et al., High-precision realization of robust
quantum anomalous Hall state in a hard ferromagnetic
topological insulator. Nat. Mater. 14, 473–477 (2015).
doi: 10.1038/nmat4204; pmid: 25730394
169. A. J. Bestwick et al., Precise quantization of the anomalous Hall
effect near zero magnetic field. Phys. Rev. Lett. 114, 187201
(2015). doi: 10.1103/PhysRevLett.114.187201; pmid: 26001016
,
170. B. Lian, X.-Q. Sun, A. Vaezi, X.-L. Qi, S.-C. Zhang, Topological
quantum computation based on chiral Majorana fermions.
Proc. Natl. Acad. Sci. U.S.A. 115, 10938–10942 (2018).
doi: 10.1073/pnas.1810003115; pmid: 30297431
171. Q. L. He et al., RETRACTED: Chiral Majorana fermion modes
in a quantum anomalous Hall insulator-superconductor
structure. Science 357, 294–299 (2017). doi: 10.1126/
science.aag2792; pmid: 28729508
172. H. Thorp, Editorial retraction. Science 378, 718 (2022).
doi: 10.1126/science.adf7575
173. M. Kayyalha et al., Absence of evidence for chiral Majorana
modes in quantum anomalous Hall-superconductor devices.
Science 367, 64–67 (2020). doi: 10.1126/science.aax6361;
pmid: 31896711
174. A. Kitaev, Anyons in an exactly solved model and beyond.
Ann. Phys. 321, 2–111 (2006). doi: 10.1016/j.aop.2005.10.005
175. A. Thomson, A. Pientka, Simulating spin systems with Majorana
networks. arXiv:1807.09291 [cond.mat-mes-hall] (2018).
176. E. Sagi, H. Ebisu, Y. Tanaka, A. Stern, Y. Oreg, Spin liquids
from Majorana zero modes in a Cooper-pair box. Phys. Rev. B
99, 075107 (2019). doi: 10.1103/PhysRevB.99.075107
11 of 12
RES EARCH | R E V I E W
177. G. Jackeli, G. Khaliullin, Mott insulators in the strong spin-
orbit coupling limit: From Heisenberg to a quantum
compass and Kitaev models. Phys. Rev. Lett. 102,
017205 (2009). doi: 10.1103/PhysRevLett.102.017205;
pmid: 19257237
178. S. Trebst, C. Hickey, Kitaev materials. Phys. Rep. 950, 1–37
(2022). doi: 10.1016/j.physrep.2021.11.003
179. Y. Kasahara et al., Majorana quantization and half-integer
thermal quantum Hall effect in a Kitaev spin liquid. Nature
559, 227–231 (2018). doi: 10.1038/s41586-018-0274-0;
pmid: 29995863
180. Y. Vinkler-Aviv, A. Rosch, Approximately quantized
thermal Hall effect of chiral liquids coupled to phonons.
Phys. Rev. X 8, 031032 (2018). doi: 10.1103/
PhysRevX.8.031032
181. M. Ye, G. B. Halász, L. Savary, L. Balents, Quantization of the
thermal Hall conductivity at small Hall angles. Phys. Rev. Lett.
121, 147201 (2018). doi: 10.1103/PhysRevLett.121.147201;
pmid: 30339425
182. D. Aasen, R. S. K. Mong, B. M. Hunt, D. Mandrus, J. Alicea,
Electrical probes of the non-Abelian spin liquid in Kitaev
Yazdani et al., Science 380, eade0850 (2023) 23 June 2023
materials. Phys. Rev. X 10, 031014 (2020). doi: 10.1103/
PhysRevX.10.031014
183. X. Mi et al., Noise-resilient edge modes on a chain of
superconducting qubits. Science 378, 785–790 (2022).
doi: 10.1126/science.abq5769; pmid: 36395220
184. J. F. Steiner, F. von Oppen, Readout of Majorana qubits.
Phys. Rev. Res. 2, 033255 (2020). doi: 10.1103/
PhysRevResearch.2.033255
185. D. Wong et al., Cascade of electronic transitions in
magic-angle twisted bilayer graphene. Nature 582,
198–202 (2020). doi: 10.1038/s41586-020-2339-0;
pmid: 32528095
ACKN OWLED GMEN TS
We thank all our collaborators on this subject. Funding: A.Yaz.
acknowledges funding from the Gordon and Betty Moore
Foundation’s Emergent Phenomena in Quantum Systems (EPiQS)
Initiative grant GBMF9469, Office of Naval Research grant N00014-
21-1-2592, National Science Foundation (NSF) Materials Research
Science and Engineering Centers (MRSEC) through the Princeton
Center for Complex Materials grant NSF-DMR-2011750, the US
Army Research Office (ARO) Multidisciplinary University Research
Initiatives (MURI) project under grant no. W911NF-21-2-0147, and
NSF grant DMR-1904442. B.I.H. acknowledges funding from the
Microsoft Corporation and NSF grant DMR-1231319. F.v.O. is
grateful for funding by the Deutsche Forschungsgemeinschaft
(CRC 183 and SFB 910) as well as the Einstein Research Unit on
Quantum Devices. A.Yac. is supported by the Quantum Science
Center (QSC), a National Quantum Information Science Research
Center of the US Department of Energy. A.Yac. is also partly
supported by the Gordon and Betty Moore Foundation through
grant GBMF 9468, by the ARO MURI project under grant no.
W911NF-21-2-0147, and by the STC Center for Integrated Quantum
Materials NSF grant no. DMR-1231319. Competing interests: The
authors have no competing interests. License information:
Copyright © 2023 the authors, some rights reserved; exclusive
licensee American Association for the Advancement of Science. No
claim to original US government works. https://www.science.org/
about/science-licenses-journal-article-reuse
Submitted 25 July 2022; accepted 22 May 2023
10.1126/science.ade0850
p
g
y
y g
,
12 of 12
RES EARCH
RESEARCH ARTICLE SUMMARY
MOLECULAR BIOLOGY
WAPL functions as a rheostat of Protocadherin
isoform diversity that controls neural wiring
Lea Kiefer†, Anna Chiosso†, Jennifer Langen†, Alex Buckley†, Simon Gaudin, Sandy M. Rajkumar‡,
Gabrielle Isabelle F. Servito‡, Elizabeth S. Cha, Akshara Vijay, Albert Yeung, Adan Horta,
Michael H. Mui, Daniele Canzio*
INTRODUCTION: The establishment of neural
connectivity patterns requires the ability of
individual neurons to distinguish self from
nonself. In mammals, clustered Protocadherin
(Pcdh) genes encode cell surface molecular
“identifiers” (i.e., barcodes) that allow neural
“self/nonself” discrimination: Neurites from
the same cell carrying identical Pcdh barcodes
recognize and repel each other, whereas neu-
rites from different cells carrying distinct Pcdh
barcodes do not. In mice, there are 116 Pcdh
genes, 58 on each of the two homologous chro-
mosomes, organized into three tandemly arranged
clusters (a, b, and g). Different neural types ex-
press distinct repertoires of Pcdh genes to instruct
their wiring processes. The most noteworthy
examples of this behavior are the deterministic
expression of a single Pcdh gene in serotonergic
neurons (5-HTs) and the stochastic expression
of a few Pcdh genes in olfactory sensory neurons
(OSNs). Deterministic expression of Pcdhac2
provides 5-HTs with a single, shared barcode.
Using this mechanism, neurites from the same
cell not only recognize and repel self, but also
neurites from other 5-HTs, thus favoring non-
overlapping tiled projections of their axons
throughout the brain. By contrast, stochastic
WAPL functions as a rheostat of cohesin processivity and Pcdh isoform diversity. In 5-HTs, high WAPL
limits cohesin translocation, which favors Pcdhac2 expression and axon tiling. Loss of WAPL results in 5-HT tiling
defects. In OSNs, low WAPL enables cohesin translocation, which favors Pcdh isoform diversity and axon convergence.
Loss of Rad21 (a cohesin subunit) results in OSN convergence defects. Only the Pcdha gene cluster is shown.
Kiefer et al., Science 380, 1236 (2023) 23 June 2023
◥ cohesin activity could be a potential mechanism
and combinatorial expression of distinct rep-
ertoires of Pcdh genes provides each OSN with
a unique barcode. The generation of this barcod-
ing diversity enables convergence of OSN axons
into tightly packed, overlapping structures. These
observations pose a fundamental question: How
do neurons choose the correct number (and type)
of Pcdh genes to support their wiring needs?
RATIONALE: The genomic architecture of Pcdh
genes poses a fundamental regulatory chal-
lenge: cis-regulatory mechanisms must over-
come genomic-proximity biases imposed by the
linear arrangement of Pcdh promoters relative
to distal enhancers. These mechanisms must
be regulated differently in distinct cell types.
For instance, in 5-HTs, choice of the Pcdhac2
gene is genomic-distance biased, because its
promoter is the most enhancer-proximal pro-
moter in the Pcdha cluster. However, in OSNs,
genomic-distance biases in promoter choice
are erased in favor of random selection. Pre-
vious studies have suggested a role for the DNA
translocase cohesin in regulating Pcdh ex-
pression by mediating enhancer-promoter in-
teractions. Building upon these studies, we
hypothesized that differential regulation of
by which neurons achieve the Pcdh expression
programs that support their wiring patterns.
RESULTS: By genetically targeting components
of the cohesin complex in 5-HTs and OSNs, we
revealed that neural type–specific Pcdh expres-
sion and axonal behavior depend on the activ-
ity of cohesin and its unloader, WAPL (wings
apart-like protein homolog). Given the linear
arrangement of Pcdh genes, high WAPL in 5-
HTs limits cohesin translocation, thereby fa-
voring the expression of the enhancer-proximal
Pcdhac2 isoform in all cells and ultimately con-
straining axon arrangements to a tiling pat-
tern. Conditional deletion of WAPL in 5-HTs
resulted in loss of Pcdhac2 expression and dis-
ruption of axon tiling. By contrast, we found that
low WAPL in OSNs enables cohesin transloca-
tion along the locus. Using this mechanism,
cohesin increases the probability of contacts
p
between the enhancers and the more distal
Pcdh promoters, thereby enabling stochastic
expression of a larger repertoire of Pcdh iso-
forms and ultimately driving axon convergence
and assembly of olfactory circuits. In OSNs, con-
ditional ablation of Rad21, an essential subunit
g
of the cohesin complex, resulted in loss of Pcdh
isoform diversity and gain of the biased expres-
sion of the enhancer-proximal Pcdhac2 gene.
Thus, Rad21 ablation turned the Pcdh expres-
sion profile of OSNs into one reminiscent of
y
5HTs. Loss of Pcdh diversity by Rad21 deletion
resulted in disruption of OSN axon convergence.
Therefore, by countering enhancer-promoter
genomic proximity biases, neural type–specific
regulation of cohesin activity by WAPL tunes
Pcdh isoform diversity and enables distinct
modes of neuronal wiring and circuit assembly.
CONCLUSION: We propose a model in which
WAPL functions as a rheostat of cohesin pro-
y g
cessivity on DNA to enable structural and tran-
scriptional modularity of the Pcdh gene cluster
required to establish different patterns of neu-
ral connectivity during brain development. More
broadly, our data suggest a new class of regu-
,
latory principles for genes organized in clusters
in which cells use rheostat-operating logics
to overcome distance biases in gene selection.
We speculate that rheostats are solutions to the
challenges imposed by the complex architec-
tures of these gene clusters that generate the
transcriptional diversity critical for cellular
and functional diversification.
▪
The list of author affiliations is available in the full article online.
*Corresponding author. Email: daniele.canzio@ucsf.edu
†These authors contributed equally to this work.
‡These authors contributed equally to this work.
Cite this article as L. Kiefer et al., Science 380, eadf8440
(2023). DOI: 10.1126/science.adf8440
READ THE FULL ARTICLE AT
https://doi.org/10.1126/science.adf8440
1 of 1
RES EARCH

◥ three clusters (22, 23), which allows individ-

RESEARCH ARTICLE ual OSNs to display different barcodes on
their cell surface (18). Overriding Pcdh diver-
MOLECULAR BIOLOGY sity by expressing high levels of specific Pcdh

WAPL functions as a rheostat of Protocadherin

isoforms in all OSNs leads to axon conver-
gence defects and disruption of glomeruli

isoform diversity that controls neural wiring

formation in the olfactory bulb (OB) (18).
The establishment of these two distinct mech-
anisms of neural wiring, tiling and convergence,
Lea Kiefer1,2†, Anna Chiosso1,2†, Jennifer Langen1,2,3†, Alex Buckley1,2,3†, Simon Gaudin1,2,4, by clustered Pcdh genes represents an example
Sandy M. Rajkumar1,2‡, Gabrielle Isabelle F. Servito1,2‡, Elizabeth S. Cha1,2, Akshara Vijay1,2, of how cells must be able to commit orthogo-
Albert Yeung5, Adan Horta6, Michael H. Mui1,2, Daniele Canzio1,2,7* nal gene regulatory logics. They must either
promote genomic distance biases in favor of
Neural type–specific expression of clustered Protocadherin (Pcdh) proteins is essential for the deterministic promoter selection to acquire
establishment of connectivity patterns during brain development. In mammals, deterministic expression the cellular uniformity required to tile, or they
of the same Pcdh isoform promotes minimal overlap of tiled projections of serotonergic neuron axons must overcome such biases in favor of sto-
throughout the brain, while stochastic expression of Pcdh genes allows for convergence of tightly chasticity to generate the cellular diversity
packed, overlapping olfactory sensory neuron axons into targeted structures. How can the same gene required to converge. The regulation of these
locus generate opposite transcriptional programs that orchestrate distinct spatial arrangements of contrasting gene expression processes poses
axonal patterns? Here, we reveal that cell type–specific Pcdh expression and axonal behavior depend on a fundamental question: How can both deter-
the activity of cohesin and its unloader, WAPL (wings apart–like protein homolog). While cohesin ministic and stochastic promoter selection

p
erases genomic-distance biases in Pcdh choice, WAPL functions as a rheostat of cohesin processivity arise from the same gene cluster?
that determines Pcdh isoform diversity.
Results
OSN progenitor cells display developmental

T
he generation of cellular diversity is vital (5, 6) (Fig. 1A). Each cluster is regulated by biases in Pcdhac2 promoter choice
to the many functions of higher organ- cluster-specific transcriptional enhancers lo- To investigate the process underlying the reg-

isms, from the detection of microbes by
the immune system to the ability to per-
ceive and interpret the outside world
by the central nervous system. Stochastic ex-
pression of genes organized in clusters, from
cated downstream (3′) of the promoters, with
distances between enhancers and promoters
that can reach up to 600,000 base pairs (7, 8)
(Fig. 1A: HS5-1, HS18-22, and HS16-17 are the
transcriptional enhancers of the Pcdha, Pcdhb,
g
ulation of neural type–specific Pcdh expression,
we first focused on studying the mechanism of
stochastic Pcdh promoter choice during the
maturation of OSNs. The mouse olfactory epi-
thelium (OE) is a pseudostratified tissue com-

immunoglobulin to the olfactory receptor to
clustered Protocadherin (Pcdh) genes, repre-
sents an extraordinary example of a mechanism
by which cells greatly expand their proteome
repertoire to achieve unparalleled diversifica-
tion (1–3). There are different challenges that
these systems must overcome to promote ran-
dom gene choice. While olfactory receptor
gene choice relies on interchromosomal con-
tacts (2, 4), the genomic architectures of the
and Pcdhg clusters, respectively).
Clustered Pcdh genes encode cell surface
proteins that “barcode” individual neurons
with unique identities (3, 9, 10). This mecha-
nism allows neurons to self-recognize and self-
avoid as they establish proper connections
with neighboring cells during brain develop-
ment (3, 9, 10). There are two broad classes of
neuronal contact patterns: tiling and conver-
gence (11–13). In mammals, neural tiling is
y
posed of distinct cell types that represent
different neurodevelopmental stages of OSNs,
all of which have finite life spans and are con-
tinuously regenerated from mitotically active
stem cells within the epithelium. To determine
the precise developmental stage at which Pcdh
genes are chosen by individual cells, we per-
formed single-cell RNA sequencing (RNA-seq)
from an isolated wild-type (WT) OE. Using es-
tablished genetic markers for the distinct cell

immunoglobulin and Pcdh loci call for cis-
regulatory mechanisms that overcome ge-
nomic distance biases imposed by the linear
arrangement of these genes relative to their
cis-regulatory elements.
epitomized by serotonergic neurons (5-HTs)
whose axons must minimize overlap with
their neighboring 5-HTs to innervate various
brain regions (14–16). Tiling of 5-HT axons is
achieved by the deterministic expression of only
y g
types of the OE, we observed developmentally
regulated Pcdh gene choice (Fig. 1B). Con-
sistent with previous observations (24), no Pcdh
expression was observed in the intracellular ad-
hesion protein 1–positive (Icam1+) cells (hori-

In mice, there are a total of 116 clustered
Pcdh genes (120 in humans), 58 on each of the
two homologous chromosomes, organized
into three tandemly arranged gene clusters
(14 Pcdha, 22 Pcdhb, and 22 Pcdhg) spanning
nearly 1 million base pairs of genomic DNA
1
Weill Institute for Neurosciences, University of California,
San Francisco, San Francisco, CA 94158, USA. 2Department of
Neurology, University of California, San Francisco, San
Francisco, CA 94158, USA. 3Neuroscience Graduate Program,
University of California, San Francisco, San Francisco, CA
94158, USA. 4Ecole Normale Superieure de Lyon, 69432 Lyon,
France. 5Department of Molecular and Cell Biology, University
of California, Berkeley, Berkeley, CA 94720, USA. 6Pura Vida
Investments, New York, NY 10106, USA. 7Chan-Zuckerberg
Biohub, San Francisco, CA 94158, USA.
*Corresponding author. Email: daniele.canzio@ucsf.edu
†These authors contributed equally to this work.
‡These authors contributed equally to this work.
Pcdhac2, the most HS5-1 enhancer–proximal
Pcdha gene, which conveys an identical Pcdh
barcode to the cell surfaces of this class of
neurons (14, 16). Deletion of Pcdhac2 results
in a marked tiling defect of 5-HT projections
(14, 16), a self-avoidance phenotype (i.e., neurites
clumping together) similar to that previously
observed in starburst amacrine cells upon ge-
netic removal of the Pcdhg genes (17). Con-
vergence is instead exemplified by olfactory
sensory neurons (OSNs) whose axons express-
ing the same olfactory receptor (OR), but dis-
tinct repertoires of clustered Pcdh genes, must
coalesce into tightly packed structures known
as glomeruli (18–21). These distinct Pcdh reper-
toires are generated by stochastic and com-
binatorial Pcdh promoter choice from all
,
zontal basal stem cells, HBCs), which represent
the quiescent stem cell population of the OE,
whereas all Pcdha isoforms were stochasti-
cally expressed in postmitotic OSNs, positive
for the olfactory marker protein (Omp) and
negative for the growth-associated protein 43
(Gap43) (mature olfactory sensory neurons,
mOSNs) (Fig. 1B). However, we observed a tran-
scriptional bias toward choice of the Pcdhac2
promoter over the rest of the cluster, reminis-
cent of the deterministic expression of Pcdhac2
in 5-HTs. This bias was first observed in the
globose basal cells, continued into early OSN
precursors, and disappeared in mOSNs (Fig.
1B). Down-regulation of Pcdhac2 in mOSNs
coincided with the onset of stochastic expres-
sion of the rest of the Pcdha promoters (Fig. 1B).

Kiefer et al., Science 380, eadf8440 (2023) 23 June 2023 1 of 11

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

Fig. 1. Cohesin erases genomic A 600kb

distance biases in Pcdh 300kb 200kb
promoter choice in mature Pcdhα Pcdhβ Pcdhγ
OSNs. (A) Genomic architecture
α1 α12 αc1 αc2 β1 β22 γa1 γa12

HS5-1

HS16-17
HS18-22
of the mouse clustered Pcdh
gene locus. Pcdha, Pcdhb, and
Pcdhg variable 5′ exons are variable c-type constant variable c-type constant
exons exons exons exons exons exons
shown in white, red, and blue,
respectively. The variable exons B Olfactory epithelium (OE) C OSN-WT
of the Pcdha and Pcdhg gene HBC GBC INP iOSN mOSN
OSN-RAD21-cKO
(Icam+) (Ascl1+) (Tex15+) (Gap43+) (Omp+, Gap43-)
clusters are further subdivided 80
into alternate and c-types, α

% OSNs
60
the latter of which includes the OSN maturation
40
20
Pcdhac2 gene (orange). Down- UMI count
0 15 30
0
stream of the Pcdhac2 and Mean Omp
Pcdhgc5 exons, there are three
# of cells expressing
Pcdhα isoforms

400 80
small constant exons that splice

% OSNs
60
300
to the chosen variable exons. 40
200 20
Cluster-specific transcriptional
100 0
enhancers (HS5-1, HS16-17, and OSN-WT
0
HS18-22) are shown in gray. 1 5 10 αc2 1 5 10 αc2 1 5 10 αc2 1 5 10 αc2 1 5 10 αc2

p
# of cells expressing
(B) Pcdha expression during 80
1500 50
γ

% OSNs
Pcdhα isoforms
40 60
OSN maturation in OSN-WT and 30 40
1000
20
OSN-RAD21-cKO mice. The Omp cre 10 20
distinct cell types that result in loxP loxP 500 0 0
1 8 αc2
OSN-RAD21-cKO Rad21 0 2 4 6
mOSNs are schematized on top. 0 Number of Pcdh
Heat map shows the Omp 1 5 10 αc2 1 5 10 αc2 isoforms chosen per OSN

g
expression level. UMI, unique D E
molecular identifier. (C) Histo- OSN-WT OSN-RAD21-cKO
gram of the number of Pcdh Olfr1264 Olfr1264
Olfactory Olfactory
isoforms chosen for expression α7-β20-γb1
Epithelium (OE) Bulb (OB)
per mOSN in OSN-WT and OSN-

y
β1-β4-β17
RAD21-cKO. Data represent one α3-β1-γb6
single-cell RNA-seq replicate per
α β γ α β γ
condition. (D) Left: Subset of
Olfr167 Olfr167
Pcdh expression profiles from WT
α2-β1-γb6
OSNs expressing either the OR
Olfr1264 or Olfr167. Each row α3-γb2-γa11
Glomeruli
indicates a single cell, each column α9-β1 OSNs
a Pcdh isoform from all three α β γ α β γ
clusters. Right: Schematic of αc2 αc2
16µm 16µm
OSN axons from the OE anterior posterior

y g
projecting to the OB, where OSNs F G H m71
DAPI
m71
DAPI
m71
DAPI
expressing the same OR but
OSN-RAD21-cKO OSN-RAD21-het

VGLUT2 VGLUT2
distinct Pcdhs converge into DAPI DAPI
OSN-WT
glomeruli. (E) As in (D) but from OSN-RAD21-cKO anterior posterior
OSN-RAD21-cKO animals. ****
% unique barcodes

,
(F) Pcdh diversity across all
per “glomerulus”

100
OSNs expressing the same OR 80
(OSNs converging into the same 60 m71 Glomerulus
“glomerulus”). (G) IHC against 40 m71-expressing OSNs
VGLUT2 in coronal sections of OSN- OSN-
RAD21-het RAD21-cKO
the OB from OSN-RAD21-het
fl/+
[Omp(iresCre/+);Rad21 ;
tdTomatofl/+] and OSN-RAD21-cKO [Omp(iresCre/+);Rad21fl/fl;tdTomatofl/+]. Green, VGLUT2; blue, DAPI. Dotted white circle is an example of a glomerulus. Scale bar,
100 mm. (H) Left: Schematic of the coronal sections of the OB. Right: IHC against OR m71 in coronal sections from OSN-RAD21-het and OSN-RAD21-cKO. Green, m71; blue,
DAPI. Scale bar, 50 mm.

Cohesin erases genomic distance biases in Pcdh mammalian three-dimensional (3D) folding of (CTCF) protein that are bound by CTCF (29–32).
promoter choice in mature OSNs the genome through a mechanism known as Within the context of the Pcdha gene locus,
Recent studies have implicated the cohesin DNA loop extrusion. By this mechanism, cohesin cohesin has been shown to mediate long-range
protein complex in regulating the expression of loads onto DNA and extrudes chromatin until engagement between individual promoters
Pcdha genes (24–28). Cohesin is a DNA trans- it stalls at boundary elements, usually bearing and their distal HS5-1 enhancer (24–28). To de-
locase that has been proposed to regulate binding sites for the CCCTC-binding factor termine whether cohesin functions to promote

Kiefer et al., Science 380, eadf8440 (2023) 23 June 2023 2 of 11

RES EARCH | R E S E A R C H A R T I C L E
the transcriptional switch from the determi-
nistic expression of Pcdhac2 to the stochastic
expression of all Pcdhs in mature OSNs, we
conditionally deleted the Rad21 gene. Rad21
encodes the kleisin subunit of cohesin, and its
deletion renders the entire cohesin complex
inactive. Deletion of Rad21 was accomplished
using a Cre driver under the control of the Omp
promoter [Omp(iresCre/+);Rad21fl/fl;tdTomatofl/+,
OSN-RAD21-cKO] (fig. S1A). As shown by our
single-cell RNA-seq data, Omp expression ini-
tiates in Gap43+ immature OSNs and con-
tinues as OSNs mature into Omp+,Gap43–
mOSNs (Fig. 1B). OSNs from WT animals were
isolated by green fluorescent protein (GFP)
using Omp(iresGFP/+) mice, and OSNs from
OSN-RAD21-cKO were isolated using tdTomato
(fig. S1B).
To filter for only mOSNs, we computation-
ally selected Omp+,Gap43– cells. From all cells
identified as mOSNs from WT mice, 95.8% had
a detectable Pcdh barcode composed of, on av-
erage, three Pcdh isoforms (Fig. 1, C and D,
and fig. S1C). The probability of each individ-
ual isoform to be expressed was low, except for
Pcdhb1, which was chosen in 78.4% of mOSNs
in which a Pcdh barcode was detected (Fig. 1,
C and D). Our data further showed that Pcdh
isoform choice was independent from the
choice of ORs, demonstrating that essentially
all individual OSNs bear a unique identity
defined by the expression of both Pcdhs and
ORs (fig. S1D).
Upon conditional deletion of the Rad21 gene,
we observed a loss of Pcdh isoform diversity in
all three Pcdh gene clusters and an increase
in the number of mOSNs without a detectable
Pcdh barcode (OSN-WT, 4.2%; OSN-RAD21-
cKO, 10.4%) (Fig. 1, B to E, and fig. S1C). Al-
though a reduction of promoter choice was
observed across all three clusters, this loss
appeared to follow a genomic distance depen-
dent trend in which the most enhancer-distal
promoters were the most affected (fig. S1E).
Conversely, the frequency of choice of the
Pcdhac2 promoter increased ~3-fold (OSN-
WT, 12.3%; OSN-RAD21-cKO, 38.0%) (Fig. 1, D
and E, and fig. S1E). This genomic distance–
biased expression of Pcdha genes correlated
with a decrease in chromatin contacts be-
tween the HS5-1 enhancer and the distal
promoters (fig. S1F). Finally, contrary to the
rest of the Pcdhb genes, expression of Pcdhb1
did not change (OSN-WT, 79.4%; OSN-RAD21-
cKO, 78.3%), suggesting that the activation of
the Pcdhb1 promoter is cohesin independent
(Fig. 1, D and E).
These data suggest that cohesin favors sto-
chastic Pcdh promoter choice by erasing ge-
nomic distance biases in promoter selection
and that, in the context of the Pcdha gene
cluster, loss of cohesin activity in mOSNs re-
sults in a Pcdh transcriptional profile remi-
niscent of a different class of neurons—5-HTs,
Kiefer et al., Science 380, eadf8440 (2023) 23 June 2023
where Pcdhac2 is the only Pcdha isoform ex-
pressed (14, 16).
Cohesin activity is required for normal assembly
of an olfactory map
OSN axons from the OE project to stereotypic
positions in the OB, where OSNs expressing the
same OR converge to form glomeruli (33, 34)
(Fig. 1D). Pcdh isoform diversity is necessary
for OSN convergence and glomeruli formation
(18). Given that we observed a significant loss of
Pcdh isoform diversity within mOSNs express-
ing the same OR upon Rad21 knockout (Fig. 1F),
we wondered whether Rad21 deletion resulted
in any phenotypic abnormalities of the glomeruli
structures. To investigate this, we compared the
OB of OSN-RAD21-cKO mice with that of hetero-
zygous littermates [Omp(iresCre/+);Rad21fl/+;
tdTomatofl/+, OSN-RAD21-het] in which Pcdh
expression was not altered relative to OSN-WT
(fig. S1G). Unlike OSN-Rad21-het, glomeruli
structures were disrupted throughout the OB
in OSN-RAD21-cKO mice, as assayed by stain-
ing for the vesicular glutamate transporter 2
(VGLUT2) (Fig. 1G). Although we cannot ex-
clude that this effect was caused by additional
changes in gene expression upon Rad21 knock-
out, we note that the expression of OR genes,
as well as of other axon guidance molecules
implicated in OSN axon targeting, were un-
changed (fig. S1, H and I). To further probe
whether OSN axons expressing the same OR
were capable of projecting to their stereo-
typic positions in the OB, we performed im-
munohistochemistry (IHC) against the mouse
OR m71 (OLFR151) and observed that these
OSNs targeted to their predicted positions (Fig.
1H). Consistent with the VGLUT2 staining, m71
glomeruli appeared malformed in OSN-RAD21-
cKO mice relative to their OSN-RAD21-het litter-
mates (Fig. 1H). These data are consistent with
the established paradigm that Pcdh protein
isoform diversity plays a critical role in the as-
sembly of olfactory circuits (18–21) and suggest
that the activity of the cohesin protein complex
generates the Pcdh isoform barcoding diver-
sity required for OSN axons to properly converge
into glomeruli structures.
Taken together, these data suggest a logic
for the regulation of Pcdh promoter choice
whereby cohesin activity determines the cellu-
lar commitment to either a deterministic and
uniform Pcdh barcoding profile (where Pcdh
promoter choice is biased toward the enhancer-
proximal Pcdhac2 promoter) or a diverse
Pcdh barcoding profile (where Pcdh promoter
choice is random and enhancer distance
independent).
Developmental regulation of WAPL suggests a
model for the generation of Pcdh diversity
We next investigated how cohesin activity could
be regulated to drive the developmental switch
from the genomic distance–biased expression
of Pcdhac2 to the stochastic expression of all
Pcdh genes during OSN maturation. Given the
linear arrangement of Pcdh promoters span-
ning almost 1 Mb of DNA and the location of
the transcriptional enhancers (Fig. 1A), we rea-
soned that the distance that cohesin can traverse
along the cluster could determine the prob-
ability of Pcdh promoters to be chosen in in-
dividual neurons by the enhancers (Fig. 2A).
One mechanism known to regulate cohesin
processivity on DNA is the activity of the
cohesin “unloader” protein WAPL (wings apart–
like protein homolog) (35–37). In the context of
the Pcdh gene locus, our model would predict
that high WAPL levels would result in frequent
unloading of the cohesin complex from DNA
and would limit the distance that cohesin can
traverse on DNA, therefore restricting the choice
to the most enhancer-proximal Pcdh promoters
(Fig. 2A). Conversely, low WAPL levels would
increase cohesin processivity and promote
p
long-distance cohesin translocation on DNA,
therefore allowing the enhancer-distal Pcdh
promoters to be chosen.
On the basis of this model, we wondered
whether the distinct cell types within the OE
express different levels of Wapl and Rad21
g
mRNAs. We quantified the mRNA levels of
Wapl and Rad21 in the different cell types
within the OE from our single-cell RNA-seq
studies and bulk RNA-seq studies from sorted
cells (fig. S2, A and B). Although Rad21 mRNA
y
levels appeared mostly constant across the
OE (fig. S2C), we observed a decrease of Wapl
mRNA levels that correlated with the onset of
stochastic Pcdh expression whereby globose
basal cells expressed the highest levels of Wapl
mRNA and mOSNs the lowest (Fig. 2B). We
found a correlation between the levels of Wapl
and the frequency of choice of the Pcdhac2
promoter in individual cells (Fig. 2C). This
correlation agreed with the trajectory of OSN
y g
maturation (Fig. 2C).
To visualize this correlation spatially in the
OE, we performed RNA fluorescence in situ
hybridization (FISH) using probes targeting
the mRNA of Wapl, Pcdhac2, and Omp. We
,
observed that Wapl mRNA levels were high-
est in the cells closer to the basal layer, where
most early OSN precursors reside, and grad-
ually declined toward the apical layer, where
most mOSNs reside. This spatial distribu-
tion of Wapl mRNA within the OE positively
correlated with that of Pcdhac2 mRNA but
negatively correlated with the levels of ex-
pression of Omp mRNA (fig. S2D).
Finally, to determine whether WAPL pro-
tein levels correlated with its mRNA levels and
tissue distribution, we performed IHC in the
OE for WAPL and OMP. Consistent with the
single-cell RNA-seq and RNA-FISH data, WAPL
protein levels decreased as OSNs matured, as
evidenced by the loss of staining in the apical
side of the OE (Fig. 2D). Unlike WAPL, the levels
3 of 11
RES EARCH | R E S E A R C H A R T I C L E

A B C

Fraction of Pcdhαc2 containing barcodes

1.00
cohesin CTCF sites

Wapl mRNA levels (UMI per cell)

Low cohesin processivity 9
0.75
(High WAPL) α c2 GBC
INP
6
promoters HS5-1 0.50
iOSN
High cohesin processivity 3
0.25
(Low WAPL)
α c2 mOSN
0
0.00
HBC GBC INP iOSN mOSN
promoters HS5-1 0.0 0.5 1.0 1.5 2.0
Mean Wapl mRNA levels (UMI)
D
Olfactory epithelium (OE)
HBC GBC INP iOSN mOSN
(Icam+) (Ascl1+) (Tex15+) (Gap43+) (Omp+, Gap43-)

p
OSN maturation

DAPI WAPL OMP WAPL OMP WAPL OMP DAPI

Distance from basal lamina (µm)

OMP WAPL

20 40 60 80

g
OE

50 µm

y
1.0 0.8 0.6 0.4 0.2 0.0

DAPI RAD21 OMP RAD21 OMP RAD21 OMP DAPI OMP RAD21

Distance from basal lamina (µm)

20 40 60 80
OE

50 µm

0
1.0 0.8 0.6 0.4 0.2 0.0
Signal intensity

Fig. 2. WAPL is developmentally regulated during OSN maturation. (A) Model
of WAPL controlling cohesin processivity and thereby defining the probability of
choice of Pcdh promoters. (B) Quantification of Wapl expression levels from single-
cell RNA-seq in distinct cell types of the OE. Dots are individual cells; black line in
each box plot is the mean expression level (UMI). (C) Fraction of cells in the OE
y g
The observed correlation follows the developmental trajectory of OSNs. (D) Top: IHC
against WAPL and OMP in coronal sections from the OE of OSN-WT mice. Green,
WAPL; magenta, OMP; blue, DAPI. Scale bars, 50 mm. Bottom: Same as the top but
showing RAD21 levels across the OE. Green, RAD21. For both WAPL and RAD21,
quantification of protein levels relative to OMP along the axis of the OE represents

,
expressing a Pcdha barcode containing Pcdhac2 versus mean Wapl mRNA levels. the average of two biological replicates.

of the RAD21 protein remained constant through-
out the OE (Fig. 2D).
Low WAPL promotes Pcdh isoform
diversity in OSNs
To determine whether developmental down-
regulation of WAPL in mOSNs promotes Pcdh
diversity, we engineered a transgenic mouse line
in which we overexpressed WAPL in mOSNs
[Omp(tTa);tetO-WAPL-mCherry, OSN-WAPL-XP]
and performed both bulk and single-cell RNA-
seq (Fig. 3A and fig. S3, A to C). Consistent with
our hypothesis, increasing WAPL levels resulted
in the biased expression of Pcdhac2 over the rest
of the Pcdha genes in mOSNs (Fig. 3B), in which along the Pcdh locus, thus favoring the expres-
under WT conditions, Pcdha choice is other- sion of Pcdh promoters located closer, in linear
wise random and unbiased (Fig. 1B). We also genomic sequence, to their respective enhancers.
observed a genomic distance–biased expression If low WAPL levels promote Pcdh isoform
of the Pcdhg and Pcdhb genes (fig. S3D), which diversity in individual neurons, then our model
resulted in a preferential usage of the Pcdh would predict that decreasing WAPL levels be-
clusters that correlated with their genomic sizes low an already low WT level in mOSNs would
(Fig. 3C, Pcdhg > Pcdha > Pcdhb). Consistent increase the ability of cohesin to translocate long
with these changes in Pcdh expression, Pcdh en- distances along DNA, thus increasing the prob-
hancers displayed genomic distance biases in ability of choosing promoters that are located
their contacts with their respective promoters, farther from their respective enhancers rela-
as assayed by in situ chromatin conformation tive to WT.
capture (Hi-C) (Fig. 3D and fig. S3E). These data To test this, we conditionally deleted WAPL
suggest that high WAPL limits cohesin extrusion in OSNs [Omp(iresCre/+);WAPLfl/fl;tdTomatofl/+,

Kiefer et al., Science 380, eadf8440 (2023) 23 June 2023 4 of 11

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

Fig. 3. WAPL overexpression in mature A B

log2 fold change in isoform choice

OSNs results in distance-biased Pcdh

(OSN-WAPL−XP/OSN-WT)
expression. (A) Schematic of the transgenic 1
mouse line used to overexpress WAPL in mOSNs Omp tTA
0
(OSN-WAPL-XP) and its mechanistic prediction.
α10 α11 αc2
(B) Change in Pcdh expression of Pcdha genes OSN-WAPL-XP tetO Wapl −1 α6 α8 αc1
between OSN-WT and OSN-WAPL-XP as a α3 α5 α9 α12
−2 α2 α7
function of the relative distance of their α4
promoters to the HS5-1 enhancer. A total of cohesin −3 α1
seven and eight biological replicates of OSN-WT 300 200 100 0
and OSN-WAPL-XP were used. (C) Relative Distance from HS5−1 enhancer (kb)
usage of Pcdh gene clusters between OSN-
C D

log2 fold change in contact probability

WAPL-XP and OSN-WT. The distance between 1.4

Relative usage of Pcdh clusters

α8

(OSN-WAPL-XP/OSN-WT)
the cluster-specific enhancer to the most 1 γa8

(OSN-WAPL-XP/OSN-WT)
γa1 γa7 γa9
distal promoter of that same cluster is shown γa3 γb5 γa10
1.3
in parentheses. (D) Change in contact proba- β6 β15 β22 α3 α9 γb6 γa2
β1 α11γa6 αc2
0 α6 γb4
β16 β17 α2 γa11γc5
bility determined by in situ Hi-C between Pcdh β2 β10 α12γa5 γa4 γb7
1.2 β5 β21 α4α10
promoters and their respective enhancers as a β11 β19 α1 γb1 γa12 γc4
β3 β20 α5 γb2
function of the genomic distance between −1 β13 β14 α7 αc1 γc3
1.1 β7 β12
enhancer-promoter pairs. β18
β4
−2

p
1.0 β8 β9
β α γ
600 400 200 0
(600kb)(300kb)(200kb)
Distance from respective enhancer (kb)

OSN-WAPL-cKO] (Fig. 4A and fig. S4, A and B). (up to 200,000 bp away), were the least af- our RNA-seq data from FACS-isolated mOSNs
Consistent with the role of WAPL in regulat- fected, as assayed by bulk and single-cell RNA- and found that Wapl expression was ~5-fold

ing cohesin processivity (35–37), we observed
that cohesin enabled a longer reach of the Pcdh
enhancers along the cluster in OSN-WAPL-cKO
relative to OSN-WT and OSN-RAD21-cKO, as
assayed by in situ Hi-C and virtual 4C (Fig. 4B
g
seq (Fig. 4E and fig. S4D). These changes were higher in 5-HTs compared with mOSNs (Fig.
also evidenced by the mean number of isoforms 5B and fig. S5A). To determine whether high
expressed per cell whereby mOSNs displayed WAPL promotes Pcdhac2 expression, we con-
an increase in only the Pcdhb isoforms in OSN- ditionally removed Wapl in 5-HTs using a Cre
WAPL-cKO compared with OSN-WT (fig. S4, driver under the control of the promoter of the

and fig. S4C). We next calculated the contact
probability as a function of genomic distance
[P(s)] and its derivative to probe the proces-
sivity of cohesin. By this metric, cohesin proces-
sivity in OSN-WAPL-cKO relative to OSN-WT
increased by almost 3-fold (Fig. 4C). We then
measured the size of anchored loops and
observed a 1.5-fold increase between OSN-
WAPL-cKO and OSN-WT (Fig. 4D). Although
the P(s) in OSN-WAPL-XP remained mostly
y
E and F). serotonin transporter Slc6a4 [Sert(Cre/+);
To measure the impact of WAPL levels on Waplfl/fl;tdTomatofl/+, 5-HT-WAPL-cKO] (Fig.
Pcdh isoform diversity, we quantified the prob- 5C, top, and fig. S5B). We then used RNA FISH
ability of choice of Pcdh promoters in OSN-WT, to measure the levels of Pcdhac2 mRNA using
OSN-WAPL-XP, and OSN-RAD21-cKO relative tryptophan hydroxylase 2 (Tph2) as a marker
to OSN-WAPL-cKO. Increasing WAPL levels for serotonergic somas in the raphe. Consistent
resulted in cluster biases in Pcdh diversity that with our model (Fig. 2A), removal of WAPL
reflected the genomic size of the individual Pcdh resulted in down-regulation of expression of
gene clusters and the relative distances of their Pcdhac2 (Fig. 5, C and D).
promoters with their respective cluster-specific If high WAPL favors the biased expression

unchanged compared with OSN-WT (Fig. 4C),
we observed a reduction in the size of anchored
loops genome wide (Fig. 4D), consistent with
the shorter reach of contacts of Pcdh enhancers
with their promoters (Fig. 3D and fig. S3E). Fi-
y g
enhancers (Fig. 4F, Pcdhg > Pcdha > Pcdhb, of Pcdhac2 in 5-HTs by promoting low cohesin
OSN-WT versus OSN-WAPL-XP). The distance processivity, then Pcdhac2 promoter choice
biases in Pcdh isoform diversity that were de- should be independent of cohesin activity in
pendent on WAPL concentration required these neurons, as we show to be the case in
cohesin activity (Fig. 4F, OSN-RAD21-cKO). mOSNs (Fig. 1B). To test this, we removed Rad21
in 5-HTs [Sert(Cre/+);Rad21fl/fl;tdTomatofl/+,

nally, only two anchored loops were detected in
OSN-RAD21-cKO (Fig. 4D), consistent with pre-
vious studies of cohesin depletion in cells (38).
Increasing cohesin processivity on the Pcdh
locus by removing WAPL increased the prob-
ability of choice of promoters that are located
farther from their respective enhancers (Fig.
4E and fig. S4D). These changes scaled with
the relative distances of each Pcdh promoter
to its enhancer across all three clusters and
were consistent with the increased loop sizes
observed at the Pcdh locus and genome wide.
The distal Pcdhb promoters, which are located
farthest from their respective enhancers (up
to 600,000 bp away), had a larger fold change
in activation, whereas the Pcdhg promoters,
which are located nearest to their enhancers
,
These data indicate that during OSN devel-
opment, cohesin processivity by WAPL tunes 5-HT-RAD21-cKO] (fig. S5C). Consistent with
Pcdh isoform diversity and suggest a model our hypothesis, ablation of Rad21 did not change
for how the deterministic expression of the Pcdhac2 expression compared with Pcdhac2
enhancer-proximal Pcdhac2 could be favored expression in Rad21 heterozygous littermates
in other neural types (e.g., in 5-HTs). [Sert(Cre/+);Rad21fl/+;tdTomatofl/+, 5-HT-RAD21-
het] (fig. S5D).
High WAPL favors deterministic Pcdhac2 5-HT axons project pervasively from the
expression and axon tiling in serotonergic neurons raphe nuclei through a tiling distribution to
Unlike OSN axons, which converge, 5-HT axons several other regions in the brain, including the
tile (Fig. 5A). To determine whether Pcdhac2 hippocampus (14, 16). To investigate whether
expression and axon tiling of 5-HTs require reduced Pcdhac2 expression upon Wapl dele-
high WAPL, we first investigated whether Wapl tion results in 5-HT wiring defects, we exam-
mRNA was differentially expressed between ined the ability of 5-HT axons to project to the
5-HTs and mOSNs. We compared published hippocampus (Fig. 5A). We found that 5-HT
RNA-seq data from fluorescence-activated cell axons from 5-HT-WAPL-cKO animals presented
sorting (FACS)–isolated 5-HT nuclei (39) with a clumping phenotype (Fig. 5E) that qualitatively

Kiefer et al., Science 380, eadf8440 (2023) 23 June 2023 5 of 11

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

A B View OSN-RAD21-cKO OSN-WT OSN-WAPL-cKO

Omp cre point
100 1 Mb
loxP loxP

HS18-22 HS16-17 HS5-1

2
100
OSN-WAPL-cKO Wapl 2
100
2
100
2
100
cohesin 2
100
2
C 100
2
OSN-RAD21-cKO 100
10-2 OSN-WAPL-XP 2
100
OSN-WT 2
OSN-WAPL-cKO
IC contact frequency

10-3

HS5-1

HS16-17
HS18-22
10-4
-0.5
CTCF
Slope

10-5 -1.0 motifs

-1.5
10-6
105 106 107 108 105 106 107 108

p
Separation (bp) Separation (bp)
D E F
Linear genomic distance to respective enhancers Pcdh Pcdh Pcdh Pcdh
600kb

Fold change in probability of isoform choice

300kb 200kb
Pcdh Pcdh Pcdh
OSN-RAD21-cKO OSN-WT

(relative to OSN-WAPL-cKO)

g
2

HS16-17
HS18-22
HS5-1

1 12 c1 c2 1 22 a1 a12
OSN-WAPL-XP OSN-WAPL-cKO
log2 fold change in isoform choice
(OSN-WAPL-cKO/OSN-WT)
****

3
****

y
0.5 2 1
0.4
Density

0.3 1
0.2
0.1
0
0.0 0
103 102
Loop size (kb)
−1
600 400 200 0 OSN- OSN- OSN-
Genomic distance from respective enhancers (kb) WT WAPL-XP RAD21-cKO

y g
Fig. 4. Ablation of WAPL in mature OSNs results in high cohesin processivity OSN-RAD21-cKO (two loops), OSN-WAPL-XP (2943 loops, mean loop size of 225 kb),
and an increase in Pcdh isoform diversity. (A) Schematic of the transgenic OSN-WT (4820 loops, mean loop size of 262 kb), and OSN-WAPL-cKO (7280 loops,
mouse line used to conditionally delete WAPL in OSNs (OSN-WAPL-cKO) and its mean loop size of 374 kb). (E) Top: Linear genomic distance of Pcdh promoters
mechanistic prediction. (B) Virtual 4C from in situ Hi-C of the chromatin contacts with respect to their enhancers. Bottom: Log2-fold change in Pcdh promoter choice
from the HS5-1, the HS16-17, and the HS18-22 enhancers in OSN-RAD21-cKO, calculated by single-cell RNA-seq between OSN-WT and OSN-WAPL-cKO mOSNs as
OSN-WT, and OSN-WAPL-cKO, respectively. (C) Contact frequency as a function of a function of the linear genomic distance of the promoters with respect to their

,
genomic distance (left) and the first derivative (right). Average loop sizes: OSN- enhancers. (F) Quantification of Pcdh isoform diversity. Probability of Pcdh isoform
RAD21-cKO, not applicable; OSN-WT, 117 kb; OSN-WAPL-XP, 133 kb; and OSN- choice in OSN-WT, OSN-WAPL-XP, and OSN-RAD21-cKO relative to OSN-WAPL-cKO.
WAPL-cKO, 333 kb. Results are the average of two biological replicates per For both (E) and (F), Pcdha, Pcdhb, and Pcdhg genes are shown in black, red,
genotype. (D) Loop anchor distances (corner dots called from in situ Hi-C data) for and blue, respectively; Pcdhac2 is shown in orange.

phenocopied the deletion of the Pcdhac2 gene shared across evolution, from invertebrates to grams necessary for different neural wiring
(14, 16). vertebrates (11). In flies, neurons that require patterns, from tiling to convergence (14, 18).
These data suggest that high WAPL in 5-HTs cell surface barcoding diversity for their cir- Here, we uncovered the molecular logic for
limits cohesin translocation along the Pcdh clus- cuit assembly leverage the massive alternative how the Pcdh gene cluster achieves transcrip-
ter to promote the biased expression of Pcdhac2 splicing of the Dscam1 pre-mRNA that alone tional modularity (Fig. 6). Given the astonishing
and 5-HT axon tiling. can generate >18,000 distinct protein isoforms diversity of neural patterns in the brain and the
(40–42). Conversely, neurons that require an role that Pcdh proteins play in neural self-
Discussion identical cell surface barcode to tile harness avoidance, we propose that different neural
The generation of cell surface protein isoform the Dscam2 gene (43), which instead only has types tune cohesin activity, and thereby its pro-
diversity as a mechanism to promote neural two isoforms. By contrast, in mice, the single cessivity, to generate the extent of Pcdh isoform
self-avoidance and axon wiring is a strategy Pcdh gene locus produces the barcoding pro- diversity that meets their vastly differing wiring

Kiefer et al., Science 380, eadf8440 (2023) 23 June 2023 6 of 11

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

A Hippocampus C

Ser cre
loxP loxP
5HT-WAPL-cKO Wapl cohesin
5-HT Dentate Gyrus
Tph2 Pcdh c2 Tph2 Pcdh c2 DAPI
DAPI
Raphe
nuclei

5-HT-WT
B
10.0
−log10(p−adjusted)

5-HT-WAPL-cKO
7.5
5 6 c2
5.0 4
Wapl
2.5 1
7
12 c1
0.0 OSNs 8 11 5-HTs

p
−5 0 5
log2 fold change (OSNs vs 5-HTs)

D E
SR
** **** CA1

5-HT-WT
60
intensity per TPH2+ cell
Pcdh c2 mRNA puncta

SLM
Pcdh c2 mRNA FISH

150
CA3

g
per TPH2+ cell

DG
40 MO
100
SERT

50 20
5-HT-WAPL-cKO

SR

y
CA1
0 0
SLM
5-HT- 5-HT- 5-HT- 5-HT-
WT WAPL-cKO WT WAPL-cKO CA3
DG
MO
SERT

Fig. 5. Ablation of WAPL in 5-HTs results in loss of Pcdhac2 expression intensity (right) of Pcdhac2 mRNA in Tph2+ cells in 5-HT-WT and 5-HT-WAPL-cKO
and axon tiling. (A) Schematic of 5-HT axon tiling from the raphe to the 5-HTs. Each dot represents data from an individual Tph2+ cell, with two biological
hippocampus. (B) Differential gene expression analysis between OSNs and replicates per condition. (E) 5-HT wiring defects in 5-HT-WAPLcKO mice relative to
5-HTs. Green, Wapl; orange, Pcdhac2; black, other Pcdha genes. (C) Top: schematic 5-HT-WT. Left: 5-HTs are indicated by IHC against SERT. White, SERT; blue,
of the transgenic mouse line used to conditionally delete WAPL in 5-HTs (5-HT- DAPI. CA1, cornu ammonis 1; CA3, cornu ammonis 3; DG, dentate gyrus. Right:

y g
WAPL-cKO) and its mechanistic prediction. Bottom: RNA-FISH in the raphe in Binary mask of SERT signal in presented images. SR, stratum radiatum; SLM,
5-HT-WT and 5-HT-WAPL-cKO mice. Magenta, Tph2; green, Pcdhac2; blue, DAPI. stratum lacunosum moleculare; MO, molecular layer. Red arrows indicate clumping
Scale bars, 50 mm. (D) Quantification of the number of puncta (left) and signal of 5-HT axons. Scale bar, 100 mm.

requirements. We uncovered WAPL as a key mouse models of Cornelia de Lange Syndrome neurons that extends beyond the regulation of
Pcdh gene expression (48, 49).

,
modulator of Pcdh isoform diversity (Fig. 6). (46), and in brains isolated from the cohesin-
However, given the many cohesin subunits STAG1 core subunit null embryos (47). On
and known protein regulators, we speculate the basis of these observations and the data Rheostats as strategies to overcome
that other mechanisms might exist, beyond presented here, it is reasonable to speculate transcriptional biases and achieve stochasticity
that of modulating WAPL levels, that define that the severe intellectual impairments as- of genes arranged in clusters
the combinatorial space of Pcdh isoform sociated with cohesinopathies arise from the We revealed an activity of the cohesin protein
diversity. dysregulation of clustered Pcdh expression. complex and its unloader, WAPL, in the di-
In support of this proposal, there are sev- Thus, the implications of our findings extend versification of cell surface proteomes required
eral observations that implicate genetic var- beyond the mechanism of regulating healthy for brain wiring. Aspects of our proposed mech-
iants of the cohesin protein complex and its brain development and establish a biochem- anism extend beyond the Pcdh gene locus and
regulators associated with cohesinopathies, ical ground between the genetic variants of apply more generally to the regulation of other
including Cornelia de Lange syndrome and the cohesin protein complex and Pcdh dysreg- gene clusters with expression programs that
Roberts syndrome (44), with dysregulation of ulation. More broadly, we also wonder whether are required for cellular diversification and
Pcdh expression. It was found that clustered our proposed model could provide a molec- fate. There is precedence for such a process
Pcdh genes are down-regulated in cells derived ular framework for the recently suggested in B cells, which also control their cohesin
from Cornelia de Lange syndrome patients link among cohesin, 3D genome structure, activity through WAPL to maximize antibody
(45), in NIPBL (a cohesin loader)–heterozygous maturation, and function of mouse cortical diversity using V(D)J recombination during

Kiefer et al., Science 380, eadf8440 (2023) 23 June 2023 7 of 11

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

Axon OSNs were achieved as described previously

tiling (24). Briefly, Rad21 conditional allele mice were
Pcd
Co
hi
so
crossed to Omp(iresCre/+) mice (Omptm1(cre)Jae).
he c2
sin HS5-1 Recombined cells were purified by including

fo
rm
WAPL a Cre-inducible tdTomato allele [ROSA26-

pr

div
oc
tdtomato, Gt(ROSA)26Sortm14(CAG–tdTomato)Hze/J]

ersity
ess
in the cross and selecting tdTomato+ cells by

iv
high low

ity
Serotonergic
c2
HS5-1 neurons FACS. In the text and the figures, we refer to the
Rad21 cKO in OSNs as OSN-RAD21-cKO for
homozygous deletion of the floxed allele and as
OSN-RAD21-het for heterozygous deletion of the
floxed allele. The WAPL-overexpressing mouse
was obtained by microinjection of a linear-
Axon
convergence ized DNA plasmid containing the tetO-WAPL-
Pcd
hi
mCherry cassette using the Gladstone mouse
Co so facility core at the University of California San
he
sin
fo

3 HS5-1 Francisco (UCSF). A total of four transgenic

rm

WAPL
pr

div

founders were characterized with two to three

oc

ersity
ess

biological replicates each. When crossed to

ivity

high low Omp(tTa) mice, WAPL-overexpressing OSNs

Olfactory sensory
neurons were sorted using mCherry. The WAPL condi-

p
6 HS5-1 tional allele mouse was generated as described
previously (54). WAPL floxed mice were crossed
to Omp(iresCre/+) or to Sert(Cre) (JAX strain
014554) mice to conditionally delete WAPL in
Fig. 6. Model: WAPL functions as a rheostat of cohesin processivity and Pcdh isoform diversity that
OSNs or 5-HTs, respectively. In the text and the
regulates neural wiring patterns. High WAPL limits cohesin extrusion through the Pcdh locus, favoring
figures, we refer to the WAPL cKO as OSN-WAPL-
the choice of Pcdh promoters that are located closer to their enhancers. In 5-HTs, high WAPL favors the

g
cKO and 5-HT-WAPL-cKO for homozygous
choice of the sole Pcdhac2 promoter, thus constraining 5-HT axon wiring to a tiling pattern. As WAPL levels
deletion of the floxed allele in OSNs and 5-HTs,
decrease, cohesin processivity increases. Cohesin extrusion through the Pcdh locus erases genomic distance
respectively. Primers used for genotyping are
biases in Pcdh promoter choice by the distal enhancers. Therefore, as WAPL levels decrease, Pcdh isoform
listed in table S1.
diversity increases. In OSNs, low WAPL favors stochastic Pcdh promoter choice independent of enhancer-

y
promoter distance, a mechanism that generates sufficient Pcdh protein isoform diversity required for convergence FACS sorting of mouse OSNs
of OSN axons. Only Pcdha genes (rainbow colors with Pcdhac2 in orange) are shown. Blue, cohesin; red,
Cells were dissociated into a single-cell sus-
WAPL rheostat. Barcodes correspond to the colors of the Pcdha isoform chosen.
pension by incubating freshly dissected main
OE with papain for 30 to 40 min at 37°C ac-
maturation (50, 51). In that context, cohesin On the basis of these previous studies and our cording to the Worthington Papain Dissoci-
promotes contraction of the mouse immuno- present results, we propose that a rheostat- ation System. After dissociation and filtering
globulin locus to facilitate random contacts operating logic is used by cells to enable through a 35-mm cell strainer, cells were re-
between the VH genes and the recombined promoter stochasticity in gene clusters, the suspended in 1× phosphate-buffered saline
DJH segments by the recombination center architectures of which demand mechanisms (PBS) with 2% fetal bovine serum (FBS) with
located at the 3′ end of the locus (51). Also in to overcome transcriptional biases. Given that DNase (0.0025% final concentration) and 4′,6-

this context, WAPL has been proposed to reg-
ulate the distance that cohesin can extrude
along the immunoglobulin locus to maximize
antibody diversification (50, 51). These obser-
vations suggest a rheostat-like function of
rheostat-operating mechanisms might require
large changes in the concentration of protein
factors with activity that is also essential for
cell division (53), as is the case for WAPL, we
hypothesize that the use of such strategies
y g
diamidino-2-phenylindole (DAPI). For in situ
Hi-C experiments, upon dissociation, cells were
fixed with 1% formaldehyde for 10 min at room
temperature (RT). Formaldehyde was quenched
by adding glycine to a final concentration of

WAPL also in the regulation of V(D)J recombi-
nation and antibody diversity.
In addition, there is a parallel in the logic
of promoter stochasticity between Pcdh genes
and OR genes. In fact, single OR choice in
OSNs necessitates the anatomically regulated
rheostatic-like activity of the NFIA, B, and X
transcription factors that regulate the gradi-
ent of heterochromatin assembly and genome
compartmentalization along the dorsoventral
axis of the OE to promote zonal OR promoter
choice (52). Our studies suggest that proper
assembly of olfactory maps requires the devel-
opmental regulation of yet another rheostat,
that of WAPL, which enables Pcdh diversity in
mOSNs by controlling cohesin processivity on
the Pcdh locus.
could be confined to noncycling cells such as
B cells and postmitotic neurons.
Materials and methods
Animals
Mice were treated in compliance with the rules
and regulations of the institutional animal
care and use committee under protocol num-
ber AN-170364-03F. For all experimental pro-
cedures performed in OSNs, both male and
female animals between 3 and 24 weeks of age
were used. For experiments performed to in-
vestigate Pcdh expression in 5-HTs, both male
and female animals between 2 and 4 weeks of
age were used. Primary FACS-sorted cells were
obtained from dissected main OE. OSNs were
sorted from Omp(iresGFP/+) mice. Rad21 cKO
,
0.125 M for 5 min at RT. Cells were then washed
once and resuspended in cold PBS with 2% FBS
and DNase (0.0025% final concentration). Fluo-
rescent cells were then sorted on a FACSAria II
cell sorter (BD Biosciences).
RNA isolation and sequencing studies
RNA was isolated from tissue using TRIzol. Cell
lysate was extracted with bromo-chloropropane,
and RNA was precipitated with 100% isopropa-
nol supplemented with 10 mg of glycoblue for
10 min at RT and then pelleted at 16,000g for
30 min at 4°C. The RNA pellet was washed
once with 75% ethanol and then resuspended
in RNase-free water to a maximal concentration
of 200 ng/ml. Genomic DNA contaminants were
removed by Turbo DNase. Removal of Turbo

Kiefer et al., Science 380, eadf8440 (2023) 23 June 2023 8 of 11

RES EARCH | R E S E A R C H A R T I C L E
DNase was performed by phenol:chloroform
extraction, and RNA was precipitated as de-
scribed above, resuspended in RNase-free water,
and stored at –80°C.
Sequencing libraries for total RNA were made
using the SMARTer Stranded Total RNA-Seq
Pico input mammalian RNA kit v2. The quality
and quantity of all libraries were assessed by
Bioanalyzer and Qubit. Libraries were sequenced
on a NextSeq 500/550 (UCSF Gladstone Ge-
nomic Core). Unless otherwise stated, RNA-
seq data represent two biological replicates
per condition.
Single-cell RNA-seq studies
Cells were dissociated from the main OE from
8- to 12-week-old female mice as described
above. About 16,000 cells were FACS purified
for OSNs, submitted for 10x Genomics GEM
generation using the Single Cell 5′ Gene Expres-
sion set-up, and sequenced using a NovaSeq
SP100 PE50 (UCSF Gladstone Genomic Core
and the UCSF Center for Advanced Technol-
ogy) and a NextSeq2000 P2 200. The data
were analyzed using 10x Genomics Cell Ranger
6.0.1 with a recovery of 4000 to 8000 cells per
experiment. Data analysis was performed
using Seurat (55–58) and custom scripts. Pcdh
isoform choice in mOSNs was defined as
the detection of UMI-corrected reads in cells
of high Omp expression (Omp+, Gap43–).
Single-cell RNA-seq analysis and statistics
are provided in table S2. Data for single-cell
experiments represent one biological repli-
cate per condition.
In situ Hi-C
About 500,000 sorted OSNs were lysed and in-
tact nuclei were processed through an in situ
Hi-C protocol as previously described with a
few modifications (59). Briefly, cells were lysed
with lysis buffer [50 mM Tris, pH 7.5, 0.5%
igepal, 0.25% sodium deoxycholate, 0.1% so-
dium dodecyl sulfate (SDS), 150 mM NaCl, and
protease inhibitors]. Pelleted intact nuclei
were then resuspended in 0.5% SDS and in-
cubated for 20 min at 65°C for nuclear perme-
abilization. After quenching with 1.1% Triton-X
for 10 min at 37°C, nuclei were digested with
6 U/ml of DpnII in 1x DpnII buffer overnight
at 37°C. After the initial digestion, a second
DpnII digestion was performed at 37°C for
2 hours. DpnII was heat-inactivated at 65°C
for 20 min. For the 1.5 hours fill-in at 37°C,
biotinylated dATP (Jena Bioscience) was used
instead of dATP to increase ligation efficiency.
Ligation was performed at 25°C for 4 hours.
Nuclei were then pelleted and sonicated in
sonication buffer (10 mM Tris pH 7.5, 1 mM
EDTA, 0.25% SDS) on a Covaris S220 (peak
power 105.0; duty factor 2.0; cycle/burst 200;
treatment time 960 s; temperature 4 to 8°C).
DNA was reverse cross-linked overnight at
65°C with proteinase K and RNAse A. Re-
Kiefer et al., Science 380, eadf8440 (2023) 23 June 2023
verse cross-linked DNA was purified with 2×
AMPure beads following the standard pro-
tocol. Biotinylated fragments were enriched
using Dynabeads MyOne Streptavidin T1 beads.
The biotinylated DNA fragments were pre-
pared for next-generation sequencing on the
beads by using the Nugen Ovation Ultralow
kit protocol with some modifications. After
end repair, magnetic beads were washed twice
at 55°C with 0.05% Tween, 1 M NaCl in Tris/
EDTA, pH 7.5. Residual detergent was removed
by washing the beads twice in 10 mM Tris,
pH 7.5. End repair buffers were replenished
to original concentrations, but the enzyme
and enhancer was omitted before adapter liga-
tion. After adapter ligation, beads underwent
five washes with 0.05% Tween, 1 M NaCl in Tris/
EDTA, pH 7.5, at 55°C and two washes with
10 mM Tris, pH 7.5. DNA was amplified by
10 cycles of polymerase chain reaction irre-
spective of starting material. Beads were re-
claimed, and amplified DNA fragments were
purified with 0.8× AMPure beads. The OSN-
WAPL-cKO and OSN-WAPL-XP Hi-C datasets
were generated using the Arima Hi-C kit ac-
cording to the manufacturer’s instructions.
Quality and concentration of libraries were
assessed by Agilent Bioanalyzer and Qubit.
Samples were paired-end sequenced on a
NextSeq 550 and 2000 and a NovaSeq (UCSF
Gladstone Genomic Core and the UCSF Center
for Advanced Technology). Hi-C analysis and
statistics are provided in table S3. For OSN-
WAPL-XP, two Hi-C biological replicates were
performed using cells FAC sorted from seven
and five animals per replicate. For OSN-WAPL-
cKO, two Hi-C biological replicates were per-
formed using cells FAC sorted from one animal
per replicate.
Bioinformatic analysis of sequencing data
For RNA-seq experiments, raw FASTQ files were
aligned with STAR using the mm10 reference
genome. The initial four base pairs of both
paired reads were trimmed before alignment.
Differential expression analysis was performed
using DESeq2 (60). When comparing OSNs
and 5-HTs, seven biological replicates from
FACS-isolated OSNs and three published data-
sets from FACS-isolated raphe nuclei from 5-
HTs were used (39). When comparing OSN-WT
with OSN-RAD21-cKO, OSN-RAD21-het, OSN-
WAPL-XP, or OSN-WAPL-cKO, seven biological
replicates from FACS-isolated OSN-WT, two
biological replicates from FACS-isolated OSN-
RAD21-cKO and OSN-RAD21-het, eight biological
replicates from FACS-isolated OSN-WAPL-
XP, and two biological replicates from FACS-
isolated OSN-WAPL-cKO were used.
For in situ Hi-C experiments, raw FASTQ files
were processed using HiC-Pro (61, 62) (table S3).
The Hi-C maps were generated from raw Hi-C
matrices in the cooler format by binning valid
pairs into 5-kb bins. Raw matrices for the two
replicates were merged (using cooler merge)
and iteratively corrected (using cooler bal-
ance) (63). The normalized matrices were then
smoothed and interpolated for visualization
(using cooltools adaptive_coarsegrain and interp_
nan, https://doi.org/10.5281/zenodo.5214125). The
contact probability as a function of genomic
separation P(s) was computed using cooltools.
expected_cis. Loop sizes were estimated based
on the position of the peak in the log-derivative
of P(s), with a correction factor equal to 0.6 ac-
cording to (64). Anchored loops were called
using Mustache (65) on the balanced cool files
of two merged Hi-C replicates binned at 5-kb
resolution. The threshold P value for an in-
teraction to be reported was set at 0.05 (using
option –pt). Distances between anchors were
computed based on the midpoints of the an-
chor coordinates.
Preparation of mouse brain tissue sections
p
Animals were perfused using standard trans-
cardial perfusion with ice-cold PBS and 4%
paraformaldehyde (PFA), and brain tissue was
immediately dissected. For experiments using
OE and OB tissues, the OE/OB were dissected
and immersion fixed in 4% PFA for 8 min,
g
washed in PBS for 5 min three times at RT and
incubated in 30% sucrose overnight at 4°C.
The following day, samples were incubated in
a 1:1 30% sucrose and optimal cutting tempera-
ture (OCT) solution for 1 hour and then moved
y
to OCT for 1 hour before freezing. Samples were
frozen in embedding molds using isopropanol
and dry ice. For experiments targeting 5-HT
neurons in the raphe and hippocampus, whole
brains were dissected and placed in 4% PFA
overnight at 4°C. The brains were then washed
in PBS for 5 min three times at RT and trans-
ferred to 30% sucrose at 4°C overnight. Once
sunk, OB tissue was removed from 30% su-
crose, incubated in a tube containing a 1:1
y g
solution of 30% sucrose and OCT embedding
compound (Sakura, #4583) for 1 hour at RT,
followed by another hour of RT incubation
in an embedding mold containing pure OCT.
Whole brains, once sunk, were directly trans-
,
ferred from 30% sucrose solution to an em-
bedding mold containing 2:1 30% sucrose and
OCT. To freeze the tissue in its embedding
solution, the mold was flash frozen in dry ice
and isopropanol. Tissue was stored at –80°C
until sectioning. Samples in OCT were placed
in a Leica cryostat (Leica Microsystems, Wetzlar,
Germany) for 30 to 45 min before sectioning
to equilibrate. OCT was used to freeze samples
to cryostat chucks. Sectioning was performed
with a chamber temperature of –21°C and an
object temperature of –17°C. Samples were
sectioned coronally at 16 mm (OE and OB) and
at 20 mm (hippocampus and raphe). All sec-
tions were mounted onto Superfrost Plus slides
(Fisher Scientific, #12-550-15) and stored at
–20°C until downstream use.
9 of 11
RES EARCH | R E S E A R C H A R T I C L E
Immunohistochemistry
For immunostaining of OB tissue, sections
were fixed with 4% PFA in PBS, washed for
5 min three times with PBS containing 0.1%
Triton X-100, and blocked in 4% donkey
serum in PBS plus 0.1% Triton X-100 for 30 min
at RT. For immunostaining of the hippocam-
pus region, sections were directly washed with
PBS plus 0.1% Triton X-100 and incubated with
the same blocking serum for 30 min at RT. After
the blocking step, the slides were incubated in
a humid chamber with primary antibodies di-
luted in blocking serum for 24 hours at 4°C. The
following primary antibodies were used: rabbit
anti-SERT (1:500, Millipore Sigma, #PC177L100UL),
guinea pig anti-VgGLUT2 (1:2000, Millipore Sigma,
#AB2251-I), guinea pig anti-m71 (1:1000, from
the laboratory of G. Barnea), rabbit anti-WAPL
(1:400, Invitrogen, #MA5-38145), rabbit anti-
RAD21 (1:100, Invitrogen, #PA528344), goat
anti-OMP (1:1000, Fujifilm Wako Chemicals,
#544-10001), and rabbit anti-GAP43 (1:500,
Abcam, #ab75810). After extensive washing
with PBS plus 0.1% Triton X-100, sections were
incubated with Alexa Fluor–conjugated sec-
ondary antibodies diluted in blocking serum for
2 hours before being washed again and mounted
with Vectashield Antifade Mounting Medium
with DAPI (Vector Laboratories). Alexa Fluor
488 donkey immunoglobulin G (IgG) anti-rabbit
(1:2000, Invitrogen, #A21206), and Alexa Fluor
647 goat anti–guinea pig IgG (1:2000, Invitrogen,
#A21450) were used as secondary antibodies. For
all immunohistochemistry experiments, at least
two biological replicates were characterized,
and representative images are shown.
Hybridization chain reaction experiments
on fixed tissue sections
RNA-FISH experiments were performed using
hybridization chain reaction (HCR) technol-
ogy from Molecular Instruments. Fixed tissue
sections from postnatal day 20 mice were
stored at –20°C and allowed to dry at RT for
5 min. An HCR protocol (dx.doi.org/10.17504/
protocols.io.8epv59725g1b/v2) was then per-
formed. The following probes and matching
hairpins ordered from Molecular Instruments
were used: Tph2 (lot number PRE854), Pcdhac2
(lot number PRK987), Omp (lot number PRK986),
and Wapl (lot number PRQ058). The hairpin am-
plification systems used for Tph2 and Pcdhac2
were B1 at 647 nm and B3 at 488 nm, respec-
tively. The B4 546-nm hairpin amplification
system was used for both Omp and Wapl.
Imaging studies
Sections from the OB and raphe were imaged
using a Yokogawa CSU-W1 SoRa spinning disk
confocal microscope. Sections from the hippo-
campus were imaged using a Leica DMi8 in-
verted microscope and stitched using the LAS
X software platform. All images were postpro-
cessed in Fiji software.
Kiefer et al., Science 380, eadf8440 (2023) 23 June 2023
Quantification of Pcdhac2 expression in the
raphe nuclei
Pcdhac2 and Wapl mRNA expression levels in
individual 5-HT neurons were quantified using
a custom CellProfiler pipeline (66). The pipe-
line categorized input images from HCR ex-
periments into groups of three as follows: (i)
images containing Pcdhac2 or Wapl mRNA
signal, (ii) images containing the Tph2 mRNA
signal, and (iii) manual segmentations of 5-HT
somas based on Tph2 expression performed in
Fiji. The pipeline first identified 5-HT somas
from the manual segmentations using the
two-class Otsu thresholding method with a
correction factor of 1.0. These somas were then
used as a mask over the Pcdhac2/Wapl images,
only keeping signal residing within 5-HT somas.
The masked Pcdhac2/Wapl images were then
enhanced for speckle features with a maxi-
mum size of 10 pixels to make mRNA puncta
easier to identify. Pcdhac2 and Wapl puncta
were identified using the robust background
thresholding method with a correction factor
of 1.0, lower and upper outlier fractions of
0.05, and 3 SDs for Pcdhac2 images. Five SDs
were used for Wapl images to account for the
slightly lower signal-to-noise ratio generated
by the Wapl probes relative to the Pcdhac2
probes. Clumped puncta were distinguished
by shape, and dividing lines between clumped
objects were drawn based on intensity. Finally,
the RelateObjects module was used to assign
Pcdhac2 and Wapl puncta to their parent 5-HT
somas. Data for all Pcdhac2 and Wapl puncta
and Tph2 soma objects were exported using
the ExportToSpreadsheet module. To quantify
Pcdhac2 and Wapl mRNA abundance, both
the number of puncta in Tph2+ somas and the
total intensity of all puncta within a soma were
measured. Both metrics agreed on the reported
effects on Pcdhac2 and Wapl expression.
To determine sample size for experiments
comparing Pcdhac2 and Wapl expression in
5-HT neurons, the number of cells required to
have an 80% power, at probability level 0.05,
of detecting a difference of 10 puncta per cell
between an experimental condition and a con-
trol condition with a mean of 60 puncta per
cell was calculated, with both conditions having
SDs of 20 puncta. Our calculations estimated
65 cells per condition, and between 65 and
140 cells per condition from two biological
replicates were used in each experiment. The
mean and SD values were determined from
pilot experiments in WT tissue.
Quantification of IHC and RNA-FISH signal
in the OE
For quantification of IHC images, signal inten-
sity profiles were generated in Fiji and aver-
aged across three biological replicates. WAPL,
RAD21, and OMP signals were quantified using
equally sized rectangular selections spanning
from the basal lamina of the OE to the apical
sustentacular cell layer. For quantification of
RNA-FISH signal, Pcdhac2, Wapl, and Omp
puncta were identified using the same Cell-
Profiler pipeline as described above. Two im-
ages were quantified for each mRNA, with each
being from a different biological replicate. In
each of these images, a line was manually drawn
in Fiji outlining the basal lamina of the OE.
These lines were exported as a series of (x,y)
coordinates. For each punctum identified in
the CellProfiler pipeline, the shortest distance
was calculated between the punctum and the
basal lamina using the punctum’s (x,y) coordi-
nates and the coordinates of the basal line.
These distances were plotted as histograms
with 25 bins for each RNA.
Quantification and statistics
Statistical tests were performed using two-
sided Wilcoxon rank-sum tests. Statistically
significant effects are reported in the respective
p
figure panels using asterisks as *P < 0.05,
**P < 0.01, ***P < 0.001, and ****P < 0.0001;
otherwise, the P values associated with the
observed transcriptional changes are provided
in Figure 5B.
g
REFERENCES AND NOTES
1. N. Maizels, Immunoglobulin gene diversification. Annu. Rev.
Genet. 39, 23–46 (2005). doi: 10.1146/annurev.
genet.39.073003.110544; pmid: 16285851
2. K. Monahan, S. Lomvardas, Monoallelic expression of olfactory
y
receptors. Annu. Rev. Cell Dev. Biol. 31, 721–740 (2015).
doi: 10.1146/annurev-cellbio-100814-125308; pmid: 26359778
3. G. Mountoufaris, D. Canzio, C. L. Nwakeze, W. V. Chen,
T. Maniatis, Writing, Reading, and Translating the Clustered
Protocadherin Cell Surface Recognition Code for Neural Circuit
Assembly. Annu. Rev. Cell Dev. Biol. 34, 471–493 (2018).
doi: 10.1146/annurev-cellbio-100616-060701; pmid: 30296392
4. K. Monahan, A. Horta, S. Lomvardas, LHX2- and LDB1-
mediated trans interactions regulate olfactory receptor choice.
Nature 565, 448–453 (2019). doi: 10.1038/s41586-018-0845-0;
pmid: 30626972
5. Q. Wu, T. Maniatis, A striking organization of a large family
of human neural cadherin-like cell adhesion genes. Cell 97,
779–790 (1999). doi: 10.1016/S0092-8674(00)80789-8;
y g
pmid: 10380929
6. Q. Wu, T. Maniatis, Large exons encoding multiple ectodomains
are a characteristic feature of protocadherin genes. Proc. Natl.
Acad. Sci. U.S.A. 97, 3124–3129 (2000). doi: 10.1073/
pnas.97.7.3124; pmid: 10716726
7. S. Yokota et al., Identification of the cluster control region
for the protocadherin-beta genes located beyond the
,
protocadherin-gamma cluster. J. Biol. Chem. 286,
31885–31895 (2011). doi: 10.1074/jbc.M111.245605;
pmid: 21771796
8. P. Kehayova, K. Monahan, W. Chen, T. Maniatis, Regulatory
elements required for the activation and repression of the
protocadherin-a gene cluster. Proc. Natl. Acad. Sci. U.S.A. 108,
17195–17200 (2011). doi: 10.1073/pnas.1114357108;
pmid: 21949399
9. D. Canzio, T. Maniatis, The generation of a protocadherin
cell-surface recognition code for neural circuit assembly.
Curr. Opin. Neurobiol. 59, 213–220 (2019). doi: 10.1016/
j.conb.2019.10.001; pmid: 31710891
10. J. L. Lefebvre, Neuronal territory formation by the atypical
cadherins and clustered protocadherins. Semin. Cell Dev. Biol.
69, 111–121 (2017). doi: 10.1016/j.semcdb.2017.07.040;
pmid: 28756270
11. S. L. Zipursky, W. B. Grueber, The molecular basis of self-
avoidance. Annu. Rev. Neurosci. 36, 547–568 (2013).
doi: 10.1146/annurev-neuro-062111-150414; pmid: 23841842
12. S. L. Zipursky, J. R. Sanes, Chemoaffinity revisited: Dscams,
protocadherins, and neural circuit assembly. Cell 143, 343–353
(2010). doi: 10.1016/j.cell.2010.10.009; pmid: 21029858
10 of 11
RES EARCH | R E S E A R C H A R T I C L E
13. J. L. Lefebvre, J. R. Sanes, J. N. Kay, Development of dendritic
form and function. Annu. Rev. Cell Dev. Biol. 31, 741–777 (2015).
doi: 10.1146/annurev-cellbio-100913-013020; pmid: 26422333
14. W. V. Chen et al., Pcdhac2 is required for axonal tiling and
assembly of serotonergic circuitries in mice. Science 356,
406–411 (2017). doi: 10.1126/science.aal3231; pmid: 28450636
15. S. Katori et al., Protocadherin-alpha family is required for
serotonergic projections to appropriately innervate target brain
areas. J. Neurosci. 29, 9137–9147 (2009). doi: 10.1523/
JNEUROSCI.5478-08.2009; pmid: 19625505
16. S. Katori et al., Protocadherin-aC2 is required for diffuse
projections of serotonergic axons. Sci. Rep. 7, 15908 (2017).
doi: 10.1038/s41598-017-16120-y; pmid: 29162883
17. J. L. Lefebvre, D. Kostadinov, W. V. Chen, T. Maniatis,
J. R. Sanes, Protocadherins mediate dendritic self-avoidance
in the mammalian nervous system. Nature 488, 517–521
(2012). doi: 10.1038/nature11305; pmid: 22842903
18. G. Mountoufaris et al., Multicluster Pcdh diversity is required
for mouse olfactory neural circuit assembly. Science 356,
411–414 (2017). doi: 10.1126/science.aai8801; pmid: 28450637
19. S. Hasegawa et al., The protocadherin-a family is involved
in axonal coalescence of olfactory sensory neurons into glomeruli
of the olfactory bulb in mouse. Mol. Cell. Neurosci. 38, 66–79
(2008). doi: 10.1016/j.mcn.2008.01.016; pmid: 18353676
20. S. Hasegawa et al., Constitutively expressed protocadherin-a
regulates the coalescence and elimination of homotypic olfactory
axons through its cytoplasmic region. Front. Mol. Neurosci. 5,
97 (2012). doi: 10.3389/fnmol.2012.00097; pmid: 23087612
21. S. Hasegawa et al., Distinct and cooperative functions for the
protocadherin-a, -b and -g clusters in neuronal survival and
axon targeting. Front. Mol. Neurosci. 9, 155 (2016).
doi: 10.3389/fnmol.2016.00155; pmid: 28066179
22. B. Tasic et al., Promoter choice determines splice site selection in
protocadherin alpha and gamma pre-mRNA splicing. Mol. Cell 10,
21–33 (2002). doi: 10.1016/S1097-2765(02)00578-6;
pmid: 12150904
23. X. Wang, H. Su, A. Bradley, Molecular mechanisms governing
Pcdh-g gene expression: Evidence for a multiple promoter and
cis-alternative splicing model. Genes Dev. 16, 1890–1905
(2002). doi: 10.1101/gad.1004802; pmid: 12154121
24. D. Canzio et al., Antisense lncRNA transcription mediates DNA
demethylation to drive stochastic protocadherin a promoter
choice. Cell 177, 639–653.e15 (2019). doi: 10.1016/j.
cell.2019.03.008; pmid: 30955885
25. Y. Guo et al., CTCF/cohesin-mediated DNA looping is required
for protocadherin a promoter choice. Proc. Natl. Acad. Sci. U.S.A.
109, 21081–21086 (2012). doi: 10.1073/pnas.1219280110;
pmid: 23204437
26. Y. Guo et al., CRISPR inversion of CTCF sites alters genome
topology and enhancer/promoter function. Cell 162, 900–910
(2015). doi: 10.1016/j.cell.2015.07.038; pmid: 26276636
27. K. Monahan et al., Role of CCCTC binding factor (CTCF)
and cohesin in the generation of single-cell diversity of
protocadherin-a gene expression. Proc. Natl. Acad. Sci. U.S.A.
109, 9125–9130 (2012). doi: 10.1073/pnas.1205074109;
pmid: 22550178
28. Z. Jia et al., Tandem CTCF sites function as insulators to
balance spatial chromatin contacts and topological enhancer-
promoter selection. Genome Biol. 21, 75 (2020). doi: 10.1186/
s13059-020-01984-7; pmid: 32293525
29. M. Merkenschlager, E. P. Nora, CTCF and cohesin in genome
folding and transcriptional gene regulation. Annu. Rev.
Genomics Hum. Genet. 17, 17–43 (2016). doi: 10.1146/
annurev-genom-083115-022339; pmid: 27089971
30. G. Fudenberg et al., Formation of chromosomal domains by
loop extrusion. Cell Rep. 15, 2038–2049 (2016). doi: 10.1016/
j.celrep.2016.04.085; pmid: 27210764
31. E. Alipour, J. F. Marko, Self-organization of domain structures
by DNA-loop-extruding enzymes. Nucleic Acids Res. 40,
11202–11212 (2012). doi: 10.1093/nar/gks925; pmid: 23074191
32. A. L. Sanborn et al., Chromatin extrusion explains key features
of loop and domain formation in wild-type and engineered
genomes. Proc. Natl. Acad. Sci. U.S.A. 112, E6456–E6465
(2015). doi: 10.1073/pnas.1518552112; pmid: 26499245
33. K. J. Ressler, S. L. Sullivan, L. B. Buck, Information coding in
the olfactory system: Evidence for a stereotyped and highly
organized epitope map in the olfactory bulb. Cell 79,
1245–1255 (1994). doi: 10.1016/0092-8674(94)90015-9;
pmid: 7528109
34. F. Wang, A. Nemes, M. Mendelsohn, R. Axel, Odorant receptors
govern the formation of a precise topographic map. Cell 93,
47–60 (1998). doi: 10.1016/S0092-8674(00)81145-9;
pmid: 9546391
Kiefer et al., Science 380, eadf8440 (2023) 23 June 2023
35. S. Kueng et al., Wapl controls the dynamic association of
cohesin with chromatin. Cell 127, 955–967 (2006).
doi: 10.1016/j.cell.2006.09.040; pmid: 17113138
36. G. Wutz et al., Topologically associating domains and
chromatin loops depend on cohesin and are regulated by
CTCF, WAPL, and PDS5 proteins. EMBO J. 36, 3573–3599
(2017). doi: 10.15252/embj.201798004; pmid: 29217591
37. J. H. I. Haarhuis et al., The cohesin release factor WAPL
restricts chromatin loop extension. Cell 169, 693–707.e14
(2017). doi: 10.1016/j.cell.2017.04.013; pmid: 28475897
38. S. S. P. Rao et al., Cohesin loss eliminates all loop domains.
Cell 171, 305–320.e24 (2017). doi: 10.1016/j.cell.2017.09.026;
pmid: 28985562
39. E. Södersten et al., A comprehensive map coupling histone
modifications with gene regulation in adult dopaminergic and
serotonergic neurons. Nat. Commun. 9, 1226 (2018).
doi: 10.1038/s41467-018-03538-9; pmid: 29581424
40. D. Hattori et al., Dscam diversity is essential for neuronal
wiring and self-recognition. Nature 449, 223–227 (2007).
doi: 10.1038/nature06099; pmid: 17851526
41. B. J. Matthews et al., Dendrite self-avoidance is controlled by
Dscam. Cell 129, 593–604 (2007). doi: 10.1016/
j.cell.2007.04.013; pmid: 17482551
42. W. M. Wojtowicz, J. J. Flanagan, S. S. Millard, S. L. Zipursky,
J. C. Clemens, Alternative splicing of Drosophila Dscam
generates axon guidance receptors that exhibit isoform-
specific homophilic binding. Cell 118, 619–633 (2004).
doi: 10.1016/j.cell.2004.08.021; pmid: 15339666
43. S. S. Millard, J. J. Flanagan, K. S. Pappu, W. Wu, S. L. Zipursky, Dscam2
mediates axonal tiling in the Drosophila visual system. Nature 447,
720–724 (2007). doi: 10.1038/nature05855; pmid: 17554308
44. E. Watrin, F. J. Kaiser, K. S. Wendt, Gene regulation and
chromatin organization: Relevance of cohesin mutations to
human disease. Curr. Opin. Genet. Dev. 37, 59–66 (2016).
doi: 10.1016/j.gde.2015.12.004; pmid: 26821365
45. P. Garcia et al., Disruption of NIPBL/Scc2 in Cornelia de Lange
Syndrome provokes cohesin genome-wide redistribution
with an impact in the transcriptome. Nat. Commun. 12, 4551
(2021). doi: 10.1038/s41467-021-24808-z; pmid: 34315879
46. S. Kawauchi et al., Multiple organ system defects and
transcriptional dysregulation in the Nipbl(+/-) mouse, a model
of Cornelia de Lange syndrome. PLOS Genet. 5, e1000650
(2009). doi: 10.1371/journal.pgen.1000650; pmid: 19763162
47. S. Remeseiro, A. Cuadrado, G. Gómez-López, D. G. Pisano,
A. Losada, A unique role of cohesin-SA1 in gene regulation and
development. EMBO J. 31, 2090–2102 (2012). doi: 10.1038/
emboj.2012.60; pmid: 22415368
48. Y. Fujita et al., Decreased cohesin in the brain leads to
defective synapse development and anxiety-related behavior.
J. Exp. Med. 214, 1431–1452 (2017). doi: 10.1084/
jem.20161517; pmid: 28408410
49. L. Calderon et al., Cohesin-dependence of neuronal gene
expression relates to chromatin loop length. eLife 11, e76539
(2022). doi: 10.7554/eLife.76539; pmid: 35471149
50. L. Hill et al., Wapl repression by Pax5 promotes V gene
recombination by Igh loop extrusion. Nature 584, 142–147
(2020). doi: 10.1038/s41586-020-2454-y; pmid: 32612238
51. H.-Q. Dai et al., Loop extrusion mediates physiological Igh locus
contraction for RAG scanning. Nature 590, 338–343 (2021).
doi: 10.1038/s41586-020-03121-7; pmid: 33442057
52. E. Bashkirova et al., Homeotic regulation of olfactory receptor
choice via NFI-dependent heterochromatic silencing and
genomic compartmentalization. bioRxiv 274035 [Preprint]
(2020); https://doi.org/10.1101/2020.08.30.274035.
53. I. F. Davidson, J.-M. Peters, Genome folding through loop extrusion
by SMC complexes. Nat. Rev. Mol. Cell Biol. 22, 445–464
(2021). doi: 10.1038/s41580-021-00349-7; pmid: 33767413
54. A. Tedeschi et al., Wapl is an essential regulator of chromatin
structure and chromosome segregation. Nature 501, 564–568
(2013). doi: 10.1038/nature12471; pmid: 23975099
55. Y. Hao et al., Integrated analysis of multimodal single-cell data.
Cell 184, 3573–3587.e29 (2021). doi: 10.1016/
j.cell.2021.04.048; pmid: 34062119
56. T. Stuart et al., Comprehensive integration of single-cell
data. Cell 177, 1888–1902.e21 (2019). doi: 10.1016/
j.cell.2019.05.031; pmid: 31178118
57. A. Butler, P. Hoffman, P. Smibert, E. Papalexi, R. Satija,
Integrating single-cell transcriptomic data across different
conditions, technologies, and species. Nat. Biotechnol. 36,
411–420 (2018). doi: 10.1038/nbt.4096; pmid: 29608179
58. R. Satija, J. A. Farrell, D. Gennert, A. F. Schier, A. Regev, Spatial
reconstruction of single-cell gene expression data. Nat. Biotechnol.
33, 495–502 (2015). doi: 10.1038/nbt.3192; pmid: 25867923
59. S. S. P. Rao et al., A 3D map of the human genome at kilobase
resolution reveals principles of chromatin looping. Cell 159,
1665–1680 (2014). doi: 10.1016/j.cell.2014.11.021; pmid: 25497547
60. M. I. Love, W. Huber, S. Anders, Moderated estimation of fold change
and dispersion for RNA-seq data with DESeq2. Genome Biol. 15,
550 (2014). doi: 10.1186/s13059-014-0550-8; pmid: 25516281
61. N. Servant et al., HiC-Pro: An optimized and flexible pipeline
for Hi-C data processing. Genome Biol. 16, 259 (2015).
doi: 10.1186/s13059-015-0831-x; pmid: 26619908
62. R. H. van der Weide et al., Hi-C analyses with GENOVA: A case
study with cohesin variants. NAR Genom. Bioinform. 3, lqab040
(2021). doi: 10.1093/nargab/lqab040; pmid: 34046591
63. N. Abdennur, L. A. Mirny, Cooler: Scalable storage for Hi-C data and
other genomically labeled arrays. Bioinformatics 36, 311–316
(2020). doi: 10.1093/bioinformatics/btz540; pmid: 31290943
64. K. Polovnikov, B. Slavov, S. Belan, M. Imakaev, H. B. Brandão,
L. A. Mirny, Crumpled polymer with loops recapitulates key
features of chromosome organization. bioRxiv 478588
[Preprint] (2023); https://doi.org/10.1101/2022.02.01.478588.
65. A. Roayaei Ardakany, H. T. Gezer, S. Lonardi, F. Ay, Mustache:
Multi-scale detection of chromatin loops from Hi-C and Micro-C
maps using scale-space representation. Genome Biol. 21, 256 (2020).
doi: 10.1186/s13059-020-02167-0; pmid: 32998764
66. A. E. Carpenter et al., CellProfiler: Image analysis software for
identifying and quantifying cell phenotypes. Genome Biol. 7,
R100–R111 (2006). doi: 10.1186/gb-2006-7-10-r100;
pmid: 17076895
p
67. L. Kiefer et al., WAPL functions as a rheostat of Protocadherin
isoform diversity that controls neural wiring (2023),
doi: 10.5281/zenodo.7613545
AC KNOWLED GME NTS
We thank T. Maniatis, S. Lomvardas, G. Narlikar, E. Nora, H. Madhani,
V. Ramani, M. Paredes, G. Mountoufaris, E. Cannavo, H. Shayya, and
members of the Canzio and Nora laboratories for helpful discussions and
suggestions; J.-M. Peters for the generous gift of the Wapl(fl/fl) mouse;
g
G. Barnea for the antibody for the olfactory receptor m71; the UCSF
Gladstone Mouse Transgenic Core for help in the generation of
transgenic animals; M. Bernardi and H.-R. Lin of the Gladstone
Genomics Core for assistance with bulk and single-cell sequencing
experiments; and S. Kim and the staff of the UCSF Center for Advanced
y
Light Microscopy for support with image acquisition and analysis.
Funding: This work was supported by the National Institute of General
Medicine (grant R00 GM121815-05 to D.C.); the National Institute of
Mental Health (grant DP2 MH129955-01 to D.C.); a CZ Biohub
Investigator grant (D.C.); and the National Science Foundation
(graduate research fellowship grant 2034836 to A.B.). Any opinions,
findings, and conclusions, or recommendations expressed in this
material are those of the authors and do not necessarily reflect the
views of the National Science Foundation. Authors contributions:
D.C. and L.K. conceived the project. L.K. designed, performed,
and interpreted the bulk of the genomic experiments. A.C. and A.B.
designed, performed, and interpreted the studies in 5-HT. J.L.
designed, performed, and interpreted the IHC studies in the OB. S.G.
y g
helped with the analysis and interpretation of the in situ Hi-C data.
G.I.F.S. and S.M.R. established the initial protocols to perform genomic
experiments in OSNs with help from E.S.C.. A.H. helped with the
establishment of the in situ Hi-C protocols. A.V. and A.J. assisted with
immunohistochemistry studies and mouse genotyping. M.H.M.
assisted with mouse colony management. D.C. supervised the project
and wrote the paper with substantial help from L.K., A.C., A.B., J.L.,
and S.G. Competing interests: The authors declare no competing
,
interests. Data and materials availability: All data are available
in the main text or the supplementary materials, and all reagents can
be made available from the corresponding author upon request.
Source data have been deposited in NCBI’s Gene Expression Omnibus
and are accessible through GEO series accession number GSE191195.
Custom scripts can be found at Zenodo (67). License information:
Copyright © 2023 the authors, some rights reserved; exclusive
licensee American Association for the Advancement of Science. No
claim to original US government works. https://www.science.org/
about/science-licenses-journal-article-reuse
SUPPLEMENTARY MATERIALS
science.org/doi/10.1126/science.adf8440
Figs. S1 to S5
Tables S1 to S3
MDAR Reproducibility Checklist
View/request a protocol for this paper from Bio-protocol.
Submitted 17 November 2022; accepted 7 May 2023
10.1126/science.adf8440
11 of 11
RES EARCH

◥ tions and functions, we hypothesized that the

RESEARCH ARTICLE SUMMARY partner of MFN2 on the ER could be a splice
variant of the MFN2 gene. We studied MFN2
CELL BIOLOGY splice variants dubbed ERMIN2 (ER mitofu-

Splice variants of mitofusin 2 shape the endoplasmic

sin 2) and ERMIT2 (ER mitofusin 2 tether)
that we found in human skeletal muscle and

reticulum and tether it to mitochondria

mouse fibroblasts.

RESULTS: ERMIN2 and ERMIT2 were expressed

Déborah Naón*, María Isabel Hernández-Alvarez, Satoko Shinjo, Milosz Wieczor, Saska Ivanova, in multiple human tissues and were induced
Olga Martins de Brito, Albert Quintana, Juan Hidalgo, Manuel Palacín, Pilar Aparicio, upon stresses that drive ER-mitochondria
Juan Castellanos, Luis Lores, David Sebastián, Sonia Fernández-Veledo, Joan Vendrell, Jorge Joven, proximity. We combined imaging and bio-
Modesto Orozco, Antonio Zorzano*†, Luca Scorrano*† chemical assays of subcellular localization and
found that the MFN2 variants were exclu-
sively at the ER. ERMIT2 and MoV-MFN2
INTRODUCTION: In the cytosol of eukaryotic drial fusion protein mitofusin 2 (MFN2) also were enriched at the ER-mitochondria inter-
cells, organelles are closely juxtaposed at mem- functions as a structural ER-mitochondria face with their N and C termini exposed to the
brane contact sites. Membrane contact sites tether, but its partner on the ER membrane is cytosol. Neither variant was able to correct the
provide hotspots for the metabolic and sig- unknown. disrupted morphology of Mfn2−/− mitochon-
naling cascades shared between the interact- dria. However, ERMIN2 was essential to shape
ing organelles. Protein bridges called tethers RATIONALE: Interactions between mitofusins the ER, and ERMIT2 and MoV-MFN2 were

stabilize membrane contact sites to generate
the microenvironment essential for the ex-
change of metabolites and second messengers.
The interaction between the endoplasmic reti-
culum (ER) and mitochondria at membrane
contact sites allows ER-to-mitochondria Ca2+
on separate mitochondria drive organelle fu-
sion. MFN2 can also support heterotypic organ-
elle interaction, as indicated by the ability of an
artificial ER-restricted MFN2 mutant to physi-
cally interact with mitofusins on mitochondria.
Because of this MFN2 propensity for homo-
p
essential to tether it to mitochondria. The coiled
coil (CC) domain of ER-located ERMIT2 inter-
acted with the CC regions of the mitochondrial
mitofusins, forming a complex that was energet-
ically stable enough to tether the two organelles
at least in molecular modeling simulations.

and phosphatidylserine (PS) transfer and serves
as a site of autophagy and mitochondrial fission
initiation. In mammalian cells, the mitochon-
typic interactions, and because mitochondrial
dynamics genes are alternatively spliced in
variants with divergent subcellular localiza-
g
Functionally, organelle tethering mediated by
ERMIT2 allowed mitochondrial uptake of ER-
released calcium in cells. Furthermore, organelle
tethering by ERMIT2 allowed for mitochondrial
uptake of ER-derived phospholipids in the

y
liver of genetic and diet-induced models of
Alternatively spliced, Mitofusin 2 (MFN2) gene reduced ER-mitochondria juxtaposition. This
ER-specific MFN2 var- Exons tethering ultimately helped to alleviate hepatic
iants ERMIN2 and 5' 3' ER stress and inflammation.
ERMIT2 expand MFN2 Transcription
function beyond the CONCLUSION: In this work, we identified al-
mitochondria. (Top) ternatively spliced, ER-restricted variants of
Alternative splicing of the MFN2 ERMIT2 ERMIN2 the single MFN2 gene. These variants acted as
human MFN2 gene results 5' 3' 5' 3' 5' 3' the molecular mediators of MFN2 extramito-
in two additional tran- chondrial functions in ER morphology and

y g
scripts coding for two tethering to mitochondria. We also identified
MFN2
proteins restricted at the the role of the tethering function of MFN2 in
ERMIT2 ERMIN2
ER. (Bottom left) ERMIT2 liver pathophysiology. In particular, in the
is enriched at the ER- PS
context of nonalcoholic steatohepatitis, ERMIT2
PS PS
mitochondria interface could correct the altered phospholipid transfer

and is the ER partner of
mitochondrial mitofusins
in tethering, allowing effi-
cient Ca2+ and phospho-
lipid transfer between the
organelles. (Bottom right)
ERMIN2 is not enriched in
mitochondria-associated
membranes (MAMs), and
it regulates ER morphol-
ogy. IMM, inner mitochon-
drial membrane; OMM,
outer mitochondrial
membrane; PE,
phosphatidylethanolamine.
,
PE from ER to mitochondria and the ensuing
hepatic distress. Thus, alternative splicing can
Ca2+
extend the function of the mitochondria-
shaping machinery beyond the mitochondria,
IMM OMM MAM and altered ER-mitochondria communication
plays a key role in the pathobiology of meta-
▪
Mitochondria ER ER
bolic liver diseases.
ER
The list of author affiliations is available in the full article online.
*Corresponding author. Email: luca.scorrano@unipd.it
(L.S.); antonio.zorzano@irbbarcelona.org (A.Z.);
deborah.naon@gmail.com (D.N.)
†These authors contributed equally to this work.
Cite this article as D. Naón et al., Science 380, eadh9351
ER-mitochondria (2023). DOI: 10.1126/science.adh9351
contact sites
ER-mit Ca2+ transfer READ THE FULL ARTICLE AT
ER stress
ER-mit PS transfer
https://doi.org/10.1126/science.adh9351

Naón et al., Science 380, 1237 (2023) 23 June 2023 1 of 1

RES EARCH

◥ We decided to look for the partner of MFN2

RESEARCH ARTICLE that enables ER-mitochondrial tethering. Given
this propensity of mitochondrial dynamics
CELL BIOLOGY genes to undergo alternative splicing in va-

Splice variants of mitofusin 2 shape the endoplasmic

riants with different subcellular localizations
and functions, and because of MFN2’s pro-

reticulum and tether it to mitochondria

pensity to engage in homotypic interactions
(33), we hypothesized that a MFN2 gene splice
variant could be the ER partner of MFN2.
Déborah Naón1,2,3*, María Isabel Hernández-Alvarez4,5,6, Satoko Shinjo1,2, Milosz Wieczor3,7,
Saska Ivanova3,4,5, Olga Martins de Brito2, Albert Quintana8,9, Juan Hidalgo8,9, Manuel Palacín3,4,10, Human and mouse mitofusin 2 gene are
Pilar Aparicio11, Juan Castellanos11, Luis Lores12, David Sebastián3,4,5, Sonia Fernández-Veledo5,13,14, alternatively spliced
Joan Vendrell5,13,14, Jorge Joven14,15, Modesto Orozco3,4, Antonio Zorzano3,4,5*†, Luca Scorrano1,2*† We amplified human skeletal muscle cDNAs
by polymerase chain reaction (PCR) using pri-
In eukaryotic cells, different organelles interact at membrane contact sites stabilized by tethers. mers within intron 2 and exon 19 of the ca-
Mitochondrial mitofusin 2 (MFN2) acts as a membrane tether that interacts with an unknown partner on nonical MFN2 DNA sequence. In addition to
the endoplasmic reticulum (ER). In this work, we identified the MFN2 splice variant ERMIT2 as the ER the known MFN2 sequence, we identified two
tethering partner of MFN2. Splicing of MFN2 produced ERMIT2 and ERMIN2, two ER-specific variants. shorter amplicons that we cloned and sequenced.
ERMIN2 regulated ER morphology, whereas ERMIT2 localized at the ER-mitochondria interface and The 1330–base pair (bp)–long mRNA of the first
interacted with mitochondrial mitofusins to tether ER and mitochondria. This tethering allowed efficient variant dubbed ERMIN2 (for ER mitofusin 2)
mitochondrial calcium ion uptake and phospholipid transfer. Expression of ERMIT2 ameliorated the resulted from alternative splicing between exons

p
ER stress, inflammation, and fibrosis typical of liver-specific Mfn2 knockout mice. Thus, ER-specific 3a and 3c and between exons 6a and 15b. The
MFN2 variants display entirely extramitochondrial MFN2 functions involved in interorganellar tethering mRNA of the second variant that we called
and liver metabolic activities. ERMIT2 (for ER mitofusin 2 tether) was 1220
bp long and resulted from alternative splicing

I
between exons 4a and 13b (Fig. 1A). Ribonu-
n eukaryotic cells, individual organelles (11). The ER-mitochondria interface is stabi- clease protection assay (RPA) confirmed that
the mRNAs of ERMIN2 and ERMIT2 were

interact at membrane contact sites (MCSs)—
areas of close juxtaposition that facilitate
the exchange of ions, metabolites, and lipids
(1, 2). The endoplasmic reticulum (ER)–
mitochondria juxtaposition epitomizes the
lized by 16- to 30-nm protein bridges between
the two organelles (12). The mitochondrial
fusion protein mitofusin (MFN) 2 also tethers
the two organelles and regulates communica-
tion between them (13–22). Several additional
g
expressed in HeLa cells (Fig. 1B). We next set
up real-time PCR assays to specifically detect
mRNAs of ERMIN2, ERMIT2, and MFN2 in
human samples (fig. S1A). We found that

importance of MCSs in cell biology. Mito-
chondria-ER MCSs are required for Ca2+
signaling (3, 4), apoptosis (5, 6), mitochondrial
fission (7), autophagosome formation (8, 9),
phosphatidylserine metabolism (10), and even
for nutrient sensing in hypothalamic neurons
1
Department of Biology, University of Padua, 35121 Padova,
ER-mitochondria tethers have also been iden-
tified, including the ERMES complex in yeast
(23) and the ER VAPB-mitochondrial PTPIP51
couple in mammals (24). These tethers are
commonly formed by a protein interacting
in trans with its partner on the opposing or-
ganelle. However, the molecular partner of
MFN2 on the ER that allows tethering without
heterotypic organelle fusion is unknown. This
uncertainty complicates our understanding of
y
ERMIN2 accounted for 20 to 25%, whereas
ERMIT2 accounted for 7 to 52% of MFN2
mRNA expression in white adipose tissue
(WAT), skeletal muscle, and liver from human
subjects (fig. S1B). We next monitored whether
specific exogenous stresses affected the expres-
sion of ERMIN2, ERMIT2, and MFN2. Their
mRNA levels were unchanged in starved HeLa
cells, whereas they increased in cells exposed
to the mitochondrial complex I inhibitor rote-

Italy. 2Veneto Institute of Molecular Medicine, 35129 Padova,
Italy. 3Institute for Research in Biomedicine (IRB Barcelona),
Barcelona, Spain. 4Departament de Bioquímica i Biomedicina
Molecular, Universitat de Barcelona, Barcelona, Spain.
5 (10, 25, 26). It is unclear whether Mfn2 de-
CIBER de Diabetes y Enfermedades Metabólicas Asociadas
(CIBERDEM), Madrid, Spain. 6IBUB, Universitat de Barcelona,
Barcelona, Spain. 7Department of Physical Chemistry,
Gdansk University of Technology, 80-233 Gdańsk, Poland.
the consequences of MFN2 deficiency, includ-
ing both mitochondrial and ER alterations
letion affects the ER because of its role in
tethering or because of disrupted mitochon-
y g
none. The sarcoplasmic ER Ca2+ adenosine
triphosphatase (SERCA) inhibitor thapsigargin
that causes ER stress induced the expression
of ERMIN2 and ERMIT2 but not of MFN2
(Fig. 1C). The well-characterized dimerization-

8
Institute of Neurosciences, Universitat Autònoma de
Barcelona, Bellaterra, Spain. 9Department of Cellular Biology,
Physiology and Immunology, Animal Physiology Unit, Faculty
of Biosciences, Universitat Autònoma de Barcelona,
Bellaterra, Spain. 10CIBER de Enfermedades Raras
(CIBERER), Instituto de Salud Carlos III, Madrid, Spain.
11
Department of Orthopaedics and Trauma Surgery, Parc
Sanitari Sant Joan de Déu, Sant Boi de Llobregat, Barcelona,
Spain. 12Pneumology Department, Hospital General Parc
Sanitari Sant Joan de Déu, Sant Boi de Llobregat, Barcelona,
Spain. 13Department of Endocrinology and Nutrition, Hospital
Universitari de Tarragona Joan XXIII, Institut d’Investigació
Sanitària Pere Virgili (IISPV), Tarragona, Spain. 14Medicine
School, Universitat Rovira i Virgili, Tarragona and Reus,
Spain. 15Unitat de Recerca Biomèdica, Institut d’Investigació
Sanitària Pere Virgili, Hospital Universitari de Sant Joan,
Reus, Spain.
*Corresponding author. Email: luca.scorrano@unipd.it (L.S.);
antonio.zorzano@irbbarcelona.org (A.Z.);
deborah.naon@gmail.com (D.N.)
†These authors contributed equally to this work.
drial dynamics that commonly disturbs the
ER (27–29). For example, in nonalcoholic stea-
tohepatitis models, the observed reduction in
MFN2 levels can affect mitochondrial dynam-
ics or ER-mitochondria communication—two
very different potential pathogenic mecha-
nisms (10). Knowledge of the ER partner of
MFN2 could help disentangle them.
Alternative splicing expands the genome
functional repertoire by generating multiple
proteins with different intracellular localiza-
tions and functions (30). For example, alter-
native splicing of the mitochondrial fission
executor dynamin related protein 1 gene en-
codes a brain-specific isoform that regulates
dendrite formation independently of mito-
chondrial fission (31, 32).
,
dependent green fluorescent protein (ddGFP)
ER-mitochondria proximity probe (21) reported
higher juxtaposition between these two organ-
elles in cells treated with rotenone and thap-
sigargin, which suggests that ERMIN2 and
ERMIT2 expression was induced when ER-
mitochondria proximity increased (fig. S1C).
The proteins encoded by ERMIN2 and
ERMIT2 were predicted to be shorter than
MFN2—partially or completely lacking its
guanosine triphosphatase (GTPase) and coiled-
coil 1 (CC1) domains but retaining its trans-
membrane (TM) and coiled-coil 2 (CC2) regions
(Fig. 1A and fig. S2). Exogenous hemagglutinin
(HA)–tagged ERMIN2 and ERMIT2 expressed
in HeLa cells were detected at around their
theoretical molecular masses—i.e., 41 and

Naón et al., Science 380, eadh9351 (2023) 23 June 2023 1 of 13

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

A B D Muscle Liver

1 2 3abc 4ab 5 6ab 7 8 91011 12 13ab 14 15ab161718 19 RPA 70 MFN2

MFN2

ERMIN2-
ORF

Probe
RNA ( g)
MFN2 G1G2 G3 G4 G5 CC1 TM TM CC2 ERMIT2
40 ERMIN2
0 2 4
2 3abc 4ab 5 6a 15b161718 19 Mfn2 (aa661-751)
V1-MFN2
(ERMIN2)
100
Vinculin

ERMIT2-
ERMIN2 G1G2 TM TM CC2
WAT

Probe
RNA ( g)
2 3ab 4a 13b 14 15ab1617 18 19
V2-MFN2 0 2 4
(ERMIT2) MFN2

ERMIT2 TM TM CC2

ERMIT2
ERMIN2
C MFN2 ERMIN2 ERMIT2

p
Mfn2 (aa661-751) long exposure
4000 * 70 35 *
*
3500 60 * 30 Vinculin
RNA levels (relative to PPIA)

*
6
3000 50 25

protein/vinculin (a.u.)

g
2500
40 20
4
2000
30 15
1500
20 10 2

y
1000

500 10 5
0

ERMIN2

ERMIN2

ERMIN2
MFN2

ERMIT2

MFN2

ERMIT2

MFN2

ERMIT2
0 0 0
TG
TG
TG

EBSS
EBSS
EBSS

Rot
Rot

control
control
Rot
control

Muscle Liver WAT

Fig. 1. MFN2 undergoes alternative splicing. (A) Schematic representation of cells. ERMIN2- and ERMIT2-specific probes covered the 5′- and 3′- flanking
the MFN2 splicing variants. Different PCR-amplification products obtained from sequences of the skipped region, and 20% of their length was a nonmatching

y g
human skeletal cDNA were cloned and sequenced. PCR was performed with sequence as positive control for RNAse digestion. (C) Means ± standard
primers located on intron 2 and exon 19. Exons are numbered 1 to 19, and the errors (SEs) of mRNA of MFN2, ERMIN2, and ERMIT2 levels (normalized to PPIA)
splicing events generate new exons (3a, 3b, 3c, 4a, 4b, 6a, 6b, 13a, 13b, 15a, in HeLa cells 4 hours after incubation in Earle’s buffered salt solution (EBSS)
and 15b). MFN2 corresponds to sequence NM014874. Variant 1 (V1-MFN2, (starvation), treatment with rotenone (Rot; 5 mM), or thapsigargin (TG; 1 mM).
ERMIN2) mRNA is 1330 bp long, is produced by exons 3b and 6b to 15a skipping, N = 8 independent experiments. *P < 0.05 in a one-way analysis of
and it encodes for ERMIN2, a 41-kDa (predicted molecular weight) protein. variance (ANOVA). (D) Equal amounts (40 mg) of lysates from the indicated

,
Variant 2 (V2-MFN2, ERMIT2) mRNA is 1220 bp long, is produced by exon 4b to human adult tissues were separated by SDS-PAGE and immunoblotted using
13a skipping, and it encodes ERMIT2, a 43-kDa (predicted M.W.) protein. Exons the indicated antibodies. The bottom graph reports means ± SEs of
involved in the alternative splicing process and the domains in the produced protein/vinculin ratios from N = 3 independent immunoblotting experiments.
proteins are indicated. G1 to G5 are GTPase domain motifs. (B) RPA in HeLa a.u., arbitrary units.

43 KDa—and, like full-length MFN2, were We next tested whether alternatively spliced, similar to ERMIT2 (fig. S3, A to C). Thus, human
recognized by antibodies raised against MFN2 shorter MFN2 versions were present in Mus and mouse Mitofusin 2 genes are alternatively
amino acids 557 to 576 or CC2 domain (fig. musculus, where MFN2 also tethers ER to spliced to generate proteins shorter than full-
S1D). Using the former antibody, we retrieved mitochondria. By amplifying cDNA from mouse length MFN2, expressed in multiple tissues
ERMIN2 and ERMIT2 running at their pre- embryonic fibroblasts (MEFs) using Mfn2 gene and induced when ER-mitochondria juxtaposi-
dicted molecular weights in human skeletal open reading frame exon 1 and 19 primers, we tion is pharmacologically increased.
muscle, liver, and WAT extracts. The ratio be- obtained and sequenced an amplicon corre-
tween ERMIN2 and MFN2 levels was 0.39 in sponding to a shorter Mfn2 transcript (fig. S3, A ERMIN2, ERMIT2, and MoV-MFN2 localize to
skeletal muscle, 1.8 in liver, and 0.17 in WAT, to C). This mouse Mfn2 variant (MoV-Mfn2) the ER
whereas the ERMIT2/MFN2 ratio was 0.3 to mRNA encoded for a protein containing the Given that ERMIN2 and ERMIT2 lack seve-
0.35 in all tissues analyzed (Fig. 1D). MFN2 TM and CC2 domains and hence is very ral domains found in full-length MFN2, we

Naón et al., Science 380, eadh9351 (2023) 23 June 2023 2 of 13

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

investigated whether they were targeted to
mitochondria. We first examined the distri-
bution of ERMIN2 or ERMIT2 in subcellular
fractions purified from Mfn2 knockout livers
(Mfn2LKO) where we had expressed the two
variants using adenovirus vectors. We detected
ERMIN2 and ERMIT2 in the ER-enriched
light membrane (LM) fraction as well as in
mitochondria-associated membranes (MAMs)
but not in the pure mitochondria fraction (Fig.
2A). We confirmed that endogenous ERMIN2
was present exclusively in LMs and ERMIT2
was present in LMs and MAMs but not in
the pure mitochondria fraction of HeLa cells,
which differed from full-length MFN2 that
was retrieved in all three fractions (13) (fig. S4).
We further tested the selective ER and MAMs
localization of ERMIN2 and ERMIT2 by
coexpressing GFP-tagged MFN2, ERMIN2,
and ERMIT2 with a mitochondrially targeted
cyan fluorescent protein (mtCFP) and a dsRED
fluorescent protein targeted to the ER (ER-RFP)
in Mfn2−/− MEFs. Although reexpressed MFN2
localized mostly in mitochondria and only
partially in the ER, ERMIN2 and ERMIT2 dis-
played a pattern like that of the ER marker.
Upon closer inspection, ERMIN2 was uniformly
distributed on the ER, and ERMIT2 displayed
a punctate staining that colocalized with ER
and mitochondria, compatible with the retrieval
of the endogenous protein in MAMs (Fig. 2,
C and D; individual channels in fig. S5). Simi-
larly, we found GFP-moV-MFN2 at the interface
between ER and mitochondria in Mfn2−/− MEFs
(fig. S3, D and E) and endogenous moV-MFN2
in MAMs and LM fractions purified from MEFs
(fig. S3F). Thus, ERMIN2 and ERMIT2 localize
only at the ER, and ERMIT2 and moV-MFN2
are enriched at the ER-mitochondria interface.
We next explored the determinants of
ERMIT2 ER targeting. By coexpressing GFP-
tagged ERMIT2 mutants with ER-RFP in
Mfn2−/− MEFs, we found that the ER localiza-
tion of ERMIT2 was not affected by deletion of
the CC2, the pre-CC2 region, or the N-terminal
domain (fig. S6, A to C). The low level of
mitochondrial colocalization of ERMIT2 was
unaffected in these deletion mutants (fig. S6,
D and E). Because these results suggested a
role for the TM region of ERMIT2/MFN2 as
an ER localization signal, we generated an
ERMIT2/MFN2 TM region–GFP chimera that
indeed localized at the ER (fig. S7, A and B).
Addition of CC1 and CC2 to this chimera re-
duced its ER localization and increased its
mitochondrial localization (fig. S7).
Topology of ERMIN2 and ERMIT2 at the ER
To determine the topology of ERMIN2 and
ERMIT2 at the ER, we devised a SPLIT-GFP
assay to detect the orientation of the N ter-
minus of ERMIN2 and ERMIT2. We coex-
pressed a nonfluorescent cytosolic GFP1-10

p
protein and chimeras of GFP11 b-strand fused
A C to the N terminus of MFN2, ERMIN2, or
Pure mito

GFP GFP-MFN2 GFP-ERMIN2 GFP-ERMIT2 ERMIT2. Fluorescence of the SPLIT-GFP sys-

MAMs
Total

GFP mito ER tem is reconstituted only if both GFP1-10 and

Mito

LM

the GFP11 chimeras are in the same cellular

40
compartment. GFP fluorescence was detected
ERMIN2

g
when we coexpressed GFP1-10 with GFP11-MFN2,
GFP11-ERMIN2, and GFP11-ERMIT2, which indi-
35 PS1 cates that the N termini of MFN2, ERMIN2, and
25
ERMIT2 face the cytosol (fig. S8A). To under-
25 stand whether the C termini of ERMIN2 and

y
TOM20
15 ERMIT2 were exposed to the lumen of the ER,
we performed limited trypsin proteolysis assays
140
SERCA2 on LM fractions purified from Mfn2LKO livers
100 infected with adenoviruses expressing ERMIN2
and ERMIT2. If ERMIN2 and ERMIT2 C
B D GFP GFP-MFN2 GFP-ERMIN2 GFP-ERMIT2
Pure mito

1.1 termini were exposed to the lumen, trypsin

* * *
Colocalization with organelle
MAMs

(Manders' coefficient, a.u.)

1.0 would eliminate the epitopes recognized by

Total
Mito

LM

the antibody against their N terminus, whereas

0.9
the CC2 epitope would be protected, and vice
ERMIT2 0.8
40 versa if the N terminus was facing the ER

y g
0.7 lumen (fig. S8B). Both the N-terminal and the
100 0.6 CC2 epitopes were lost upon trypsin treat-
70 FACL4 0.5 ment, indicating that both the N terminus and
0.4 the CC2 domain of ERMIN2 and ERMIT2
25 0.3 were accessible to the protease (fig. S8C).
TOM20

,
15 0.2 Thus, ERMIN2 and ERMIT2 N and C termini
0.1 face the cytosol (fig. S8D).
140
SERCA2 0.0
100 Mito ER Mito ER Mito ER Mito ER ERMIN2 regulates ER morphology
To functionalize ERMIN2 and ERMIT2, we
Fig. 2. MFN2 variants localize at the ER, not the mitochondria. (A and B) Mfn2LKO mice were injected first analyzed mitochondrial and ER morphol-
with adenoviruses encoding ERMIN2 (A) or ERMIT2 (B), and after 72 hours, liver subcellular fractions were ogy in Mfn2 −/− MEFs reconstituted with
prepared. Equal amounts (40 mg) of protein from total extracts (Total), crude (Mito), pure mitochondria MFN2, ERMIN2, and ERMIT2. Inspection
(Pure mito), MAMs, and LMs were separated by SDS-PAGE and immunoblotted using antibodies against of volume-rendered three-dimensional (3D)
MFN2, the MAM markers PS1 (presenilin 1) and FACL4 (long-chain acyl-CoA synthetase 4), the outer reconstructions of z-stacks of confocal images
mitochondrial membrane marker TOM20 (translocase of the outer mitochondrial membrane 20), and the of mtRFP revealed that, in contrast to MFN2,
ER marker SERCA2 (sarcoplasmic ER Ca2+ ATPase). (C) Representative confocal images of Mfn2−/− MEFs ERMIN2 or ERMIT2 did not rescue Mfn2−/−
cotransfected with the indicated plasmids (green), mitochondrially targeted CFP (mito; blue) and ER-dsRED mitochondrial morphology (Fig. 3, A and B).
(ER; red). Insets are magnified 7×. Scale bars, 10 mm. (D) Averages ± SEMs of Manders’ coefficient of Expression of MFN2 and ERMIN2 corrected
mitochondria-ER pseudocolocalization. Dots indicate individual cells (n = 150) from N = 3 independent the ER morphology of Mfn2−/− cells (Fig. 3C).
experiments. Mfn2−/− MEFs were cotransfected with ER-dsRED or mito-dsRED and the indicated Fluorescent recovery after photobleaching
GFP-tagged constructs. *P < 0.05 in a one-way ANOVA with Tukey’s mean comparison. (FRAP) assays further corroborated the finding

Naón et al., Science 380, eadh9351 (2023) 23 June 2023 3 of 13

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

that expression of MFN2 and ERMIN2 restored ERMIT2 do not modulate mitochondrial mor- reconstructions of z-axis stacks of confocal
ER connectivity in Mfn2−/− MEFs. Although phology, and endogenous ERMIN2 is required images of mtRFP and ER–yellow fluorescent
ERMIT2 slightly ameliorated ER morphology, for ER connectivity even when MFN2 is present. protein (YFP). Reexpression of MFN2 as well
it did not improve ER connectivity (Fig. 3, C as of ERMIT2 in Mfn2−/− MEFs improved ER-
and D). We then turned to a loss-of-function ERMIT2 and moV-MFN2 tether ER mitochondria pseudocolocalization. Conversely,
approach using small interfering RNAs (siRNAs) to mitochondria ERMIN2 that restored ER morphology did
to specifically down-regulate ERMIN2 or ERMIT2 Given that ERMIT2 is found in MAMs and not modify ER-mitochondria juxtaposition
in HeLa cells (fig. S9). Depletion of ERMIN2 or does not regulate ER or mitochondrial mor- (Fig. 4, A and B; individual channels in fig. S10).
ERMIT2 did not affect mitochondrial morphol- phology, we posited that it participated in Second, the fluorescence resonance energy
ogy (Fig. 3, E and F). Conversely, ERMIN2 down- ER-mitochondria tethering (13). We performed transfer (FRET)–based ER-mitochondria proxi-
regulation altered peripheral ER morphology and several orthogonal assays to verify this hypoth- mity probe (FEMP) (16) showed that expression
diminished ER interconnectivity, as determined esis. First, we evaluated juxtaposition between of MFN2 and ERMIT2 in Mfn2−/− MEFs in-
by FRAP analysis (Fig. 3G). Thus, ERMIN2 and the two organelles in volume-rendered 3D creased steady-state (basal) and maximal

A EV MFN2 B
110
C EV MFN2

Cells with fragmented mitochondria (% of total)

100
90
80

p
70
60
50
ERMIN2 ERMIT2
ERMIT 40
ERMIN2 ERMIT2
30
*

g
20
10
0

y
MFN2
ERMIN2
ERMIT2
EV

D E F G
siRNA Scr 12
110
110
ER-YFP fluorescence (% of initial value)

Mitochondria ER
ER-YFP fluorescence (% of initial value)
100 100
Mitochondria Aspect Ratio (a.u.)

10
90 90
80 80
8
70 siRNA ERMIN2 70

y g
60 60
6
50 50
40 40
4
30 30 siRNA:
EV

,
siRNA ERMIT2 20
Scr
20 MFN2 ERMIN2
2
ERMIN2 10 ERMIT2
10
ERMIT2
0 0
0 5 10 15 20 25 30 0 0 5 10 15 20 25 30
Time (sec) siRNA: Scr ERMIN2ERMIT2 Time (sec)

Fig. 3. ERMIN2 regulates ER morphology. (A) Representative volume-rendered at 100% laser power, and then recording continued for the duration of the
3D reconstructions of confocal z-stacks of Mfn2−/− MEFs cotransfected with the experiments. Data are means ± SEs of three independent experiments (12 cells
indicated HA-tagged constructs and mt-dsRED. Scale bars, 10 mm. (B) Means ± SEs per condition). (E) Representative Volume Viewer 3D reconstructions of confocal
of mitochondrial fragmentation (N = 3 independent experiments, n = 50 cells z-stack images of HeLa cells cotransfected with mt-dsRED, ER-YFP, and the
per condition) in experiments as in (A). *P < 0.05 in a one-way ANOVA with indicated siRNA. Scale bars, 10 mm. (F) Means ± SEs of mitochondrial aspect
Tukey’s mean comparison among EV and the other conditions. (C) Experiments ratio (N = 3 independent experiments, n = 27 cells per condition) in experiments as
were performed as in (A), except that Mfn2−/− MEFs were cotransfected with in (E). (G) ER-YFP fluorescence recordings from ER-YFP FRAP experiments
ER-YFP. Scale bars, 10 mm. (D) ER-YFP fluorescence recordings from FRAP in HeLa cells cotransfected with the ER-YFP and the indicated siRNA. Experiments
experiments in Mfn2−/− MEFs cotransfected with ER-YFP and the indicated were performed as in (D). Data are means ± SEs from three independent
HA-tagged plasmid. After 1 s, the ROI was photobleached by illumination experiments (12 cells per condition).

Naón et al., Science 380, eadh9351 (2023) 23 June 2023 4 of 13

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

A Mfn2-/- B C the fraction of Mfn2−/− ddGFP+ cells, whereas

FEMP FRET Ratio (FRET/CFP)

0.8 * 2.0 *

Mitochondria-ER colocalization
+EV +MFN2 * * * * expression of MFN2 or ERMIT2 led to a 2- and

(Manders’ coefficient, a.u.)

0.7 1.9 1.75-fold increase in ddGFP positivity (Fig. 4, D
0.6
1.8 and E). Live confocal imaging confirmed that
0.5 expression of MFN2 and ERMIT2, but not of
mito ER 0.4
1.7
ERMIN2, in Mfn2−/− cells led to a punctate
1.6 ddGFP signal, reminiscent of ER-mitochondria
+ERMIN2 +ERMIT2 0.3
contacts (fig. S11). Fourth, we ultrastructurally
1.5
0.2 analyzed ER-mitochondria contacts in trans-
0.1 1.4 mission electron microscopy images from
0.0 BM BM BM BM where we calculated the ER-mitochondria
EV MFN2 ERMIN2 ERMIT2 EV MFN2ERMIN2ERMIT2 contact coefficient (ERMICC) that computes
D E the average ER-mitochondria distance and
102 103 104 105

105

EV MFN2
10 * interaction length over the mitochondrial
GFP FITC-A

104

ddGFP+ cells (% of mCherry+ cells)

9 perimeter (16). MFN2 and ERMIT2, but not
ERMIN2, increased ER-mitochondria juxta-
103

8
* * position and ERMICC in Mfn2−/− cells (Fig. 4,
7
102

F and G). MFN2 and ERMIT2 decreased the

6 average ER-mitochondria distance measured
50 100 150 200 250 50 100 150 200 250
SSC-A (x1,000) SSC-A (x1,000) 5 from >70 interorganellar interactions occurring
in the ≤30-nm range in Mfn2−/− cells (fig. S12). In
105
102 103 104 105

ERMIN2 ERMIT2 4

p
parallel, we analyzed whether moV-MFN2, which
GFP FITC-A

3
104

resembles ERMIT2, was also involved in ER-

2
mitochondria tethering. Expression of moV-MFN2
103

1 in Mfn2−/− cells increased ER-mitochondria

102

0 juxtaposition, as reflected by increased pseudo-

EV MFN2 ERMIN2ERMIT2
localization of fluorescent proteins targeted to
50 100 150 200 250 50 100 150 200 250

g
SSC-A (x1,000) SSC-A (x1,000) ER and mitochondria (fig. S2, G and H). Thus,
F EV MFN2 G
0.020 * * *
expression of ERMIT2 and moV-MFN2 increases
ER-mitochondria juxtaposition.
ERMICC (A.U.)

0.016 ERMIT2 requires mitofusins on mitochondria

y
to tether them to the ER
0.012 A feature of heterotypic tethers is their inter-
action in trans with homo- or heterotypic
ERMIN2 ERMIT2 0.008 partners on the surface of the opposite organ-
elle. Given the propensity of MFNs to engage in
0.004
homo- and hetero-oligomerization (34), we
0.000 tested whether mitochondrial MFNs were
the mitochondrial partners of ERMIT2. First,
EV MFN2 ERMIN2ERMIT2 we addressed whether ERMIT2 (and ERMIN2)
were able to interact with MFN1 and/or MFN2

y g
Fig. 4. ERMIT2 reconstitutes ER-mitochondria interaction in Mfn2−/− MEFs. (A) Representative volume- in pull-down experiments. GFP-tagged ERMIN2
rendered 3D reconstructions of confocal z-stacks of Mfn2−/− MEFs cotransfected with ER-YFP, mt-dsRED, and or ERMIT2 immunoprecipitated V5-tagged
plasmids coding for the indicated HA-tagged proteins. Yellow indicates pseudocolocalization of the two MFN2ActA and Flag-tagged MFN1 that are re-
organelles. (B) Means ± SEs of Manders’ coefficient (n = 3 independent experiments, 30 cells per condition stricted to mitochondria (13), although ERMIN2
per experiment; open dots) of ER-mitochondria pseudocolocalization in experiments as in (A). *P < 0.05 in a immunoprecipitated these mitochondrial Mfns
one-way ANOVA with Tukey’s mean comparison among the indicated groups. (C) Means ± SEs of basal

,
somewhat less efficiently (Fig. 5A). Next, we
(B) and maximal (M) FEMP FRET values (n = 6 independent experiments, >200 cells per condition per used genetics to verify whether this biochemical
experiment; open dots). *P < 0.05 in a one-way ANOVA with Tukey’s means comparison between the interaction was functionally relevant for ER-
indicated groups. (D) Dot plots of ddGFP fluorescence versus SSC in the mCherry+ gated population from mitochondria tethering. We expressed MFN2,
flow cytometric analysis of Mfn2−/− MEFs cotransfected for 48 hours with ddGFP A-B monomers, cytosolic ERMIN2, and ERMIT2 in MEFs lacking Mfn1
mCherry, and the indicated plasmids. (E) Means ± SEs of the percentage of ddGFP+ cells (n = 6 independent and Mfn2 (Mfn1, Mfn2−/−). MFN2 and ERMIT2
experiments; open dots) in experiments as in (D). *P < 0.05 in a one-way ANOVA with Tukey’s means were unable to increase ER-mitochondria
comparison between the indicated groups. (F) Representative electron microscopy images of Mfn2−/− MEFs tethering unless coexpressed with MFN1 that
cotransfected with GFP and the indicated plasmid. GFP-positive cells were sorted, fixed, processed, and is found only on the surface of mitochondria
analyzed by transmission electron microscopy (TEM). Scale bars, 500 nm. (G) Means ± SEs of ERMICC (13) (Fig. 5, B and C). Thus, ER ERMIT2
(n = 50 to 182 measurements per condition; open dots) from N = 3 independent experiments as in (A). interacts with mitochondrial MFN1 for ER-
*P < 0.05 in a one-way ANOVA with Tukey’s means comparison between the indicated groups. mitochondria tethering.

ER-mitochondria juxtaposition (obtained by Third, we quantified by flow cytometry the per- Coiled-coil domains mediate ERMIN2/
a brief rapamycin pulse to engage the FEMP centage of Mfn2−/− cells positive to the ddGFP ERMIT2-mitofusin interaction
FKBP-FRB dimerization motif), whereas ERMIN2 ER-mitochondria proximity probe (16, 21). Al- We generated multiple deletion mutants of
increased only basal FEMP signal (Fig. 4C). though ERMIN2 did not significantly increase MFN1, MFN2, ERMIN2, and ERMIT2 to address

Naón et al., Science 380, eadh9351 (2023) 23 June 2023 5 of 13

RES EARCH | R E S E A R C H A R T I C L E
A B
Mfn1-/-,Mfn2-/-
+EV
FLAG-MFN1
ER
GFP-ERMIN2
GFP-ERMIT2
GFP
mito
IP: anti GFP
GFP
merge
FLAG
V5-MFN2ActA
GFP-ERMIN2
GFP-ERMIT2
GFP
IP: anti GFP
GFP
V5
Fig. 5. ERMIT2 interacts with mitochondrial mitofusins, tethering ER to
mitochondria. (A) Mfn1−/−, Mfn2−/− MEFs were cotransfected with the indicated
plasmids and after 24 hours lysed. Equal amounts (400 mg) of protein were
immunoprecipitated (IP) using the indicated antibodies, and immunoprecipitates
were separated by SDS-PAGE and immunoblotted using the indicated antibodies.
(B) Representative volume-rendered 3D reconstructions of confocal z-stacks of
which domains were required for their hetero-
typic interaction. A MFN2 mutant containing the
CC1, TM, and CC2 domains, as well as MFN2
mutants alternatively containing the CC1 or
the CC2 domain, immunoprecipitated ERMIN2
and ERMIT2 (Fig. 6A). If we added the GTPase
domain to the MFN1 and MFN2 mutants
containing only the CC1 region, they compara-
bly immunoprecipitated ERMIN2 and ERMIT2
(Fig. 6B). Thus, MFN1 and MFN2 interact with
ERMIN2 or ERMIT2 through their CC1 or CC2
domains. Reciprocally, an ERMIT2 mutant
lacking the pre-TM domain still immunopre-
cipitated MFN1, whereas upon deletion of its
CC2 domain, MFN1 binding was lost (Fig.
6C). ERMIT2 and ERMIN2 also interacted in
HeLa cells as well as in MEFs lacking Mfn1,
Mfn2, or both mitofusins (fig. S13A). We further
tested the emerging interaction paradigm be-
tween the C-terminal domains of ERMIT2 or
ERMIN2 and the CC1 or CC2 domains of
mitochondrial mitofusins in an in silico struc-
tural modeling analysis. Guided by the Alpha-
Fold 2 predicted structures of Mfn1 and ERMIT2,
the available MFN1 CC2 homodimer crystal
structure (35, 36), the experimentally determined
Naón et al., Science 380, eadh9351 (2023) 23 June 2023
Mfn1-/-,Mfn2-/-
+MFN2 +ERMIN2 +ERMIT2
C *
Mitochondria-ER colocalization
(Manders' coefficient, a.u.) 0.6
0.5
0.4
0.3
0.2
0.1
0.0
EV MFN1 EV MFN1 EV MFN1 EV MFN1
EV
dimerization interface, and sequence alignment,
we created a model of the MFN2-ERMIT2 com-
plex where the CC1 domain of MFN2 coiled
around the CC2 domain of ERMIT2. Using in
silico mutagenesis, we prepared an identical
model of the Mfn1-ERMIT2 complex. We verified
the stability of the atomistic models anchored to
model ER and mitochondrial membranes by
relaxing the 1.2M-atom systems in a long mo-
lecular dynamics simulation in explicit solvent.
The number of contacts between residues of
Mfn1 CC1 and ERMIT2 CC2 was consistently
stable, which suggests that this interaction can
tether ER and mitochondria even at a surface-
to-surface distance of 20 nm (Fig. 6D). A contact-
based affinity prediction performed on an
isolated CC1-CC2 pair yielded almost identi-
cal stability for the Mfn1-ERMIT2 and the
MFN2-ERMIT2 complexes (estimated affinity:
MFN2-ERMIT2, −12.5 kcal/mol; Mfn1-ERMIT2,
−12.6 kcal/mol; fig. S13, B and C). Thus, this
simulation shows that the affinity of these
complexes is well within the range required
for membrane tethering, which suggests how
ERMIT2 can tether ER to mitochondria through
MFN1 in the absence of MFN2.
Mfn1-/-,Mfn2-/- +MFN1
+EV +MFN2 +ERMIN2 +ERMIT2
+EV
p
*
g
MFN2 ERMIN2 ERMIT2
Mfn1−/−, Mfn2−/− MEFs cotransfected with ER-YFP, mt-dsRED, and plasmids coding
y
for the indicated HA-tagged proteins. Where indicated (+MFN1), MEFs were also
cotransfected with MFN1. (C) Means ± SEs of Manders’ coefficient (n = 3 independent
experiments, 15 cells per condition per experiment) (open dots, EV; closed dots,
+MFN1) of ER-mitochondria pseudocolocalization in experiments as in (B). *P < 0.05
in a one-way ANOVA with Tukey’s mean comparison among the indicated groups.
ERMIT2 licenses Ca2+ and phospholipid
transfer from ER to mitochondria
We finally verified whether ERMIT2 like
MFN2 influenced two facets of ER-mitochondria
y g
communication: mitochondrial uptake of ER-
released Ca2+ (37, 38) and lipid transfer (2, 10, 39).
MFN2, ERMIN2, or ERMIT2 lowered steady-
state ER Ca2+ levels measured in empty vector
(EV) transfected Mfn2−/− MEFs (13, 16) (fig. S14, A
,
and B). We titrated the inositol triphosphate (IP3)–
generating agonist adenosine 5′-triphosphate
(ATP) to induce the same ER Ca2+ release in
Mfn2−/− MEFs transfected with all the vectors
used in this study (40, 41) (fig. S14, C and D). Upon
equalized ER Ca2+ release, MFN2 and ERMIT2,
but not ERMIN2, increased mitochondrial
Ca2+ uptake compared with EV (Fig. 7, A and
B). A genetically encoded ATP probe targeted
to the mitochondrial matrix (3) indicated
that MFN2 and ERMIT2 expression increased
the Ca2+ uptake–dependent priming of mito-
chondrial ATP generation (16, 42) (Fig. 7,
C and D). Thus, ERMIT2 expression in
Mfn2−/− cells licenses mitochondrial uptake
of ER-released Ca2+ and Ca2+-primed mito-
chondrial ATP production that depends on
6 of 13
RES EARCH | R E S E A R C H A R T I C L E

A CC1 TMTM CC2 TMTM CC2 CC1 TMTM B GTPase CC1 TMTM
GFP-MFN2390-757 GFP-MFN2611-757 GFP-MFN2390-647 His-MFN1Δ630-741 GFP-MFN2Δ647-757
HA + + + HA + +
HA-ERMIN2 + + + HA-ERMIN2 + +
HA-ERMIT2 + + + HA-ERMIT2 + +
GFP His GFP

HA HA HA

IP: anti GFP IP : anti His IP: anti GFP

C D
Flag-MFN1
GFP +

TMTM CC2 GFP-ERMIT2 +

p
TMTM GFP-ERMIT2Δ647-738 +

TMTM CC2 GFP-ERMIT2Δ549-610 +

g
GFP

y
Flag

IP: anti GFP

Fig. 6. Ermit2-mitofusins interact via their coiled-coil domains. antibodies. (D) Structural modeling of the interaction between a truncated
(A to C) HeLa cells were cotransfected with plasmids encoding for the version of MFN1 and ERMIT2. The initial knowledge-based model was relaxed

indicated constructs and after 24 hours lysed. Schemes depict the domains
found in the transfected chimeras. Equal amounts (400 mg) of protein were
immunoprecipitated (IP) using the indicated antibodies, and immunoprecipitates
were separated by SDS-PAGE and immunoblotted using the indicated
y g
in a long, fully atomistic molecular dynamics simulation, revealing one of
the possible arrangements of the GTPase domain. The magnified region
shows the interface between MFN1 (green) CC1 and ERMIT2 (orange)
CC2 domain.

matrix Ca2+-activated Krebs’ cycle dehydro-

genases (43).
In the liver, phosphatidylserine (PS) is
synthesized in the ER and transferred to
mitochondria where it is then converted to
phosphatidylethanolamine (PE). In Mfn2LKO
hepatocytes, PS synthesis and transfer to mito-
chondria are impaired, resulting in ER stress
and inflammation (10). We thus measured
whether ERMIT2 expression normalized lipid
transfer between ER and mitochondria. In
control (LacZ) infected Mfn2LKO mice, hepatic
incorporation of L-serine (L-Ser) into PS and
PE was reduced. When we infected Mfn2LKO
livers with ERMIT2 adenoviruses (fig. S15A),
L-Ser incorporation into PE and PS was corrected
(Fig. 7, E and F). In addition, in the Mfn2LKO
cohort (10), ERMIT2 reduced the accumulation
of H2O2 (fig. S15B), reduced the levels of ER stress
markers phosphorylated eIF2a and CHOP (fig.
S15, C and D), and reduced the levels of the
inflammation marker tumor necrosis factor–a
(TNF-a) (fig. S15E). Thus, ERMIT2 sustains ER
to mitochondria lipid transfer and corrects the
liver ER stress and inflammation observed in
the Mfn2LKO mice. Mice fed a methionine-
choline–deficient (MCD) high-fat diet develop
a nonalcoholic steatohepatitis (NASH)–like con-
dition, an orthogonal model of liver ER stress
and inflammation compared with MFN2 defi-
ciency (10) (fig. S16A). In the MCD-fed mice, we
observed altered hepatic L-Ser labeling of PS
and PE (Fig. 7, G and H). Adenoviruses en-
coding ERMIT2 delivered at day 7 to mice
,
undergoing a 3-week MCD feeding protocol
increased ERMIT2 but not MFN2 levels (fig. S16,
A and B). ERMIT2 delivery corrected L-Ser label-
ing of PE and PS (Fig. 7, G and H) and reduced
the accumulation of markers of ER stress (fig.
S16, C and D) and inflammation (fig. S16E) with-
out altering body weight or glycemia (fig. S15, F
and G). Thus, ERMIT2 sustains ER to mito-
chondria lipid transfer and corrects the liver ER
stress and inflammation in genetic and envi-
ronmental mouse models of reduced liver Mfn2
expression, highlighting the importance of ER-
mitochondria juxtaposition in models of NASH.
Discussion
The mechanism by which MFN2 tethers ER to
mitochondria has been elusive. In contrast to

Naón et al., Science 380, eadh9351 (2023) 23 June 2023 7 of 13

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

A ATP B C ATP D
*

mt-Luc luminescence (% of initial value)

14 125
* *

mt-Luc luminescence (AUC x103)

12
EV 500 * 7.25 * *
MFN2 120
450 *
ERMIN2
10 400

[Ca2+]mit AUC (mM)

ERMIT2 7.00
115
[Ca2+]mit (mM)

8 350
300 110 6.75
6 250
4 200 105 6.50
150 EV
2 100 100 MFN2 6.25
ERMIN2
0 50 ERMIT2
0 95 6.00
0 10 20 30 40 50 60 70 EV MFN2 ERMIN2 ERMIT2 0 10 20 30 40 50 60 70 EV MFN2 ERMIN2ERMIT2
Time (s) Time (s)

E F G H
* * * * * * * *

H-L-Ser in PS (DPM x105/mg protein)

H-L-Ser in PE (DPM x104/mg protein)

300 1.75 3.5
3000
H-L-Ser in PS (DPM/mg protein)

H-L-Ser in PE (DPM/mg protein)

p
250 1.50
2500 3.0
200 1.25
2000
2.5
1.00
1500 150
0.75 2.0

g
1000 100
0.50
1.5
500 50
0.25

0 0 0.00 1.0

y
Chow MCD Chow MCD
Mfn2f/f Mfn2LKO Mfn2f/f Mfn2LKO Mfn2f/f Mfn2LKO Mfn2f/f Mfn2LKO
LacZ ERMIT2 LacZ ERMIT2 LacZ ERMIT2 LacZ ERMIT2
3

Fig. 7. ERMIT2 licenses ER-mitochondria Ca2+ and lipid transfer in vitro
and in vivo. (A) Representative mitochondrial aequorin (mt-AEQ) measurements
of [Ca2+]mit in response to ATP (0.2 mM) in Mfn2−/− MEFs cotransfected with
the indicated plasmid and mt-AEQ. Averages ± SEMs peak mitochondrial Ca2+:
EV = 3.04 ± 0.27, N = 5; MFN2 = 9.30 ± 1.14, N = 8; ERMIN2 = 5.58 ± 0.63,
N = 7; ERMIT2 = 10.79 ± 1.42, N = 6. P < 0.05 in a one-way ANOVA with Tukey’s
test for the EV-MFN2, EV-ERMIT2, and ERMIT2-ERMIN2 pairs. (B) Means ± SEs of
3
experiments as in (D) (n = 10 independent experiments). *P < 0.05 in a one-way
ANOVA with Tukey’s mean comparison among the indicated conditions. (E and
F) Five days after tail-vein injection of adenoviruses encoding LacZ or ERMIT2 in
Mfn2f/f or Mfn2LKO mice, livers were explanted and means ± SEMs 3H–L-Ser
incorporation into PS (E) and PE (F) in hepatic mitochondria-associated ER-enriched
fractions (n = 3 to 6 mice per condition; open dots) were measured. *P < 0.05
in a one-way ANOVA with Tukey’s mean comparison among the indicated conditions.

y g
the area under the curve (AUC) in experiments as in (A) (n = 10 independent (G and H) Wild-type mice were fed with chow or MCD for 3 weeks. On day 7, mice
experiments). *P < 0.05 in a one-way ANOVA with Tukey’s mean comparison were tail-vein injected with adenoviruses encoding LacZ or ERMIT2 (n = 7 to 11 mice
among the indicated conditions. (C) Representative traces of mitochondrial per group). At the end of the diet protocol, livers were explanted and means ± SEMs
ATP measurements in Mfn2−/− MEFs cotransfected with mt-luciferase and the 3
H–L-Ser incorporation into PS (G) and PE (H) in hepatic mitochondria-associated
indicated plasmids. Where indicated, cells were perfused with ATP (0.2 mM) ER-enriched fractions were measured. *P < 0.05 in a one-way ANOVA with Tukey’s

,
to initiate IP3-mediated ER Ca2+ release. (D) Means ± SEs of the AUC in mean comparison among the indicated conditions. DPM, disintegrations per minute.

other tether pairs or complexes (23, 44, 45), the mitochondrial outer membrane proteins fusion like the ER-specific GTPase atlastin
MFN2 lacked a known ER partner, and its ER- BAK (5), BCL-2 (47), and cytochrome b5 (48), (49). ERMIT2 tethers the ER to mitochondria
mitochondria tethering function was explained can also localize at the ER, the model of ER by interacting with MFN1 and/or MFN2 on
by its retrieval also on the ER, from where it sculpting and tethering in trans by a “mis- the mitochondrial outer membrane. The struc-
interacted in trans with mitochondrial MFN2 localized” fraction of MFN2 did not explain tural organization of ERMIT2 seems adequate
or MFN1 (13, 16). Our results identify a MFN2 how a fusion GTPase could participate in mito- to support this function: ERMIT2 lacks the
splice variant enriched at the ER interface chondrial and ER fusion and at the same time GTPase domain essential for mitochondrial
with mitochondria as the ER tethering partner tether ER and mitochondria without fusing fusion (34) but retains the CC2 domain (33)
of MFN2. them. The identification of ERMIN2 and ERMIT2, that is required for the interaction with CC1
The identification of extramitochondrial two alternatively spliced, ER-restricted MFN2 or CC2 of mitochondrial mitofusins. In silico
MFN2 splice variants can explain the pleiotropic variants, contributes to clarifying these issues. modeling supports this paradigm and high-
effects of Mfn2 ablation on mitochondrial and ERMIN2 that retains part of the MFN2 GTPase lights that the ERMIT2-MFN1/2 interaction
ER morphology and on ER-mitochondria tether- domain restores Mfn2−/− ER connectivity, which can indeed stabilize the juxtaposition of the
ing (11, 13, 15, 16, 22, 26, 46). Although MFN2, like suggests that this variant can participate in ER two organelles.

Naón et al., Science 380, eadh9351 (2023) 23 June 2023 8 of 13

RES EARCH | R E S E A R C H A R T I C L E
Functionally, ERMIT2 corrected the ER-
mitochondria communication defects caused
by reduced MFN2 levels. ERMIT2 was suffi-
cient to license mitochondrial uptake of Ca2+
released from the ER when we equalized the
amount of agonist-induced ER Ca2+ release.
We surmise that ERMIT2 operates by allow-
ing the formation of the high Ca2+ micro-
domains at the interface between the two
organelles required for efficient mitochondrial
Ca2+ uptake (37, 38, 50). In addition, ERMIN2
and ERMIT2 also normalized ER Ca2+ in
Mfn2-deficient cells by modulating the facets
of Ca2+ signaling altered by Mfn2 deletion:
capacitative Ca2+ entry, SERCA activity, or ER
Ca2+ leak (40, 41).
In vivo ERMIT2 delivery to the liver correct-
ed most of the hepatic alterations associated
with Mfn2 genetic ablation or depletion in a
diet-induced model of NASH. ERMIT normal-
ized PS transfer from the ER to mitochondria,
a process linked to alterations in lipid metab-
olism, ER stress, inflammation, and fibrosis
(10), which suggests that it participates in PS
transfer to mitochondria and in document-
ing the importance of MFN2-mediated ER-
mitochondria tethering in NASH.
Several pathogenic MFN2 mutations causing
defects in axonal trafficking of mitochondria
and resulting in the peripheral neuropathy
Charcot-Marie-Tooth (CMT) 2A (51–53) affect
the CC2 region found also in the ER-specific
variants (54). Thus, altered ER-mitochondria
tethering or ER morphology might be a common
feature of CMT2A and hereditary sensory
neuropathy type I caused by mutations in
atlastin that control ER fusion (49, 55) and
restored by the small molecules normalizing
CMT2A neuromuscular dysfunction (52).
Overexpression of MFN2 in acute myeloid leu-
kemia (AML) causes excessive ER-mitochondria
tethering that contributes to venetoclax resist-
ance, reversed by a small molecule binding the
CC1-CC2 MFN2 interface and reducing ER-
mitochondria juxtaposition (56). Thus, the dis-
covery of MFN2 ER-specific variants offers
pathobiological and therapeutic insights into
metabolic and neuromuscular diseases and
cancer. Our work elucidates the importance
of alternative splicing of mitochondrial dynam-
ics genes for their function (31, 57, 58) and
subcellular localization (32, 59) and suggests
that alternative splicing provides a mechanism
for how a single mitochondrial gene can modu-
late manifold biological processes.
Materials and methods
Molecular biology and real-time PCR
For ERMIN2 and ERMIT2 amplification, PCRs
were performed using human skeletal muscle
cDNA and the following primers: (i) Mfn2
intron2: 5′-AAGATCTCTCAGCATCCAAAAA-3′
and (ii) Mfn2 exon19: 5′-ATGGCACTTAGGGC-
TGGCAGCA-3′.
Naón et al., Science 380, eadh9351 (2023) 23 June 2023
Different products were cloned into Topo-TA
vector (Invitrogen) and sequenced with M13
forward and reverse primers using BigDye
3.1 terminator (Applied Biosystem). Inserts were
subcloned in frame into the mammalian expres-
sion vectors pCMV-HA and pEGFP (Clontech)
using Bgl II (NEB) and Asp 718 (Roche)
enzymes.
Mouse variant amplification was performed
using cDNA from MEFs and the primers
5′-AAGCTTGGACAGGTGGAGTCA-3′ and 5′-
CAACCAGCCAGCTTTATTCC-3′. Different moti-
lity products were cloned into pCR8 Topo-GW
vector (Invitrogen). First-Strand cDNA synthe-
sis was performed using 4 mg of total RNA and
the specific Mfn2 primer 5′-TGGCAAGAAGG-
GAGGCAAGTC-3′ and oligo dT primers incu-
bated at 50°C for 4 hours with SuperScript IV
Reverse Transcriptase (Invitrogen).
Split super-folder GFP (sfGFP) previously
engineered for efficient self-complementation
(10) breaks the sequence of sfGFP between the
10th and the 11th b-strand into two parts:
GFP1-10 and GFP11, a short, 16–amino acid pep-
tide (60). The GFP1-10 fragment, which contains
the three residues that constitute the GFP
chromophore, is nonfluorescent because
chromophore maturation requires the con-
served E222 residue located on GFP11. Upon
complementation, the reconstituted GFP
becomes fluorescent after the chromophore
maturation reaction is completed. GFP11 were
fused to the N terminus of ERMIN2 or ERMIT2
and coexpressed with GFP1-10.
pcDNA3.1-GFP(1-10) (Addgene plasmid no.
70219) and pEGFP-GFP11-Clathrin light chain
(Addgene plasmid no. 70217) were from Bo
Huang. GFP11 was generated by digestion of
pEGFP-GFP11-Clathrin light chain using BglII
and BclI. GFP11-Mfn2, GFP11-Ermin2 and GFP11-
Ermit2 were generated using Infusion (Takara
Technology) and the following primers: (i) Vec-
tor: 5′-TAGGGATCCACCGGATCTAGATAA-3′
and 5′-GCCTGTAATCCCAGCAGCATTTAC-3′. (ii)
Mfn2, Ermin2, and Ermit2: 5′-GCTGGGATTA-
CAGGCGGGATGTCCCTGCTCTTC-3′ and 5′-TCC-
GGTGGATCCCTAGGGCTATCTGCTGGGCTG-3′.
Mt-RFP, ER-RFP, ER-YFP, FEMP, ddGFP, mtCFP,
and aequorins (AEQs) were previously described
(13, 16, 61).
siRNA against ERMIN2 and ERMIT2 were
synthesized from the following sequences: 5′-
GTGATGTGGCCCAACTCTA-3′ and 5′-GACAT-
GATAGATGGCTTGA-3′, respectively; scrambled
siRNA was used as control. All siRNAs were
obtained from Sigma. Experiments were per-
formed 24 hours after transfection.
For RNA isolation, human and mouse tissues
were immediately frozen. RNA from liver tissues
was extracted using a protocol combining
TRIzol reagent (Invitrogen, Carlsbad, CA, USA)
and RNAeasy minikit columns (Qiagen, Alameda,
CA, USA), following the manufacturer’s ins-
tructions. RNA was reverse transcribed with
the SuperScript RT IV kit (Invitrogen, Carlsbad,
CA, USA). PCR was performed using the ABI
Prism 7900 HT real-time PCR machine (Applied
Biosystems, USA) and the SYBR Green PCR
Master Mix (Applied Biosystems, USA). Mea-
surements were normalized to PPIA (Cyclo-
philin A), ARP, or b-actin.
Absolute cDNA quantification in real-time,
PCR was performed using a standard curve
method. For standard curve construction, the
amount of plasmid MFN2, ERMIN2, and
ERMIT2 was measured using Qubit fluoro-
metric quantification (ThermoFisher scientific)
and copy number was calculated using the
online application Calculator for determining
the number of copies of a template (URI
Genomics and Sequencing Center).
Real-time PCR was performed using a
QuantStudio 6 Real-Time PCR system (Thermo-
Fisher) using Applied biosystems power Sybr
green PCR master mix (AB Applied Bioscience)
p
and the following primers: (i) MFN2: 5′-
GACTGAGCTGGGCGTGGTGG-3′ and 5′-GTA-
AACCTGCTGCTCCCGAGC-3′. (ii) ERMIN2: 5′-
GAACCAGCTGGCCCATGCCCT-3′ and 5′-
TCCTGGGTGAGCGAGCCCTG-3′. (iii) ERMIT2:
5′-TCCCTGCTCTTCTCTCGATG-3′ and 5′-
g
ATTCTTATAAACCTTGAGGACT-3′.
PCR was run at standard conditions: 50°C
for 2 min, predenaturation at 95°C for 10 min,
40 cycles of denaturation at 95°C for 15 s,
followed by annealing plus extension at 60°C
y
for 1 min. Melting curve analysis was performed
following the manufacturer’s instructions.
RPA
Probes were designed for ERMIN2 and
ERMIT2 so that they covered the flanking se-
quences at 5′- and -3′ of the skipped region and
a nonmatching sequence representing ~20%
of the probe length. PCR was performed using
primers that incorporate a SP6 and T7 phage
y g
promoter sequence. PCR products were cloned
into pGEMT Easy Vector (Promega) and se-
quenced. Cloned probes were linearized with
Sac II restriction enzyme. Retrotranscription
and antisense labeled probes were performed
,
using T7 or SP6 phage RNA polymerases
(MaxiScript Kit, Ambion) labeled with [a-32P]
UTP (Perkin Elmer), following manufacturer’s
instructions. Deoxyribonuclease (DNase) diges-
tion of template was achieved by phenol/
chloroform extraction precipitation with
NH4Acetate and linear acrylamide as carrier.
Hybridization and digestion were performed
using RPA-III Kit (Ambion), Target RNA and
positive control RNA obtained from over-
expressing cells were hybridized with 4.4 ×
105 cpm of probe overnight at 56°C. Ribonu-
clease (RNase) digestion of unhybridized RNA
was performed at 33°C for 60 min with RNase
A/T1 mix. Separation and detection of pro-
tected fragments were carried out using 8 M
urea denaturing 5% polyacrylamide gel. The
9 of 13
RES EARCH | R E S E A R C H A R T I C L E
dried gel was exposed at −80°C overnight using
an x-ray film (Amersham) or a Molecular Dy-
namics (Amersham Pharmacia) PhosphoImager
plate, which was then scanned with a Ty-
phoon 9200 Bio-imaging analyzer (Molecu-
lar dynamics, Amersham Pharmacia Biotech,
Piscataway, NJ).
Cell culture
HeLa and SV40 immortalized wild-type (WT),
Mfn2−/−, Mfn1−/−Mfn2−/− MEFs were grown in
Dulbecco’s minimum essential medium (DMEM)
(Gibco) supplemented with 10% (vol/vol) fetal
bovine serum (FBS) (Gibco), 1× nonessential
amino acids, 2 mM L-glutamine, 100 U/ml
penicillin, and 100 mg/ml streptomycin (Gibco).
Cells were maintained at 80 to 90% confluency.
Microscopy experiments were performed at
70% confluency. MEFs were cultured and
transfected as described previously using
Transfectin (Biorad) or Lipoafectamine 2000
(Invitrogen), following the manufacturer’s ins-
tructions (13). siRNA transfection was per-
formed at a final concentration of 250 nM
using Oligofectamine (Invitrogen). For imaging
studies, 1.6 × 104 cells were seeded onto 24-mm
round glass coverslips or 4 × 103 cells onto
13-mm round glass coverslips.
Imaging
For confocal imaging of live cells, 2 × 105 cells
seeded onto 24-mm round glass coverslips,
treated as indicated, were placed on the stage
of a Nikon Eclipse TE300 inverted microscope
equipped with a PerkinElmer Ultraview LCI
confocal system, a piezoelectric z axis motor-
ized stage (Pifoc, Physik Instrumente), and an
Orca ER 12-bit CCD camera (Hamamatsu
Photonics). Acquisition, deconvolution, 3D
reconstruction, volume rendering of the stacks,
as well as analysis of the mitochondria-to-ER
interaction were performed as described pre-
viously (16). For colocalization analyses, images
stacks were acquired using a Leica SP5 inverted
microscope equipped with confocal or Leica
TCS SP2 AOBS, using a 63X1.4 NA Plan Apo
objective. Cells were excited using 488 nm
Ar/ArKr and 561 DPSS (diode pump solid
state) lasers.
For FEMP FRET imaging, unless differently
specified, 9000 Mfn2−/− MEFs/ml were seeded
on a Cell carrier 384-well plate (PerkinElmer)
and transfected as indicated. After 24 hours,
cells were transfected with FEMP cDNA using
Genjet (SignaGen) according to manufacturer’s
instructions, and after further 24 hours, cells
were imaged using an Operetta CLS High
Content imaging system (Perkin Elmer) using
the following filters: CFP (excitation 410 to
430, emission 460 to 500), YFP (excitation 490
to 510, emission 520 to 560), and YFPFRET
(excitation 410 to 430, emission 520 to 560).
To image the maximum FRET intensity
(FRETmax), cells were treated with 100 nM
Naón et al., Science 380, eadh9351 (2023) 23 June 2023
Rapamycin for 15 min and then fixed with
1% formaldehyde for 10 min. Images were
analyzed using PerkinElmer Harmony 3.5 image
analysis software. The YFP channel was chosen
to mark the region of interest (ROI), and around
each ROI, a second boundary was drawn to
measure the background (bg) intensity. FRETbasal
and FRETmax were calculated as
ðFYFP  FRETcell  FYFP  FRETbg Þ
FRET ¼
ðFCFPcell  FCFPbg Þ
and the FRET ratio was calculated as
ðFRETmax  FRETbasal Þ
FRETratio ¼
ðFRETbasal Þ
FRAP
After 24 hours, cells plated and transfected
with erYFP as described were incubated in
HBSS and placed on the stage of a laser scan-
ning microscope (TCS SP5, Leica). Using the
LasAF software (Leica), regions measuring
16 mm2 were manually defined to be bleached
avowing the perinuclear region. The selected
cells showed a moderate expression of erYFP.
To bleach the YFP fluorescence, one z-plane
was bleached for a total of 3 s using 100%
power of the 488-nm laser line with a 63X,
1.4 NA objective. The postbleaching images
were taken at 1-s intervals for a total of 90 s.
Intensities of the photobleached regions were
measured and normalized to the intensities of
the same region before photobleaching (photo-
bleaching effect) and the neighboring regions
at the same time point (auto-bleaching effect
during the recording).
Morphometric and MCS analysis
For morphometric analysis of mitochondria,
the length of the major axis and the roundness
index of each identified object were calculated.
Cells were scored with elongated mitochondria
when >50% of the objects in the image (i.e.,
mitochondria) displayed a major axis longer
than 3 mm and a roundness index below 0.3
(maximum value is 1). Alternatively, mitochon-
dria aspect ratio (AR) was calculated from
automatically thresholded images in ImageJ
(NIH). For analysis of the mitochondria-ER
interaction, images of cells expressing mtRFP
or erYFP were processed using the automatic
threshold function of ImageJ, followed by
deconvolution, 3D reconstruction, and surface
rendering using the VolumeJ plugin of ImageJ.
The interaction between the ER and mitochon-
dria was analyzed using Manders’ colocaliza-
tion coefficient.
For ddGFP imaging, Mfn2−/− MEFs were
cotransfected with the indicated plasmids
and with ddGFP-A and ddGFP-B. Forty-eight
hours after transfection, ddGFP fluorescence
was acquired as previously described (16).
Subcellular fractionation
Subcellular fractions of HeLa cells (109) were
obtained as described (13). Cells were washed
with phosphate-buffered saline (PBS), sus-
pended in IB (200 mM sucrose, 1 mM EGTA-
Tris, and 10 mM Tris-MOPS, pH 7.4), and then
disrupted by dounce homogenization. The
homogenate was spun at 800 × g for 10 min.
The supernatant was recovered and further
centrifuged for 10 min at 8000 × g. The
resulting pellet (mitochondrial fraction) was
collected while the supernatant was further
spun for 30 min at 100,000 × g. The pellet was
discarded, and the supernatant was centri-
fuged for 1 hour at 100,000 × g. The resulting
pellet is the LM fraction, and the supernatant
is the cytosolic fraction. The mitochondrial
fraction was further purified by centrifuging
twice at 8000 × g for 10 min. The obtained
pellet was purified by centrifugation at 95,000
× g for 30 min on a 30% Percoll gradient in IB.
p
The mitochondrial layer obtained was washed
free of Percoll and resuspended in IB. MAMs
were identified as an intermediate layer be-
tween the LMs and the mitochondrial fraction
on the Percoll gradient, as previously described.
Forty micrograms of protein was separated by
g
SDS–polyacrylamide gel electrophoresis (SDS-
PAGE) and immunoblotted as indicated in the
figure legends.
Liver fractions were purified as previously
described (62, 63). After homogenization of
y
~1 g of liver with a Teflon potter in IB (225 mM
mannitol, 75 mM sucrose, 0.5% BSA, 0.5 mM
EGTA, and 30 mM Tris-HCl, pH 7.4), cellular
debris and nucleus were removed with two
centrifugations at 740 × g for 5 min. A small
volume of supernatant was taken, this was
called the homogenate fraction. Crude mito-
chondria were collected by centrifugation at
9000 × g for 10 min, and the pellet was resus-
pended in Mitochondria Buffer (MB) (250 mM
y g
mannitol, 5 mM HEPES and 0.5 mM EGTA,
pH 7.4). We then proceeded as described for
subcellular fractionation of cells. Proteins were
determined using the PierceTM BCA Protein
Assay kit (ThermoFisher scientific).
,
Trypsin protection assay
Adenoviruses Ad-ERMIN2 and Ad-ERMIT2
(1 × 109 inclusion-forming units per mouse)
were tail-vein injected to reach the liver of
Mfn2LKO mice. LMs (1 mg/ml) were isolated
from livers 3 days after adenovirus delivery
and incubated with the indicated trypsin
concentrations for 15 min on ice. SBTI (5 mg/ml)
was added and incubated for 20 min on ice.
Laemmli sample buffer was added, and sam-
ples were boiled for 5 min and separated by
12% SDS-PAGE.
Immunoprecipitation
Forty microliters of DynaBeads were cross-
linked with 3 ml of anti-GFP (Invitrogen) or
10 of 13
RES EARCH | R E S E A R C H A R T I C L E
4 ml of anti-HA (Roche) using BS3 (Sulfo-
DSS, Thermo Scientific Pierce). Total protein
extracts (400 mg) were dissolved in RIPA
buffer (1% TritonX-100, 0.1% SDS, 0.5% de-
oxycholate, 0.15 M NaCl, 5 mM Tris, pH8)
and incubated with antibody-crosslinked
Dynabeads protein G (Thermo fisher Scien-
tific) for 16 hours 4°C in rotation. After three
washes for 30 min at 4°C in rotation with PBS
0.2Tween 20%, immunoprecipitates were
boiled in 40 ml elution buffer [glycine 50 mM
2X NuPage loading buffer (Invitrogen) 4%
b-mercaptoethanol]. Proteins were separated
by SDS-PAGE and immunoblotted using the
indicated antibodies.
Immunoblotting
Homogenates for Western blot analyses were
obtained from either cell cultures or tissues.
Cells (1.8 × 106 or 80% of confluence) were
harvested 24 hours (for siRNA experiments)
or 48 hours after transfection and lysed in
150 mM NaCl, 1% Nonidet P-40/0, 0.25% deo-
xycholate, 1 mM EDTA, and 50 mM Tris, pH 7.4
in the presence of complete protease inhibitor
mixture (Sigma-Aldrich).
Tissue samples were homogenized in 10
volumes of lysis buffer using a polytron. Homo-
genates were rotated for 1 hour at 4°C in an
orbital shaker and centrifuged at 13,000 rpm
for 15 min at 4°C.
Forty micrograms of extracted proteins was
separated by 4 to 12% Bis-Tris SDS-PAGE
(NuPAGE; Invitrogen), transferred onto poly-
vinylidene difluoride membranes (Bio-Rad
Laboratories), and probed using the following
antibodies: actin (1:2000; EMD Millipore);
rabbit polyclonal anti Mfn2 (1:1000, homemade
raised against a peptide located between amino
acids 557 and 576 of the C terminus of Mfn2);
anti-Mfn2 (1:1000; Abcam ab 50838, against Mfn2
amino acids 38 to 55 or Abnova H00009927-
M3, against Mfn2 amino acids 661 to 675);
chicken anti-Mfn1 (1:1000, Abcam); mouse
anti-GFP (1:1000, Invitrogen); rabbit anti-
Tom20 (1:5000, Santa Cruz Biotechnology);
mouse anti-VDAC1 (1:1000, ab14734, Abcam);
goat anti-LDH (1:500, Rockland); goat anti-
FACL4 (Epitomic S0101); rabbit anti-calreticulin
(Crt, 1:750, Upstate); rabbit anti-Calnexin
(1:1000; Stressgen); rat anti-HA (1:1000;
Roche); and mouse anti-SERCA2 (1:1000;
Thermo Fisher MA3-919). Peroxidase-conjugated
anti-mouse, anti-goat, or anti-rabbit immunoglo-
bulins were used as secondary antibodies. The
following antibodies were used in tissue extracts:
Mfn2 (1:1000, Abcam); p-eIF2a; eIF2a; p-PERK;
PERK (1:1000, Cell Signaling); CHOP (1:1000,
Santa Cruz Biotechnologies); b-actin (1:10,000,
Sigma); and a-tubulin (1:8000, Sigma). The
specific proteins were detected by the ECL
Western blotting detection analysis system
(Amersham). Densitometry was performed
using ImageJ (NIH).
Naón et al., Science 380, eadh9351 (2023) 23 June 2023
Aequorin measurements of Ca2+ concentration
For ER Ca2+ measurement, cells grown on
13-mm round glass coverslips at 50% con-
fluence were cotransfected with erAEQ and
the plasmids indicated. For reconstituting
erAeq, the luminal [Ca2+]ER was reduced by
incubating cells for 5 min at 37°C with Krebs-
Ringer bicarbonate (KRB) supplemented with
1% FBS, 1 mM EGTA, 10 mM ionomycin, and
0.1 mM 2,5-di-(tert-butyl)-1,4-benzohydroquinone
(tBuBHQ). Cells were then washed three times
with KRB containing 1 mM EGTA and 5% FBS
and reconstituted for 3 hours at 4°C with KRB
supplemented with 0.1 mM tBuBHQ, 1 mM
EGTA, and 5 mM Coelentarazine N. All AEQ
measurements were carried out in KRB. Ag-
onists and other drugs were added to the same
medium.
For cytosolic and mitochondrial Ca2+ mea-
surement, cells were cotransfected with CytAEQ
and mtAEQ and the indicated plasmid. Re-
constitution was performed in KRB containing
1 mM Ca2+, and measurement and calibration
were performed in KRB containing 100 mM
EGTA. ATP was used as IP3R agonist.
Measurement of mitochondrial ATP production
Cells cotransfected with mitochondrial lucifer-
ase (mtLuc) and the indicated plasmids were
constantly perfused with a modified KRB
containing 125 mM NaCl, 5 mM KCl, 1 mM
Na3PO4, 1 mM MgSO4, 1 mM CaCl2, 20 mM
luciferin, and 20 mM Hepes (pH 7.4 at 37°C).
Agonists were perfused in the same buffer
containing 20 mM luciferin. Cell luminescence
were measured in a luminometer.
Flow cytometry and sorting
Analysis of ddGFP fluorescence was performed
as described (16, 21). Mfn2−/− MEFs were first
cotransfected with cytosolic mCherry and with
ddGFP-A and ddGFP-B. After 24 hours, cells
were transfected with the indicated plasmids.
Twenty-four hours after the second transfec-
tion, GFP and mCherry fluorescence were
evaluated by flow cytometry analysis. Acquisi-
tion was stopped when 50,000 FL2+ events
were reached. Data are shown as dot plot of
FL1 versus side scatter derived from FL2+
gated events. Cell size (forward scatter, FSC),
cytosolic mCherry level (red signal intensity
detected by FL2 channel), and ddGFP fluores-
cence (GFP signal intensity detected by a 15-mW
argon ion laser tuned at 488 nm) analyses
were performed on a FacsCalibur Flow Cytome-
ter using CellQuest software (Becton Dickinson
Biosciences).
For sorting using FACSAria (Becton Dickinson
Biosciences), cells were analyzed through a
530-nm band-pass filter as they traversed the
beam of an argon ion laser (488 nm, 100 mW).
Sorted GFP+ cells were cultured in DMEM
supplemented with 2 mM glutamine, 1 mM
penicillin/streptomycin, and 20% (vol/vol)
FBS at 37°C in a fully humidified atmosphere
of 95% air and 5% CO2. On the following day,
the medium was replaced to remove dead
cells. Growing cells were fixed after 8 hours
and prepared for EM images.
Electron microscopy
MEFs cotransfected with GFP and the indi-
cated plasmids and sorted as described above
were fixed with 1.25% (vol/vol) glutaraldehyde
in 0.1 M sodium cacodylate at pH 7.4 for 1 hour
at room temperature. Thin sections were
imaged on a Tecnai-20 electron microscope
(Philips-FEI).
Morphometric measurements were carried
out using ImageJ (NIH). For calculations of
mitochondria-ER distance, >5 mitochondria per
image in 70 images per condition were con-
sidered, and the minimal distance of the ER
located in a 30- or 20-nm radius from the con-
sidered mitochondria was computed. ERMICC
p
was calculated as previously described (16).
Human biopsies
Liver biopsies were obtained from NAFLD
patients undergoing bariatric surgery; liver
biopsies from normal individuals were not
g
collected due to ethical issues. Biopsies of sub-
cutaneous adipose tissue and skeletal muscle
were obtained from nonobese subjects. All
patients gave written informed consent. The
study protocols conformed to the Ethical
y
Guidelines of the 1975 Declaration of Helsinki,
revised in 2000, as reflected in a priori approval
by the Hospital Sant Joan de Reus (Institutional
Review Board, project code: INFLAMED/15-04-
30/4prog7), Hospital Joan XXIII, and Human
Ethics Committee.
Animal care and experiments
All animal work was approved and conducted
following established guidelines. This project
y g
was assessed favorably by the Institutional
Animal Care and Use Committee of the Parc
Cientific de Barcelona (IACUC-PCB), which
considers that the above-mentioned project
complies with standard ethical regulations
,
and meets the requirements of current ap-
plicable legislation (RD 53/2013 Council Di-
rective; 2010/63/UE; Order 214/1997/GC).
Control and Mfn2LKO mice were littermates
and described previously (10). The MCD diet
protocol was described previously (10). Mice
were kept under a 12-hour light-dark period
and provided with a standard chow diet and
water ad libitum. Eight-week-old C57Bl6/J
males were fed a standard diet. On the ex-
perimental day, mice were anesthetized using
isoflurane and euthanized by cervical disloca-
tion. Tissues used for RNA purification, protein
extraction, or histology were prepared as de-
scribed (64, 65). Sample size was not predeter-
mined. For the MCD protocol, male and female
mice were randomized to chow or MCD diet
11 of 13
RES EARCH | R E S E A R C H A R T I C L E
after a blinded draw of their ID. Researchers
were not blinded to the genotype, the insert of
the adenovirus injected, or the diet treatment.
All animals were included in the analyses.
Adenoviral transduction
For adenoviral transduction of 8- to 10-week-old
mice, adenoviruses (1 × 109 inclusion-forming
units per mouse) were tail-vein injected. Livers
were isolated 5 days after adenovirus delivery.
The following adenoviruses were used in
this study: Ad-LacZ, Ad-ERMIN2, and Ad-ERMIT2
(encoding for human ERMIN2 and ERMIT2).
ERMIN2 and ERMIT2 were cloned by recom-
bination into the pAdeno-CMV-V5 adenoviral
vector (Invitrogen) using the Gateway system.
Adenoviruses were generated by transfection
of the adenoviral expression vectors in a human
embryonic kidney cell line (HEK 293). The
adenoviruses generated were then amplified
at the Unitat de Producció de Vectors Virals-
CBATEG (Universitat Autònoma de Barcelona).
3
H–L-serine incorporation into phospholipids in
subcellular fractions
Liver was homogenized use Teflon-glass
homogenizer in Isolation Buffer (IB) (225 mM
Mannitol, 25 mM Hepes-KOH, 1 mM EGTA,
pH 7.4 and protease inhibitors) at a ratio of
4 ml of IB for every gram of tissue. The homo-
genate was pelleted for 10 min at 1500 × g at 4°C.
The supernatant was transferred to a new tube
and pelleted again as above, transferred again
to a new tube, and pelleted at 13,000 × g for
20 min at 4°C. This new pellet contained the
crude mitochondria fraction, comprising mito-
chondria and MAM, and was used to measure
lipid transfer. The supernatant contained the
ER fraction, which was pelleted at 100,000 × g
for 1 hour at 4°C. This pellet was used as a con-
trol in the assay. One milligram of the fraction
was pelleted again and resuspended in 200 ml
of PS assay buffer [25 mM Hepes-KOH, 10 mM
CaCl2, 0.4 mM of 3H-Ser (20 to 30 mci/mmol)
pH 7.4]. The mixture was incubated for 45 min
at 37°C and the reaction was stopped by adding
3 volumes of chloroform:MeOH (2:1). Lipids
were extracted using the Folch Method, dried
on an N2 flow, and separated by TLC as de-
scribed (39, 66).
Statistical analysis
Data are shown as means ± SEM values of the
indicated number of independent experi-
ments, unless otherwise noted in the figure
legends. Data from individual experiments
are plotted as dots except in the AEQ and mt-
Luc experiments which represent means ±
SEMs of the indicated number of indepen-
dent experiments. OriginPro 2022 (OriginLab
Corp., Northampton MA, USA) was used for
statistical analysis. The sample size was prede-
termined on the basis of published literature and
previous laboratory experience. No statistical
Naón et al., Science 380, eadh9351 (2023) 23 June 2023
methods were used to predetermine the sample
size. Normal distribution of data was verified
by a Shapiro–Wilkinson test. Homogeneity of
the variance was verified using Levene’s test. If
data fulfilled normality and homoscedasticity
criteria, parametric tests were used to evaluate
significance. When data exhibited a nonnor-
mal and unequal distribution, nonparametric
tests were applied. Used tests are specified in
the figure legends. Sample sizes and P values
are indicated in the figure legends. P < 0.05
was considered significant.
RE FERENCES AND NOTES
1. M. Eisenberg-Bord, N. Shai, M. Schuldiner, M. Bohnert,
A Tether Is a Tether Is a Tether: Tethering at Membrane
Contact Sites. Dev. Cell 39, 395–409 (2016). doi: 10.1016/
j.devcel.2016.10.022; pmid: 27875684
2. L. Scorrano et al., Coming together to define membrane
contact sites. Nat. Commun. 10, 1287 (2019). doi: 10.1038/
s41467-019-09253-3; pmid: 30894536
3. L. S. Jouaville, P. Pinton, C. Bastianutto, G. A. Rutter, R. Rizzuto,
Regulation of mitochondrial ATP synthesis by calcium:
Evidence for a long-term metabolic priming. Proc. Natl. Acad.
Sci. U.S.A. 96, 13807–13812 (1999). doi: 10.1073/
pnas.96.24.13807; pmid: 10570154
4. G. Hajnóczky, L. D. Robb-Gaspers, M. B. Seitz, A. P. Thomas,
Decoding of cytosolic calcium oscillations in the mitochondria.
Cell 82, 415–424 (1995). doi: 10.1016/0092-8674(95)90430-1;
pmid: 7634331
5. L. Scorrano et al., BAX and BAK regulation of endoplasmic
reticulum Ca2+: A control point for apoptosis. Science 300,
135–139 (2003). doi: 10.1126/science.1081208; pmid: 12624178
6. J. Prudent et al., MAPL SUMOylation of Drp1 Stabilizes an
ER/Mitochondrial Platform Required for Cell Death. Mol. Cell
59, 941–955 (2015). doi: 10.1016/j.molcel.2015.08.001;
pmid: 26384664
7. J. R. Friedman et al., ER tubules mark sites of mitochondrial
division. Science 334, 358–362 (2011). doi: 10.1126/
science.1207385; pmid: 21885730
8. D. W. Hailey et al., Mitochondria supply membranes for
autophagosome biogenesis during starvation. Cell 141,
656–667 (2010). doi: 10.1016/j.cell.2010.04.009;
pmid: 20478256
9. M. Hamasaki et al., Autophagosomes form at ER-mitochondria
contact sites. Nature 495, 389–393 (2013). doi: 10.1038/
nature11910; pmid: 23455425
10. M. I. Hernández-Alvarez et al., Deficient Endoplasmic
Reticulum-Mitochondrial Phosphatidylserine Transfer Causes
Liver Disease. Cell 177, 881–895.e17 (2019). doi: 10.1016/
j.cell.2019.04.010; pmid: 31051106
11. M. Schneeberger et al., Mitofusin 2 in POMC neurons connects
ER stress with leptin resistance and energy imbalance.
Cell 155, 172–187 (2013). doi: 10.1016/j.cell.2013.09.003;
pmid: 24074867
12. G. Csordás et al., Structural and functional features and
significance of the physical linkage between ER and
mitochondria. J. Cell Biol. 174, 915–921 (2006). doi: 10.1083/
jcb.200604016; pmid: 16982799
13. O. M. de Brito, L. Scorrano, Mitofusin 2 tethers endoplasmic
reticulum to mitochondria. Nature 456, 605–610 (2008).
doi: 10.1038/nature07534; pmid: 19052620
14. R. Filadi et al., Mitofusin 2 ablation increases endoplasmic
reticulum-mitochondria coupling. Proc. Natl. Acad. Sci. U.S.A.
112, E2174–E2181 (2015). doi: 10.1073/pnas.1504880112;
pmid: 25870285
15. P. Cosson, A. Marchetti, M. Ravazzola, L. Orci, Mitofusin-2
independent juxtaposition of endoplasmic reticulum and
mitochondria: An ultrastructural study. PLOS ONE 7, e46293
(2012). doi: 10.1371/journal.pone.0046293; pmid: 23029466
16. D. Naon et al., Critical reappraisal confirms that Mitofusin 2 is
an endoplasmic reticulum-mitochondria tether. Proc. Natl.
Acad. Sci. U.S.A. 113, 11249–11254 (2016). doi: 10.1073/
pnas.1606786113; pmid: 27647893
17. A. Mourier et al., Mitofusin 2 is required to maintain
mitochondrial coenzyme Q levels. J. Cell Biol. 208, 429–442
(2015). doi: 10.1083/jcb.201411100; pmid: 25688136
18. T. Daniele et al., Mitochondria and melanosomes establish
physical contacts modulated by Mfn2 and involved in organelle
biogenesis. Curr. Biol. 24, 393–403 (2014). doi: 10.1016/j.
cub.2014.01.007; pmid: 24485836
19. A. Sugiura et al., MITOL regulates endoplasmic reticulum-
mitochondria contacts via Mitofusin2. Mol. Cell 51, 20–34
(2013). doi: 10.1016/j.molcel.2013.04.023; pmid: 23727017
20. A. Duarte et al., Mitochondrial fusion is essential for steroid
biosynthesis. PLOS ONE 7, e45829 (2012). doi: 10.1371/journal.
pone.0045829; pmid: 23029265
21. S. C. Alford, Y. Ding, T. Simmen, R. E. Campbell, Dimerization-
dependent green and yellow fluorescent proteins. ACS Synth.
Biol. 1, 569–575 (2012). doi: 10.1021/sb300050j;
pmid: 23656278
22. D. Li, X. Li, Y. Guan, X. Guo, Mitofusin-2-mediated tethering of
mitochondria and endoplasmic reticulum promotes cell cycle
arrest of vascular smooth muscle cells in G0/G1 phase.
Acta Biochim. Biophys. Sin. 47, 441–450 (2015). doi: 10.1093/
abbs/gmv035; pmid: 25926139
23. B. Kornmann et al., An ER-mitochondria tethering complex
revealed by a synthetic biology screen. Science 325, 477–481
(2009). doi: 10.1126/science.1175088; pmid: 19556461
24. K. J. De Vos et al., VAPB interacts with the mitochondrial
protein PTPIP51 to regulate calcium homeostasis. Hum. Mol.
Genet. 21, 1299–1311 (2012). doi: 10.1093/hmg/ddr559;
pmid: 22131369
25. V. Debattisti, D. Pendin, E. Ziviani, A. Daga, L. Scorrano,
Reduction of endoplasmic reticulum stress attenuates the
defects caused by Drosophila mitofusin depletion. J. Cell Biol.
p
204, 303–312 (2014). doi: 10.1083/jcb.201306121;
pmid: 24469638
26. Y. Chen et al., Mitofusin 2-containing mitochondrial-reticular
microdomains direct rapid cardiomyocyte bioenergetic
responses via interorganelle Ca2+ crosstalk. Circ. Res. 111,
863–875 (2012). doi: 10.1161/CIRCRESAHA.112.266585;
pmid: 22777004
27. C. Tezze et al., Age-Associated Loss of OPA1 in Muscle Impacts
Muscle Mass, Metabolic Homeostasis, Systemic Inflammation,
g
and Epithelial Senescence. Cell Metab. 25, 1374–1389.e6
(2017). doi: 10.1016/j.cmet.2017.04.021; pmid: 28552492
28. L. M. Restelli et al., Neuronal Mitochondrial Dysfunction
Activates the Integrated Stress Response to Induce Fibroblast
Growth Factor 21. Cell Rep. 24, 1407–1414 (2018).
y
doi: 10.1016/j.celrep.2018.07.023; pmid: 30089252
29. L. Wang et al., Disruption of mitochondrial fission in the liver
protects mice from diet-induced obesity and metabolic
deterioration. Diabetologia 58, 2371–2380 (2015).
doi: 10.1007/s00125-015-3704-7; pmid: 26233250
30. S. Stamm et al., Function of alternative splicing.
Gene 344, 1–20 (2005). doi: 10.1016/j.gene.2004.10.022;
pmid: 15656968
31. K. Itoh et al., Brain-specific Drp1 regulates postsynaptic
endocytosis and dendrite formation independently of
mitochondrial division. eLife 8, e44739 (2019). doi: 10.7554/
eLife.44739; pmid: 31603426
32. K. Itoh et al., A brain-enriched Drp1 isoform associates with
y g
lysosomes, late endosomes, and the plasma membrane.
J. Biol. Chem. 293, 11809–11822 (2018). doi: 10.1074/
jbc.RA117.001253; pmid: 29853636
33. T. Koshiba et al., Structural basis of mitochondrial tethering by
mitofusin complexes. Science 305, 858–862 (2004).
doi: 10.1126/science.1099793; pmid: 15297672
34. N. Ishihara, Y. Eura, K. Mihara, Mitofusin 1 and 2 play distinct
roles in mitochondrial fusion reactions via GTPase activity.
,
J. Cell Sci. 117, 6535–6546 (2004). doi: 10.1242/jcs.01565;
pmid: 15572413
35. A. W. Senior et al., Improved protein structure prediction using
potentials from deep learning. Nature 577, 706–710 (2020).
doi: 10.1038/s41586-019-1923-7; pmid: 31942072
36. Y. L. Cao et al., MFN1 structures reveal nucleotide-triggered
dimerization critical for mitochondrial fusion. Nature 542,
372–376 (2017). doi: 10.1038/nature21077; pmid: 28114303
37. M. Giacomello et al., Ca2+ hot spots on the mitochondrial
surface are generated by Ca2+ mobilization from stores, but
not by activation of store-operated Ca2+ channels. Mol. Cell 38,
280–290 (2010). doi: 10.1016/j.molcel.2010.04.003;
pmid: 20417605
38. G. Csordás et al., Imaging interorganelle contacts and local
calcium dynamics at the ER-mitochondrial interface. Mol. Cell
39, 121–132 (2010). doi: 10.1016/j.molcel.2010.06.029;
pmid: 20603080
39. E. Area-Gomez et al., Upregulated function of mitochondria-
associated ER membranes in Alzheimer disease. EMBO J. 31,
4106–4123 (2012). doi: 10.1038/emboj.2012.202;
pmid: 22892566
12 of 13
RES EARCH | R E S E A R C H A R T I C L E
40. A. Kasahara, S. Cipolat, Y. Chen, G. W. Dorn2nd,
L. Scorrano, Mitochondrial fusion directs cardiomyocyte
differentiation via calcineurin and Notch signaling. Science
342, 734–737 (2013). doi: 10.1126/science.1241359;
pmid: 24091702
41. K. Singaravelu et al., Mitofusin 2 regulates STIM1 migration
from the Ca2+ store to the plasma membrane in cells with
depolarized mitochondria. J. Biol. Chem. 286, 12189–12201
(2011). doi: 10.1074/jbc.M110.174029; pmid: 21220420
42. O. M. de Brito, L. Scorrano, Mitofusin 2: A mitochondria-
shaping protein with signaling roles beyond fusion.
Antioxid. Redox Signal. 10, 621–634 (2008). doi: 10.1089/
ars.2007.1934; pmid: 18092941
43. R. G. Hansford, Dehydrogenase activation by Ca2+ in cells
and tissues. J. Bioenerg. Biomembr. 23, 823–854 (1991).
doi: 10.1007/BF00786004; pmid: 1778993
44. R. Galmes et al., ORP5/ORP8 localize to endoplasmic
reticulum-mitochondria contacts and are involved in
mitochondrial function. EMBO Rep. 17, 800–810 (2016).
doi: 10.15252/embr.201541108; pmid: 27113756
45. F. Giordano et al., PI(4,5)P2-dependent and Ca2+-regulated
ER-PM interactions mediated by the extended synaptotagmins.
Cell 153, 1494–1509 (2013). doi: 10.1016/j.cell.2013.05.026;
pmid: 23791178
46. H. Chen et al., Mitofusins Mfn1 and Mfn2 coordinately regulate
mitochondrial fusion and are essential for embryonic
development. J. Cell Biol. 160, 189–200 (2003). doi: 10.1083/
jcb.200211046; pmid: 12527753
47. P. Pinton et al., Reduced loading of intracellular Ca2+ stores
and downregulation of capacitative Ca2+ influx in Bcl-2–
overexpressing cells. J. Cell Biol. 148, 857–862 (2000).
doi: 10.1083/jcb.148.5.857; pmid: 10704437
48. N. Borgese, I. Gazzoni, M. Barberi, S. Colombo, E. Pedrazzini,
Targeting of a tail-anchored protein to endoplasmic reticulum
and mitochondrial outer membrane by independent but
competing pathways. Mol. Biol. Cell 12, 2482–2496 (2001).
doi: 10.1091/mbc.12.8.2482; pmid: 11514630
49. G. Orso et al., Homotypic fusion of ER membranes requires the
dynamin-like GTPase atlastin. Nature 460, 978–983 (2009).
doi: 10.1038/nature08280; pmid: 19633650
50. R. Rizzuto, M. Brini, M. Murgia, T. Pozzan, Microdomains with
high Ca2+ close to IP3-sensitive channels that are sensed by
neighboring mitochondria. Science 262, 744–747 (1993).
doi: 10.1126/science.8235595; pmid: 8235595
51. R. H. Baloh, R. E. Schmidt, A. Pestronk, J. Milbrandt,
Altered axonal mitochondrial transport in the pathogenesis of
Charcot-Marie-Tooth disease from mitofusin 2 mutations.
J. Neurosci. 27, 422–430 (2007). doi: 10.1523/
JNEUROSCI.4798-06.2007; pmid: 17215403
52. A. Franco et al., Burst mitofusin activation reverses
neuromuscular dysfunction in murine CMT2A. eLife 9, e61119
(2020). doi: 10.7554/eLife.61119; pmid: 33074106
53. A. G. Rocha et al., MFN2 agonists reverse mitochondrial
defects in preclinical models of Charcot-Marie-Tooth disease
type 2A. Science 360, 336–341 (2018). doi: 10.1126/
science.aao1785; pmid: 29674596
54. S. Züchner et al., Mutations in the mitochondrial GTPase
mitofusin 2 cause Charcot-Marie-Tooth neuropathy type 2A.
Naón et al., Science 380, eadh9351 (2023) 23 June 2023
Nat. Genet. 36, 449–451 (2004). doi: 10.1038/ng1341;
pmid: 15064763
55. C. Guelly et al., Targeted high-throughput sequencing identifies
mutations in atlastin-1 as a cause of hereditary sensory
neuropathy type I. Am. J. Hum. Genet. 88, 99–105 (2011).
doi: 10.1016/j.ajhg.2010.12.003; pmid: 21194679
56. C. Glytsou et al., Mitophagy promotes resistance to BH3
mimetics in acute myeloid leukemia. Cancer Discov.
CD-22-0601 (2023). doi: 10.1158/2159-8290.CD-22-0601;
pmid: 37088914
57. A. Olichon et al., OPA1 alternate splicing uncouples an
evolutionary conserved function in mitochondrial fusion from a
vertebrate restricted function in apoptosis. Cell Death Differ.
14, 682–692 (2007). doi: 10.1038/sj.cdd.4402048;
pmid: 17024226
58. V. Del Dotto, M. Fogazza, V. Carelli, M. Rugolo, C. Zanna, Eight
human OPA1 isoforms, long and short: What are they for?
Biochim. Biophys. Acta Bioenerg. 1859, 263–269 (2018).
doi: 10.1016/j.bbabio.2018.01.005; pmid: 29382469
59. S. Strack, T. J. Wilson, J. T. Cribbs, Cyclin-dependent kinases
regulate splice-specific targeting of dynamin-related protein
1 to microtubules. J. Cell Biol. 201, 1037–1051 (2013).
doi: 10.1083/jcb.201210045; pmid: 23798729
60. D. Kamiyama et al., Versatile protein tagging in cells with
split fluorescent protein. Nat. Commun. 7, 11046 (2016).
doi: 10.1038/ncomms11046; pmid: 26988139
61. S. Cipolat, O. Martins de Brito, B. Dal Zilio, L. Scorrano,
OPA1 requires mitofusin 1 to promote mitochondrial fusion.
Proc. Natl. Acad. Sci. U.S.A. 101, 15927–15932 (2004).
doi: 10.1073/pnas.0407043101; pmid: 15509649
62. J. M. Suski et al., Isolation of plasma membrane-associated
membranes from rat liver. Nat. Protoc. 9, 312–322 (2014).
doi: 10.1038/nprot.2014.016; pmid: 24434800
63. M. R. Wieckowski, C. Giorgi, M. Lebiedzinska, J. Duszynski,
P. Pinton, Isolation of mitochondria-associated membranes
and mitochondria from animal tissues and cells. Nat. Protoc. 4,
1582–1590 (2009). doi: 10.1038/nprot.2009.151;
pmid: 19816421
64. M. Liesa et al., Mitochondrial fusion is increased by the nuclear
coactivator PGC-1b. PLOS ONE 3, e3613 (2008). doi: 10.1371/
journal.pone.0003613; pmid: 18974884
65. F. X. Soriano et al., Evidence for a mitochondrial regulatory
pathway defined by peroxisome proliferator-activated
receptor-g coactivator-1a, estrogen-related receptor-a, and
mitofusin 2. Diabetes 55, 1783–1791 (2006). doi: 10.2337/
db05-0509; pmid: 16731843
66. E. Area-Gomez, Assessing the function of mitochondria-
associated ER membranes. Methods Enzymol. 547, 181–197
(2014). doi: 10.1016/B978-0-12-801415-8.00011-4;
pmid: 25416359
ACKN OWLED GMEN TS
We thank V. Hernandez, J. M. Seco, and L. Bardia (Advanced
Digital Microscopy, IRB Barcelona) and D. Bach for technical
assistance; N. Berrow (Protein Expression, IRB Barcelona) for
plasmid generation; J. Comas and R. Alvarez (Cytometry Unit, UB)
and R. Seminago and A. Amador (Genomics Unit, UB) for technical
assistance; F. Caicci and F. Boldrin (Bioimaging Facility,
Department of Biology, University of Padova) for EM samples
preparation; and A. Cabrelle (FACS facility, VIMM) for flow
cytometry and sorting. Funding: This study was supported by
MINECO PID2019-106209RB-I00 (A.Z.); Generalitat de Catalunya
grant 2017SGR1015 (A.Z.); CIBERDEM “Instituto de Salud Carlos
III” (A.Z.); Fundación Ramon Areces CIVP18A3942 (A.Z.);
Fundación BBVA (A.Z.); Fundació Marató de TV3 20132330 (A.Z.);
the European Foundation for the Study of Diabetes (EFSD) (A.Z.);
“La Caixa” Foundation (A.Z.); Health Research Grant 2021 LCF/
PR/HR21/52410007 (A.Z.); ICREA “Academia” Award Generalitat
de Catalunya (A.Z.); institutional funding from MINECO through the
Centres of Excellence Severo Ochoa Award (A.Z.); CERCA
Programme of the Generalitat de Catalunya (A.Z. and M.O.);
MICINN PDI2021-122478NB-I00 (M.O.); Generalitat de Catalunya
grant 2021SGR00863, “BioExcel-3: Centre of Excellence for
Computational Biomolecular Research” ref. no. 101093290
HORIZON-EUROHPC-JU-2021-COE-01, 101094651 — MDDB —
HORIZON-INFRA-2022-DEV-01 (M.O.); European Research Council
(ERC) GA282280 (L.S.); Muscular Dystrophy Association (MDA)
4165 (L.S.); EFSD/Novo Nordisk Program for Diabetes Research in
Europe (L.S.); Fondation Leducq TNE15004 (L.S.); Ministero
dell’Istruzione, dell’Università e della Ricerca RBAP11Z3YA_005
(L.S.); Ministero dell’Istruzione, dell’Università e della Ricerca
2017BF3PXZ (L.S.); Ministero dell’Università e della Ricerca
2020PKLEPN_002 (L.S.); and SOE_0000181, MUR Concession
Decree no. 564 of 13/12/2022, CUP C93C22007650006, funded
p
under the National Recovery and Resilience Plan (NRRP), Mission
4, Component 2, Investment 1.2, MUR Call for tender n. 367 of
7/10/2022 funded by the European Union – NextGenerationEU
(S.S.). Author contributions: Conceptualization: D.N., M.O., A.Z.,
and L.S. Investigation: D.N., M.I.H.-A., S.S., M.W., S.I., O.M.d.B., A.Q.,
J.H., M.P., P.A., J.C., L.L., D.S., S.F.-V., J.V., and J.J. Visualization:
D.N., M.I.H.-A., S.S., M.W., S.I., O.M.d.B., A.Q., J.H., M.P., P.A., J.C.,
L.L., D.S., S.F.-V., J.V., J.J., M.O., A.Z., and L.S. Funding acquisition:
S.S., M.O., A.Z., and L.S. Project administration: A.Z. and L.S.
g
Supervision: A.Z. and L.S. Writing – original draft: D.N., A.Z., and
L.S. Writing – review & editing: D.N., M.I.H.-A., S.S., M.O., A.Z., and
L.S. Competing interests: The authors declare that they have
no competing interests. Data and materials availability: All data
are available in the main text or the supplementary materials. Raw
y
data are available as supplementary materials. Materials are
available from the corresponding authors. License information:
Copyright © 2023 the authors, some rights reserved; exclusive
licensee American Association for the Advancement of Science. No
claim to original US government works. https://www.science.org/
about/science-licenses-journal-article-reuse
SUPPLEMENTARY MATERIALS
science.org/doi/10.1126/science.adh9351
Figs. S1 to S16
MDAR Reproducibility Checklist
Data S1 and S2
y g
View/request a protocol for this paper from Bio-protocol.
Submitted 23 March 2023; accepted 12 May 2023
10.1126/science.adh9351
13 of 13
RES EARCH

◥ could channel nascent chains without signal

RESEARCH ARTICLE sequences into cytosolic protein maturation
pathways, where they would first need to be
CELL BIOLOGY processed by METAPs. Consistent with such

NAC controls cotranslational N-terminal methionine

a role, knockdown of NAC by RNA interfer-
ence (RNAi) resulted in strongly reduced lev-

excision in eukaryotes
els of ribosome-associated METAP1 in both
Caenorhabditis elegans (Fig. 1B) and human
cells (fig. S2A). By contrast, ribosome binding
Martin Gamerdinger1†*, Min Jia2†, Renate Schloemer1, Laurenz Rabl1, Mateusz Jaskolowski2, of METAP2 increased in NAC-depleted animals
Katrin M. Khakzar1, Zeynel Ulusoy1, Annalena Wallisch1, Ahmad Jomaa2‡, Gundula Hunaeus1, (Fig. 1B), suggesting a compensatory mecha-
Alain Scaiola2, Kay Diederichs3, Nenad Ban2*, Elke Deuerling1* nism for the loss of METAP1. Moreover, me-
thionine cleavage of a METAP model substrate
N-terminal methionine excision from newly synthesized proteins, catalyzed cotranslationally by (fig. S2B) was markedly impaired in NAC-
methionine aminopeptidases (METAPs), is an essential and universally conserved process that plays a depleted worms and worsened when METAP2
key role in cell homeostasis and protein biogenesis. However, how METAPs interact with ribosomes was additionally knocked down (Fig. 1C). There-
and how their cleavage specificity is ensured is unknown. We discovered that in eukaryotes the nascent fore, the in vivo function of METAP1 seems to
polypeptide–associated complex (NAC) controls ribosome binding of METAP1. NAC recruits METAP1 rely on NAC.
using a long, flexible tail and provides a platform for the formation of an active methionine excision To test for a possible physical interaction
complex at the ribosomal tunnel exit. This mode of interaction ensures the efficient excision of between NAC and METAP1 on ribosomes, we
methionine from cytosolic proteins, whereas proteins targeted to the endoplasmic reticulum are performed ribosome-binding studies in vitro

p
spared. Our results suggest a broader mechanism for how access of protein biogenesis factors to using purified human NAC and METAP1. In the
translating ribosomes is controlled. absence of NAC, METAP1 bound only weakly
to ribosomes (Fig. 1D and fig. S2C). The addi-

N
tion of NAC strongly enhanced ribosome asso-
-terminal methionine excision from new- that direct proteins to the endoplasmic reticulum ciation of METAP1 (Fig. 1D and fig. S2C). By
ly synthesized proteins is a universally (ER) (fig. S1). Instead of METAPs, these sub- contrast, METAP2 did bind effectively to ribo-

conserved process that ensures the sta-
bility, folding, and function of cellular
proteins (1, 2). Cleavage is catalyzed by
specific metalloproteases called methionine
aminopeptidases (METAPs), which cotrans-
strates require binding of the ER-targeting
factor signal recognition particle (SRP) (11–14),
which competes with METAPs for access to
the ribosome exit site (10). Forced N-terminal
processing of ER signal peptides by METAPs
g
somes in vitro in the absence of NAC, and the
addition of NAC displaced METAP2 from ribo-
somes (fig. S2D). These observations suggest
a direct role of NAC in stabilizing METAP1 on
ribosomes, whereas METAP2 binds indepen-

lationally process their substrates at the exit of
the ribosomal tunnel (3–5). Cleavage requires
the free N terminus of an elongating substrate
to enter a deep methionine-binding pocket in
the core of the protease, where the active site
is located (6, 7). To reach the active site, the
second amino acid after methionine must be
small and uncharged (e.g., A, C, G, P, S, T, or V)
(8). In eukaryotes, two structurally related
METAPs are expressed in the cytosol, METAP1
[and subsequent acetylation of the neo–N ter-
minus by N-acetyltransferases (15, 16)] inhibits
translocation of proteins across the ER mem-
brane (17), suggesting that binding of METAPs
to ribosomes must be tightly regulated by an
unknown factor in vivo. However, how eukary-
otic METAPs interact with ribosomes and how
their binding is spatially and temporally co-
ordinated with other nascent chain–interacting
factors at the ribosome is unknown.
y
dently of NAC.
We then used AlphaFold (27) to predict the
structural basis of the interaction between NAC
and METAP1. The obtained model suggested
an interaction of the NACb C terminus with the
N-terminal zinc-finger domain of METAP1 (Fig.
1E and fig. S3, A and B). Zinc-finger domains
are widespread interaction modules that can
bind DNA, RNA, or proteins (28). This domain
is present only in eukaryotic METAP1, not in

and METAP2, which are thought to have sim-
ilar substrate specificity but different ribosome
interaction mechanisms (6, 7, 9, 10). These
enzymes lack a substrate recognition interface
that would allow them to distinguish between
NAC recruits METAP1 to translating ribosomes
using a long flexible tail
We sought to identify factors that regulate ac-
cess of METAPs to translating ribosomes in
y g
METAP2 or prokaryotic METAP1 (29). The en-
zymatic activity of human METAP1 lacking
the zinc-finger domain (D2-70, referred to as
DN-METAP1) was comparable to that of wild-
type METAP1 in vitro, suggesting that this do-

substrates from different cellular compart-
ments. If their ribosome binding were not
regulated, they would cleave methionine from
any nascent substrate that has a small and
uncharged amino acid at the second position,
including proteins with N-terminal presequences
1
Department of Biology, Molecular Microbiology, University of
Konstanz, 78457 Konstanz, Germany. 2Department of
Biology, Institute of Molecular Biology and Biophysics, ETH
Zurich, 8093 Zurich, Switzerland. 3Department of Biology,
Molecular Bioinformatics, University of Konstanz, 78457
Konstanz, Germany.
*Corresponding author. Email: elke.deuerling@uni-konstanz.de
(E.D.); ban@mol.biol.ethz.ch (N.B.);
martin.gamerdinger@uni-konstanz.de (M.G.)
†These authors contributed equally to this work.
‡Present address: Department of Molecular Physiology and Biological
Physics, University of Virginia, Charlottesville, VA 22903, USA.
eukaryotes. A candidate for this function is
the nascent polypeptide–associated complex
(NAC), an heterodimeric complex consisting
of NACa and NACb that is bound at the tun-
nel exit of virtually all ribosomes in the cell
(18–21). NAC consists of a central globular di-
merization domain from which four long,
flexible tails (N and C termini) protrude (Fig.
1A) (22, 23). The globular domain and the
NACb N terminus interact with the ribosome
near the tunnel exit, where NAC monitors the
N termini of nascent chains for the presence of
ER-targeting signals. NAC then directs these
substrates into the cotranslational ER protein
transport pathway by recruiting SRP to ribo-
somes through the UBA domain in the NACa
C terminus (Fig. 1A) (24–26). Conversely, NAC
,
main does not affect the protease active site
(fig. S3C). Zinc-finger domains can be unfolded
by depleting the bound zinc ions using metal
complexing agents (30). Pretreatment of puri-
fied METAP1 with the metal chelator EDTA
almost completely reversed the stabilizing ef-
fect of NAC on ribosome binding of METAP1
(fig. S3D), consistent with an interaction of NAC
with the zinc-finger domain as predicted by
AlphaFold.
According to the AlphaFold model, the METAP1
zinc-finger domain is bound by a conserved
hydrophobic motif (146VPDLV150) located near
the end of the NACb C-terminal arm (Fig. 1A and
fig. S3A). The molecular details of this interac-
tion closely resemble the interaction of METAP1
with the metallochaperone ZNG1, which binds

Gamerdinger et al., Science 380, 1238–1243 (2023) 23 June 2023 1 of 6

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

p
g
Fig. 1. A motif in the NACb C-terminal arm binds to the zinc-finger domain METAP1 in vitro in the presence of the indicated NAC variants. (E) AlphaFold
of METAP1. (A) Model showing domain architecture of NAC consisting of a model of the NACb-METAP1 interaction. The N-terminal zinc-finger domain

central heterodimeric globular domain and four flexible arms (N and C termini).
The conserved motif (146VPDLV150) in the NACb C terminus is analyzed in panels
(D) to (G). (B) Ribosome association of METAPs after knockdown of NAC in
C. elegans. Proteins in total and ribosomal pellet fractions were detected by
immunoblotting. (C) Methionine cleavage of a METAP model substrate (1433g)
after knockdown of NACb and METAP2 in C. elegans. Uncleaved substrate
containing the initiator methionine (iMet) was detected by an epitope-specific
antibody. Asterisk indicates a nonspecific band. (D) Ribosome binding of human
y
of METAP1 (green) and the C terminus of NACb (purple) are shown. Interacting
residues are shown as spheres. Arrow marks incorporation site of a photo–cross-
linking amino acid for analysis in (G). (F) Ribosome binding of zinc-finger
mutants of METAP1 in vitro in presence of wild-type NAC. (G) Photo–cross-
linking of NAC containing the UV-activated photo–cross-linking amino acid benzoyl-
phenylalanine (Bpa) at position V150 indicated in (E). Photo–cross-linking was
performed with wild-type METAP1 or a variant in which the two predicted NACb-
interacting residues were replaced with serine (L20S/F41S).

to the zinc-finger domain of METAP1 using a
similar hydrophobic motif (30, 31). To test the
predicted hydrophobic interactions between
the NACb motif residues V146 and L149 and
the residues L20 and F41 in the METAP1 zinc
bic motif in the long, flexible C-terminal arm
of NACb.
The binding affinity of the NACb C-terminal
tail to the METAP1 zinc-finger domain was high
(KD = 0.16 mM ± 0.02) and comparable to that
y g
of METAP1 in vivo. In agreement, the METAP1
mutant lacking the N-terminal zinc-finger do-
main (DN-METAP1) was almost undetect-
able in the ribosomal pellet fraction of worms
(Fig. 2B). Worms expressing mutant DC-NACb
or DN-METAP1 also showed a strong methio-

finger (Fig. 1E and fig. S4), we replaced these
residues with serines (NACb V146S/L149S and
METAP1 L20S/F41S). Additionally, we inves-
tigated deletion mutants lacking the C termi-
nus of NACb (D140-162, referred to as DC-NACb)
or the N terminus of METAP1 (DN-METAP1).
All mutations almost completely abolished
the stabilizing effect of NAC on ribosome
binding of METAP1 in vitro (Fig. 1, D and F).
Moreover, a NAC variant carrying a photo–
cross-linking probe at position V150 of NACb
directly adjacent to the predicted METAP1-
interacting residues (Fig. 1E and fig. S3B)
cross-linked strongly with wild-type METAP1,
but not with the L20S/F41S mutant (Fig. 1G).
Thus, NAC interacts with the METAP1 zinc-
finger domain through a conserved hydropho-
of ZNG1, which shares a similar interaction
motif (fig. S3, E and F) (30). This suggests that
the NACb C terminus provides key binding
affinity for METAP1 on ribosomes. We there-
fore tested the importance of this interaction
for N-terminal methionine excision in vivo.
Consistent with the in vitro studies, ribosome
binding of METAP1 was strongly reduced in
C. elegans worms expressing the DC-NACb
variant, although like wild-type NAC, this NAC
mutant efficiently bound to ribosomes (Fig. 2A).
Ribosome association of METAP1 was also
strongly decreased in human cells expressing
the V146S/L149S or DC-NACb mutant variants
(fig. S3, G and H). Thus, the interaction of
the NACb C terminus with the METAP1 zinc-
finger domain facilitates ribosome binding
,
nine excision defect in the METAP2 RNAi
background (Fig. 2, C and D). Moreover, the
viability of animals expressing the mutant
NAC or METAP1 variants was severely re-
duced compared with wild-type animals when
METAP2 expression was suppressed (Fig. 2,
E and F). Thus, tethering of METAP1 to trans-
lating ribosomes through the flexible C-terminal
NACb arm is critical for methionine excision
in vivo.
Mechanism of ribosome-nascent chain
engagement by METAP1
The C terminus of NACb and the N terminus
of METAP1 are both flexibly tethered to the
central domains of the proteins (6, 26, 30).

Gamerdinger et al., Science 380, 1238–1243 (2023) 23 June 2023 2 of 6

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

domain of METAP1 was found to interact

with both the ribosome and the globular
domain of NAC, orienting the methionine-
binding pocket toward the exit of the ribo-
somal tunnel (Fig. 3C). The distance between
the active site of the enzyme and the exit of
the ribosomal tunnel is consistent with the
cotranslational processing of nascent chains
when they reach a length of ~50 amino acids
(Fig. 3C and fig. S9). The N-terminal zinc-finger
domain of METAP1 and the C-terminal arm
of NACb were not resolved, suggesting high
flexibility even in the ribosome-bound ter-
nary complex (Fig. 3B).
In the cryo-EM reconstruction, the globular
domain of NAC was seen in contact with the
central METAP1 domain in a region where
a ribosome-anchoring domain is found in
METAP2-related proteins (Fig. 4A and fig.
S10). This explains why the globular domain
of NAC displaced METAP2 but not METAP1

p
from ribosomes (fig. S2D). The interaction
with the NAC globular domain could also ex-
plain why NAC stabilized METAP1 binding
even when it was missing the C-terminal tail,
as in the DC-NACb mutant (fig. S5A). To inves-
tigate this, we performed in vitro binding studies
with DC-NACb together with two METAP1 mu-

g
tants designed to disrupt the interaction with
the globular domain of NAC: 77PWAGY/AAAAA81
and 340WPD/AAA342 (Figs. 3D and 4A). As ex-
pected, these mutants were no longer stabi-

y
lized on ribosomes in the context of DC-NACb
(Fig. 3E). However, no substantial difference
in binding was detected in the presence of
wild-type NAC, indicating that the C-terminal
contact of NACb with the METAP1 zinc finger
is the key contributor to the binding affinity
Fig. 2. NAC recruits METAP1 to ribosomes for methionine cleavage in vivo. (A) Ribosome association of (fig. S5B).
endogenous METAP1 in C. elegans worms expressing the indicated FLAG-tagged NACb variants. Proteins The cryo-EM structure also revealed a direct
in the total and ribosomal pellet fractions were detected by immunoblotting. (B) Ribosome association of ribosomal contact of METAP1 involving the
FLAG-tagged METAP1 variants with indicated mutations in C. elegans. METAP1 levels in the total and ribosomal protein uL23 and the ribosomal

y g
ribosomal pellet fractions were analyzed by immunoblotting. (C and D) Methionine cleavage of a METAP RNA (rRNA) helix H59, a known docking site
model substrate (1433g) in worms as in (A) and (B), respectively, after knockdown of METAP2. Uncleaved for several nascent chain–associated factors
substrate (iMet) was detected with an epitope-specific antibody. (E and F) Viability or worms as in (A) and (B), (Figs. 3, B and D, and 4, B and C) (9, 14, 33, 34).
respectively, after knockdown of METAP2. Graph shows the number of progeny in each mutant strain relative To investigate the importance of these interac-
to worms expressing wild-type NACb or METAP1 (set to 100%). Data are shown as means ± SD (N = 3). tions, we mutated METAP1 residues that con-
tact uL23 (317VMK/AAA319 and 127QIK/AAA129)

Therefore, an interaction between these flex-
ible arms only ensures high local concentra-
tion of METAP1 in the vicinity of the ribosomal
tunnel exit. However, because METAP1 did
bind weakly to ribosomes in vitro in the ab-
sence of NAC (fig. S2C), some direct interac-
tions between METAP1 and the ribosome may
exist. Moreover, the DC-NACb variant still
showed a weak stabilizing effect on METAP1
ribosome binding (Fig. 1D and fig. S5A), sug-
gesting a secondary contact between NAC
and METAP1.
To understand how METAP1 is positioned
on ribosomes to engage its substrates, we puri-
fied ribosome nascent chain complexes (RNCs)
carrying a cytosolic nascent METAP substrate
(GPI; 65 amino acids) and determined its struc-
ture in complex with wild-type NAC and a metal-
binding deficient METAP1 variant with low
catalytic activity (D220N; fig. S6) (32) by single-
particle cryo–electron microscopy (cryo-EM).
The reconstruction showed the ribosome with
a peptidyl-transfer RNA, the GPI nascent chain
within the ribosomal tunnel, and both NAC
and METAP1 bound at the tunnel exit (Fig. 3,
A to C, and figs. S7 to S9). The globular domain
of NAC was found near the exit of the ribo-
somal tunnel as in previous structures (26),
including the N terminus of NACb anchored
between eL22 and eL19 (Fig. 3B). The catalytic
,
and H59 (289RS/AA290 + 308HY/AA309, referred
to as H59mut; fig. S3C). The mutated clusters
of residues are conserved and located in loop
regions of METAP1 (Fig. 4, B and C). In the
absence of NAC, the binding of these mutants
to ribosomes in vitro was strongly reduced
(Fig. 3F). Addition of NAC restored ribosome
binding of the uL23-binding mutants almost
to wild-type levels, whereas binding of the
H59mut variant was still substantially re-
duced (fig. S5B).
These data suggest that H59 is the most im-
portant ribosome contact of METAP1 and that
perturbing it would negatively affect METAP1
function. Therefore, we introduced this muta-
tion in C. elegans METAP1 for in vivo analysis.

Gamerdinger et al., Science 380, 1238–1243 (2023) 23 June 2023 3 of 6

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

p
g
y
Fig. 3. Mechanism of ribosome-nascent chain engagement by METAP1. (E) Ribosome binding of METAP1 variants carrying alanine mutations in the
(A) Cryo-EM structure of the RNC-NAC-METAP1 complex. 40S and 60S proteins are binding interfaces indicated in (D) in the presence of NAC lacking the NACb

colored in yellow and light blue, respectively; rRNA is shown in gray. Dashed box
indicates the magnified region shown in (B). (B) Close-up view of the ribosomal
tunnel exit. NACa is colored in yellow green, NACb in purple, and METAP1 in
dark green. Dashed lines show where the flexible NACb C-terminal arm and the
METAP1 zinc finger would protrude. (C) Cross section of map visualizing the
y g
C terminus (DC-NACb). (F) Ribosome binding of METAP1 variants in vitro in the
absence of NAC at low salt concentration (30 mM KOAc). Residues indicated in (D)
that interact with uL23 or H59 were replaced with alanines. (G) Ribosome binding of
METAP1 in the presence of the indicated NAC variants. KK-EE refers to a charge-
reversal NAC mutant deficient in binding its globular domain to the ribosome.

METAP1 catalytic center. Dashed line indicates a potential path of a nascent
chain 50 amino acids in length. Residues critical for catalytic activity (H203 and
H301) are shown as sticks. (D) Surface representation showing METAP1-ribosome
and METAP1-NAC interfaces (red). Left: interfaces on the ribosome and NAC.
Dashed outline indicates position of METAP1. Right: interfaces on METAP1.
,
(H) Methionine cleavage of a METAP model substrate (1433g) in C. elegans
expressing NAC variants as in (G) after knockdown of METAP2. Uncleaved substrate
(iMet) was detected by an epitope-specific antibody. (I) Ribosome binding of
METAP1 to NAC-bound RNCs translating either a cytosolic (GPI) or an ER (HSPA5)
substrate. SRP54 was used as a marker for the ER signal sequence.

Consistent with the in vitro results, disruption
of the H59 contact resulted in decreased ribo-
some binding of METAP1 in vivo (Fig. 2B).
Moreover, similar to DN-METAP1–expressing
worms, methionine cleavage of a model sub-
strate was strongly reduced (Fig. 2D). Additional
depletion of METAP2 in animals expressing
the H59 mutant also strongly impaired viabil-
ity of worms (Fig. 2F). These data suggest that
the ribosomal contact of METAP1 to H59 is
crucial for methionine excision by ensuring
the proper positioning of the enzymatic active
site in front of the tunnel exit to capture the N
terminus of a nascent chain (Fig. 3C).
ER signal sequences prevent docking of
METAP1 to NAC-bound ribosomes
Our data show that the function of METAP1
depends on the formation of a ternary complex
with NAC at the exit of the ribosomal tunnel.
In the complex, NAC is positioned on the ribo-
some with its globular domain bound in the
vicinity of the ribosomal tunnel exit (Fig. 3, A
to D). In this conformation, NAC prevents the
binding of the ER-targeting factor SRP (26).
However, a nascent chain with an N-terminal
ER-targeting signal sequence destabilizes the
globular domain of NAC to enable SRP bind-
ing (26). Therefore, considering that the con-
formation of NAC will depend on the nature
of the synthesized nascent chains, NAC will
allow functional docking of METAP1 only for
sequences that are lacking an ER-targeting

Gamerdinger et al., Science 380, 1238–1243 (2023) 23 June 2023 4 of 6

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

Fig. 4. METAP1 interactions with the ribosome and the NAC globular Fig. 3, G and H. (B) Close-up view of the uL23-METAP1 interface. The residues
domain. (A) Close-up view of the binding region between METAP1 and the shown as spheres that contact a short helix of uL23 were mutated to alanines in

NAC globular domain. METAP1 residues shown as spheres were mutated to
alanines in the analysis shown in Fig. 3E. Sequence alignment below shows their
conservation. The indicated NAC residues K78 and K43 in the ribosome-binding
helices were mutated to glutamic acid in the KK-EE NAC variant shown in
p
Fig. 3F. Conservation of residues is shown below. (C) Close-up view of the
METAP1 interaction with rRNA helix H59. METAP1 residues shown as red spheres
were mutated to alanines (H59mut) in the analyses shown in Figs. 2 and 3F.
Sequence alignment below shows their conservation.

signal when the NAC globular domain is tightly tunnel exit by destabilizing the NAC globular and the NAC globular domain to the ribosome.

bound at the exit of the ribosomal tunnel.
To investigate whether ribosome binding of
the NAC globular domain is critical for METAP1
function, we used a previously established NAC
variant with two charge reversal mutations in
domain. Therefore, the conformational change
in NAC triggered by ER-targeting signals re-
stricts METAP1 binding and at the same time
promotes SRP binding, whereas the effect is
exactly the opposite when no ER signal se-
g
This mechanism explains why the N termini
of ER-targeting presequences, of which nearly
50% would be susceptible to methionine exci-
sion in humans (fig. S1), are not modified in
vivo (17), because they disrupt the composite

the positively charged helices that attach the
globular domain to the tunnel exit: NACa K78E
and NACb K43E, referred to here as KK-EE NAC
(Fig. 4A) (26). This NAC variant still binds to
ribosomes through the high-affinity N-terminal
NACb anchor, whereas binding of the globular
domain is weakened (26). We found that re-
cruitment of METAP1 to ribosomes by KK-EE
NAC was strongly reduced in vitro (Fig. 3G),
and worms expressing this variant showed a
quence is present. These observations show
that although METAP1 cannot distinguish
between ER and cytosolic nascent substrates
on its own, the high specificity for cytosolic
nascent chains is cotranslationally mediated
by NAC.
Discussion
On the basis of our findings, we propose a mech-
anistic model for how eukaryotic METAP1 can
y
binding platform for METAP1 on the ribosome
by detaching the NAC globular domain. There-
fore, signal sequences that destabilize NAC
will simultaneously prevent METAP1 binding
while allowing SRP to access the nascent chain
at the ribosomal tunnel exit. SRP recruitment
then occurs through the UBA domain at the
end of the long, flexible C-terminal arm of NACa
to initiate ER protein targeting (26). This mech-
anism is likely to be broadly conserved in eu-

marked methionine excision defect in vivo
(Fig. 3H). Thus, methionine cleavage by METAP1
requires that the globular domain of NAC is
bound to the tunnel exit of the ribosome. This
suggests that ER-targeting signals that de-
specifically engage nascent substrates for me-
thionine excision (Fig. 5). As observed here and
in a previous study (26), the highly abundant
NAC binds to virtually all translating ribosomes
in the cytosol through its NACb anchor do-
y g
karyotes, with possible additional roles of rRNA
expansion segments in yeast, where NAC is
not essential (10, 35–37).
These findings show that NAC functions as
a molecular control hub at the ribosomal exit

stabilize the globular domain of NAC also de-
stabilize METAP1 binding. To investigate this,
we used RNCs carrying a nascent chain (65
amino acids) of the ER substrate HSPA5, which
contains an N-terminal signal sequence (26).
Binding of METAP1 to NAC-bound HSPA5-
RNCs was strongly reduced compared with
RNCs translating a cytosolic protein (GPI; 65
amino acids) (Fig. 3I), suggesting that ER sig-
nal peptides prevent a functional docking of
METAP1 to the ribosome exit site. Consistent
with this, fusion of an ER signal peptide to a
cytosolic nascent substrate effectively prevents
methionine excision in a cell-free in vitro trans-
lation system (fig. S5C). These data are con-
sistent with the idea that ER-targeting signals
disrupt the METAP1-binding platform at the
main. The NAC globular domain is positioned
next to the tunnel exit, where it can sense the
sequence of the nascent chain emerging from
the ribosomal tunnel. In the presence of an ER-
targeting signal, the globular domain of NAC
dissociates and allows SRP binding, whereas
in the absence of an ER-targeting signal, the
globular domain remains bound. METAP1 is
likely directed to the ribosome-NAC complex
through high-affinity interactions between the
long, flexible C-terminal arm of NACb and
the N-terminal zinc-finger domain of METAP1.
Our data show that formation of the produc-
tive complex, in which the catalytic pocket of
METAP1 is optimally oriented toward the exit
of the ribosomal tunnel to cleave methionine,
requires simultaneous docking of METAP1
,
site, ensuring efficient and specific binding of
METAP1 and SRP to their respective nascent
polypeptides emerging from the ribosomal
tunnel. With its two flexible C-terminal tails,
NAC ensures that the nascent chain–interacting
factors are present in the vicinity of the ribo-
somal tunnel, whereas its globular domain con-
trols the handover of the nascent polypeptide
to the designated factor in a sequence-specific
manner. This mechanism improves the spec-
ificity of N-terminal methionine excision and
the fidelity of ER protein transport, both es-
sential cellular processes. Our study further
demonstrates that methionine excision in eu-
karyotes includes METAP2 as a redundant
backup system that acts independently of NAC.
How METAP2 gains regulated access to the

Gamerdinger et al., Science 380, 1238–1243 (2023) 23 June 2023 5 of 6

RES EARCH | R E S E A R C H A R T I C L E
Fig. 5. Model of selective factor recruitment by NAC. NAC binds all ribosomes through the N-terminal
NACb anchor domain. On cytosolic RNCs, the NAC globular domain is attached to the ribosomal tunnel exit
and forms a binding platform for METAP1 while blocking SRP binding. The local concentration of METAP1
near the exit of the ribosomal tunnel is increased through tight but flexible binding between a hydrophobic motif
in the C-terminal arm of NACb and the zinc-finger domain of METAP1 (left). An emerging ER signal sequence
disrupts the METAP1-binding platform by detaching the NAC globular domain from the exit site. This also
releases the SRP-binding site at the tunnel exit, leading to specific recruitment of SRP through the C-terminal
arm of NACa and its UBA domain (right).
ribosome exit site in the presence of NAC is
unclear. In addition to SRP and METAPs, a
variety of other nascent chain–processing fac-
tors require regulated access to the ribosome
exit site, including N-terminal modifying en-
zymes (38–40), chaperones (41, 42), transport
factors (43–45) and quality control factors
(46, 47). Understanding the potential interplay
of these factors with NAC appears to be the
key to understanding the mechanisms and the
regulation of cotranslational protein biogene-
sis in human cells.
RE FE RENCES AND N OT ES
1. R. A. Bradshaw, W. W. Brickey, K. W. Walker, Trends Biochem.
Sci. 23, 263–267 (1998).
2. C. Giglione, A. Boularot, T. Meinnel, Cell. Mol. Life Sci. 61,
1455–1474 (2004).
3. C. Giglione, S. Fieulaine, T. Meinnel, Biochimie 114, 134–146
(2015).
4. G. Kramer, D. Boehringer, N. Ban, B. Bukau, Nat. Struct. Mol.
Biol. 16, 589–597 (2009).
5. C. I. Yang, H. H. Hsieh, S. O. Shan, Proc. Natl. Acad. Sci. U.S.A.
116, 23050–23060 (2019).
Gamerdinger et al., Science 380, 1238–1243 (2023)
6. A. Addlagatta, X. Hu, J. O. Liu, B. W. Matthews, Biochemistry
44, 14741–14749 (2005).
7. S. Liu, J. Widom, C. W. Kemp, C. M. Crews, J. Clardy, Science
282, 1324–1327 (1998).
8. Q. Xiao, F. Zhang, B. A. Nacev, J. O. Liu, D. Pei, Biochemistry
49, 5588–5599 (2010).
9. K. Wild et al., Nat. Commun. 11, 776 (2020).
10. Y. Nyathi, M. R. Pool, J. Cell Biol. 210, 287–301 (2015).
11. R. Gilmore, G. Blobel, P. Walter, J. Cell Biol. 95, 463–469
(1982).
12. M. Halic et al., Nature 444, 507–511 (2006).
13. R. M. Voorhees, R. S. Hegde, eLife 4, e07975 (2015).
14. A. Jomaa et al., Cell Rep. 36, 109350 (2021).
15. K. K. Starheim, K. Gevaert, T. Arnesen, Trends Biochem. Sci.
37, 152–161 (2012).
16. S. Deng, R. Marmorstein, Trends Biochem. Sci. 46, 15–27
(2021).
17. G. M. Forte, M. R. Pool, C. J. Stirling, PLOS Biol. 9, e1001073
(2011).
18. M. del Alamo et al., PLOS Biol. 9, e1001100 (2011).
19. M. Gamerdinger et al., Mol. Cell 75, 996–1006.e8 (2019).
20. B. Wiedmann, H. Sakai, T. A. Davis, M. Wiedmann, Nature 370,
434–440 (1994).
21. U. Raue, S. Oellerer, S. Rospert, J. Biol. Chem. 282, 7809–7816
(2007).
22. Y. Liu, Y. Hu, X. Li, L. Niu, M. Teng, Biochemistry 49,
2890–2896 (2010).
23. L. Wang et al., Protein Cell 1, 406–416 (2010).
23 June 2023
24. M. Gamerdinger, M. A. Hanebuth, T. Frickey, E. Deuerling,
Science 348, 201–207 (2015).
25. H. H. Hsieh, J. H. Lee, S. Chandrasekar, S. O. Shan, Nat. Commun.
11, 5840 (2020).
26. A. Jomaa et al., Science 375, 839–844 (2022).
27. J. Jumper et al., Nature 596, 583–589 (2021).
28. S. S. Krishna, I. Majumdar, N. V. Grishin, Nucleic Acids Res. 31,
532–550 (2003).
29. S. M. Arfin et al., Proc. Natl. Acad. Sci. U.S.A. 92, 7714–7718
(1995).
30. A. Weiss et al., Cell 185, 2148–2163.e27 (2022).
31. M. Pasquini et al., Cell Rep. 39, 110834 (2022).
32. S. Mitra, K. M. Job, L. Meng, B. Bennett, R. C. Holz, FEBS J.
275, 6248–6259 (2008).
33. A. Jomaa et al., Nat. Commun. 8, 15470 (2017).
34. L. Kater et al., EMBO Rep. 20, e48191 (2019).
35. K. Fujii, T. T. Susanto, S. Saurabh, M. Barna, Mol. Cell 72,
1013–1020.e6 (2018).
36. J. A. Vetro, Y. H. Chang, J. Cell. Biochem. 85, 678–688 (2002).
37. S. Zuo, Q. Guo, C. Ling, Y. H. Chang, Mol. Gen. Genet. 246,
247–253 (1995).
38. U. A. Friedrich et al., Cell Rep. 34, 108711 (2021).
39. A. G. Knorr et al., Nat. Struct. Mol. Biol. 26, 35–39 (2019).
40. S. Varland, C. Osberg, T. Arnesen, Proteomics 15, 2385–2401
(2015).
41. Y. Chen, B. Tsai, N. Li, N. Gao, Nat. Commun. 13, 3410
(2022).
p
42. Y. Zhang et al., Nat. Commun. 11, 1504 (2020).
43. S. Pfeffer et al., Nat. Commun. 6, 8403 (2015).
44. L. Smalinskaitė, M. K. Kim, A. J. O. Lewis, R. J. Keenan,
R. S. Hegde, Nature 611, 161–166 (2022).
45. R. M. Voorhees, I. S. Fernández, S. H. Scheres, R. S. Hegde,
Cell 157, 1632–1643 (2014).
46. S. Shao, A. Brown, B. Santhanam, R. S. Hegde, Mol. Cell 57,
433–444 (2015).
47. P. S. Shen et al., Science 347, 75–78 (2015).
g
AC KNOWLED GME NTS
We thank all laboratory members for discussions, especially
M. Leibundgut for the help with interpretations of electron
microscopy maps, T. Schoenhut for expressing METAP1 mutants
y
for binding studies, and M. Kovermann and Y. Werle for help with
the spectrofluorometer. Funding: This work was supported by
the German Science Foundation (grants SFB969/A01 and A07 to
E.D. and M.G.), the Swiss National Science Foundation (SNSF
grant 310030_212308 to N.B.), the National Center of Excellence
in Research RNA & Disease Program of the SNSF (grant
51NF40-205601 to N.B.), and the European Research Council
(Synergy Grant 101072047 CoTransComplex to N.B.). Author
contributions: M.G., M.Ji., N.B., and E.D. conceived the project.
M.G. and R.S. performed in vitro biochemical analyses with help
from Z.U. and A.W. M.G. and L.R. performed C. elegans in vivo
experiments. K.K. purified RNCs and performed RNC binding
studies. G.H. performed human cell-based studies. M.Ji. prepared
y g
the cryo-EM samples and collected the data with help of A.J.
and A.S. M.Ja. and M.Ji. processed the cryo-EM data and created
the structural figures. M.Ja. adjusted and refined the molecular
model. K.D. helped to interpret the structural data. M.G., M.Ji.,
N.B., and E.D. wrote the manuscript. All authors contributed
to data analysis and editing the final version of the manuscript.
Competing interests: The authors declare no competing interests.
Data and materials availability: Cryo-EM maps and model
,
coordinates have been deposited in the Electron Microscopy Data
Bank (EMDB) and Protein Data Bank (PDB) under accession
codes EMD-17367 and PDB ID 8P2K, respectively. License
information: Copyright © 2023 the authors, some rights reserved;
exclusive licensee American Association for the Advancement of
Science. No claim to original US government works. https://www.
science.org/about/science-licenses-journal-article-reuse
SUPPLEMENTARY MATERIALS
science.org/doi/10.1126/science.adg3297
Materials and Methods
Figs. S1 to S10
Tables S1 to S3
References (48–64)
MDAR Reproducibility Checklist
View/request a protocol for this paper from Bio-protocol.
Submitted 16 December 2022; resubmitted 11 April 2023
Accepted 18 May 2023
10.1126/science.adg3297
6 of 6
RES EARCH

SUPERCONDUCTIVITY tem formed dilute configurations of conven-

tional vortices that coexisted with isolated
Superconducting vortices carrying a temperature- vortex-like objects carrying a small fraction of
magnetic flux (fig. S2). To determine whether
dependent fraction of the flux quantum these objects are vortices and to estimate the
total flux trapped within them, further mea-
Yusuke Iguchi1,2*, Ruby A. Shi1,2,3, Kunihiro Kihou4, Chul-Ho Lee4, Mats Barkman5, surements were performed. By cooling through
Andrea L. Benfenati5, Vadim Grinenko6, Egor Babaev5, Kathryn A. Moler1,2,3,7 Tc with a small local magnetic field from the
scanning SQUID field coil, we also observed
Magnetic field penetrates type-II bulk superconductors by forming quantum vortices that enclose a the conventional vortices and the fractional
magnetic flux equal to the magnetic flux quantum. The flux quantum is a universal quantity that depends vortex–like objects (Fig. 1, C to E, and fig. S3).
only on fundamental constants. In this study, we investigated isolated vortices in the hole-overdoped We observed both the conventional vortex and
Ba1−xKxFe2As2 (x = 0.77) by using scanning superconducting quantum interference device (SQUID) the fractional vortex–like objects appearing at
magnetometry. In many locations, we observed objects that carried only part of a flux quantum, with a the same location in different cooling cycles,
magnitude that varied continuously with temperature. We demonstrated mobility and manipulability of where the sample was cooled down to 10.5 K
these objects and interpreted them as quantum vortices with nonuniversally quantized (fractional) from 25 K with magnetic fields (Fig. 1, C and
magnetic flux whose magnitude is determined by the temperature-dependent parameters of a E). Hence the specific area in the sample did
multicomponent superconductor. not determine the character of a flux-carrying
object. To characterize further the nature of

A
the object, a fractional antivortex object was
classical electrically conducting fluid al- etc. The experimental evidence for flux quanti- created at the same position during different

lows the creation of arbitrary vortices.
By contrast, the fundamental property
of an ordinary superconductor is the
quantization of vorticity (1). This implies
the quantization of the magnetic flux (2) that
these vortices induce. The magnetic flux F
zation comes from a number of measurements,
which in addition to scanning superconduct-
ing quantum interference device (SQUID) mag-
netometry include scanning Hall probes (6)
and magnetic force microscopy. Other probes
that do not measure magnetic flux directly but
p
cooling cycles (Fig. 1D).
To obtain the simplest estimate for the mag-
netic field penetration depth l, we simulated
the magnetic field of the conventional single-
quantum vortex with a point source magnetic
monopole field with the total flux F0 (16). We

(2–4) is defined as the integral of magnetic
field B over the surface of the sample F ¼
∫dxdyB  n^ , where n^ is a unit vector normal
to the surface. This quantization means that
magnetic fields inside a superconductor can
allow assessment of the scaling of vortex den-
sity with applied field, such as Bitter decora-
tion (7) and scanning tunneling microscopy
(8), are also consistent with integer flux quan-
tization. Under certain conditions, immobile
g
roughly estimated the magnetic field pene-
tration depth l from this point source mod-
el. We fitted it with the spacing between the
sample surface and the pickup loop center, z0 =
1250 nm (Fig. 1H), yielding a magnetic field

change only in discrete steps: An externally
applied magnetic field penetrates a type-II
superconductor in increments of F0 associated
with the entrance of quantum vortices from
the boundary of the sample (5). The magnetic
flux quantum F0 = hc/2e (in CGS units) de-
pends only on fundamental constants: the
electron charge e, the speed of light c, and the
Planck constant h. Such quantum vortices
are mobile objects that can move through the
objects carrying a half of the flux quantum
were observed in an intrinsic Josephson junc-
tion formed on grain boundaries of d-wave
superconductors (9) and on small rings of
Sr2RuO4 (10) and b-Bi2Pd (11). In supercon-
ductors consisting of almost decoupled layers
(12) as well as artificial layered structures (13),
the vortex core can move a great distance be-
tween consecutive layers, leading to partial
flux despite the standard integer phase quan-
y
penetration depth l ≈ 2.3 mm. A fitting of the
vortex object in Fig. 1C with l ≈ 2.3 mm and
the fractional point source yielded the frac-
tion of the flux quantum FF ≈ 0.3F0 at T =
11.0 K (Fig. 1F).
By measuring the magnetic flux at different
temperatures, we observed that the flux of this
object was temperature dependent. The mag-
netic field amplitude was also very different
from that of a conventional vortex and half-

superconductor subject to thermal excitation,
a pinning landscape, or forces applied by cur-
rents or electromagnetic fields. Vortex stabil-
ity is dictated by a nontrivial winding of the
phase q of the complex superconducting order
tization of the phase windings in each indi-
vidual layer. Here we report scanning SQUID
magnetometry on Ba1−xKxFe2As2 (x = 0.77).
We show that, along with the standard vortices
carrying single flux quantum, the material
y g
quantum objects that form on grain boun-
daries of a d-wave superconductor (17). We
observed that the amplitude of the peak mag-
netic flux from the fractional vortex–like ob-
ject decreased with decreasing temperature

parameter. The phase changes by 2p around
the vortex core (1).
Despite the diversity of superconducting ma-
terials, vortices carry an integer number (usu-
ally one) of flux quanta, irrespective of the other
properties of superconducting materials such
as composition, temperature, degree of purity,
1
Geballe Laboratory for Advanced Materials, Stanford
University, Stanford, CA 94305, USA. 2Stanford Institute for
Materials and Energy Sciences, SLAC National Accelerator
Laboratory, Menlo Park, CA 94025, USA. 3Department
of Physics, Stanford University, Stanford, CA 94305, USA.
4 magnetic flux on the cleaved ab plane of single
National Institute of Advanced Industrial Science and
Technology (AIST), Tsukuba, Ibaraki 305-8568, Japan.
5
Department of Physics, Royal Institute of Technology, SE-106
91 Stockholm, Sweden. 6Tsung-Dao Lee Institute, Shanghai
Jiao Tong University, Shanghai 201210, China. 7Department of
Applied Physics, Stanford University, Stanford, CA 94305, USA.
*Corresponding author. Email: yiguchi@stanford.edu
has vortex excitations that carry a fraction of
the flux quantum. Notably, the fraction is not
a rational number but instead smoothly varies
with temperature, and hence the flux quanti-
zation is nonuniversal, meaning that it is not a
function of only fundamental constants.
Temperature-dependent fractional vortex flux
To explore the magnetic properties of
Ba1−xKxFe2As2, we used the scanning SQUID
susceptometer to microscopically image the
crystals (Fig. 1, A and B) (14, 15). We conducted
measurements below the superconducting crit-
ical temperature Tc (fig. S1). By cooling the
sample in a small uniform magnetic field of
3.5 mG, we observed that below Tc, the sys-
,
(Fig. 2A). This temperature-dependent flux is
in strong contrast to the conventional vortex. The
usual vortex carries temperature-independent
single flux quantum. Hence the peak amplitude
of magnetic field increases with decreasing tem-
perature, reflecting the temperature dependence
of the magnetic field penetration depth (Fig. 2A).
To estimate the temperature dependence of
the fractional flux FF, we fitted the cross sections
of the fractional vortex–like object at several
temperatures (Fig. 2, B and C), the same way as
in Fig. 1, F and G. For the fitting of the fractional
vortex–like object, we used the same penetra-
tion depth obtained from the fitting of the
conventional vortex at the same region. We
observed fractional vortex–like objects in dif-
ferent areas of the sample (figs. S4 and S5).

Iguchi et al., Science 380, 1244–1247 (2023) 23 June 2023 1 of 4

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

A Ba0.23K0.77 Fe2As2 C D E Conventional

40 40 40
4 Fractional vortex Fractional vortex vortex
0

Φ [mΦ0 ]
region 2

Φ [mΦ0]
y [µm]

20 20 20

Φ [mΦ0]
3 T = 11.0 K
1 2 0 0 0
Φ [mΦ0]
-100 45
-20 -20 -20
-30 -30 -90 -30
-90 -70 -40 -90 -70 -40 -70 -40 y [µm]
-100 0 x [µm] y [µm] x [µm] y [µm] x [µm] -45
x [µm]

B 1 µm
F G H Φ0 point source
40 40 40 model
Pickup
0.3Φ0 point source model -0.3Φ0 point source model
loop Simulation

Φ [mΦ0 ]
Φ [mΦ0]
20 20 20
Φ [mΦ0 ]

Field coil λ = 2.3 µm

shield 0 0 0
Φ [mΦ0]
45
-20 -20 -20 10
10 10
-10 0 10 -10 0 -10 0 10 -10 0 -10 0 10 -10 0
x [µm] y [µm] x [µ m] y [µm] x [µ m] y [µm] -45

Fig. 1. Scanning SQUID imaging fractional vortices and conventional and (E) conventional vortex appear in different cooling cycles in the same area
vortices in the same area of Ba1−xKxFe2As2. (A) Optical image of the sample at 11 K after local field (C) 6 mA, (D) −6 mA, and (E) 1 mA cooling to 10.5 K
with scan regions 1 to 4. (B) Pickup loop and field coil of the SQUID from 25 K and heating to 11 K. (D) The conventional vortex was moved

p
susceptometer are covered with superconducting shields, except for the loop to this location by applying the local field at 10.8 K in the same way as in Fig. 4D.
area to detect local magnetic flux. (C to E) SQUID measurements. Isolated (F to H) Simulations of (F) 0.3 F0, (G) −0.3 F0, and (H) F0 point source
(C) fractional vortex carrying ~0.3 of the flux quantum, (D) fractional antivortex, models show similarity to the data in (C) to (E), respectively.

g
A B C vortices (18). When we cooled to 3 K, well be-
region 2 35 Fractional vortex Conventional vortex low the temperature where we can resolve
region 2 region 2 T [K] fractional vortex objects, then heated back
100 30 11.0
Magnetic flux Φ [mΦ0 ]

300 above 9 K, we observed the reemergence of

25 T [K] the fractional vortex objects with similar mag-

y
80
Peak Φ [mΦ0 ]

10.8
Φ [mΦ0]

20 11.0 10.6 netic flux pinned at the same location (the

60 200 fractional vortex object in Figs. 1D and 2B is
Conventional 15 10.4
the reemergent one; the initially created vortex
40 vortex 10 10.0 10.0
100 is not shown), but sometimes cooling and sub-
5 sequent heating resulted in disappearance of
20 Fractional
Tc the object from the scanned area.
vortex 0 9.0 0 9.0
0 To determine whether the observed vortex-
9 10 11 -10 -5 0 5 10 -10 -5 0 5 10 like object was a true quantum vortex, the mo-
T [K] Cross section [µm] Cross section [µm] bility of the object needed to be demonstrated.
We checked this by monitoring the positions

y g
Fig. 2. Temperature dependence of the magnetic field of a fractional versus conventional vortex.
(A) Comparison of the temperature dependence of the maximum flux through the pickup loop with the
of these objects when we cooled or heated the
loop above the centers of the fractional vortex and the conventional vortex. The measurement indicates that
sample. We found that fractional vortex–like
the carried fraction of the flux quantum drops continuously as temperature decreases. (B and C) Measured
objects sometimes randomly moved to other
cross sections of (B) the fractional vortex and (C) the conventional vortex along the x axis, where the
positions (Fig. 4, A and B). The observed pro-
y position is indicated by horizontal arrows in Fig. 1, C and E. Solid lines are fitting results of point source
cess was akin to the mobility of conventional

,
models. Note that London model–based ansatz for fractional vortices can overestimate the field at the
vortices moving between pinning positions.
vortex center and underestimate the field at the vortex tail owing to nonlinear effects important for fractions
Next, we were able to manually change the
other than one-half (18).
position of the fractional vortex–like objects
by applying a local repulsive force created by
the scanning SQUID field coil (Fig. 4, C and D).
Alternatively, we obtained the fractionally objects are unconventionally quantized frac- We demonstrated that the scanning SQUID
quantized flux in the fractional vortices by tional vortices. manipulated both the fractional vortex–like
fitting a fractional vortex with fitting parame- object and the conventional vortex at the same
ters of the penetration depth and the fraction Mobility of fractional vortex time (Fig. 4D).
(fig. S9) or by integrating flux over a frac- The fraction of the flux quantum gradually In superconductors with very weak inter-
tional vortex without use of the penetration decreased with decreasing temperature, and layer coupling, magnetic probe manipulation
depth (fig. S10). We also verified that the fit- below T/Tc = 0.8, it could not be resolved can break a vortex into two small magnetic
ting with the F0 point source model does not from the background noise. This temperature excitations that are described as a broken
work for the observed fractional vortices (fig. dependence could be caused by the fraction stack of pancake vortices (12). To investigate
S8). The obtained fraction FF/F0 showed a of carried flux further dropping to a very small this scenario, we manipulated integer flux
similar temperature dependence in three value at low temperatures and/or by the effect vortices in a similar way but did not observe
tested regions (Fig. 3), suggesting that these of magnetic flux delocalization of fractional such behavior, which is consistent with the

Iguchi et al., Science 380, 1244–1247 (2023) 23 June 2023 2 of 4

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

Fig. 3. Temperature and spatial A different temperature dependencies. In cer-

dependence of magnetic flux in a 1.0 tain cases, multicomponent theories arise for
vortex. (A) The total magnetic flux fields that are associated with linear combi-
0.8 region 1, antivortex
FF measured in fractional vortices located nations of superconducting gaps in different
region 2, vortex

0
in three different regions is obtained bands (27), but the argument remains rather
region 2, antivortex

Fraction ΦF /
from fittings in Fig. 2B and fig. S4 and 0.6 similar.
shows similar temperature dependence region 3, vortex There are factors that can suppress, or in
in all three regions. (B) Temperature some cases prohibit, the formation of frac-
0.4 Fractional vortex
dependence of the penetration depth tional vortices in multiband superconduc-
obtained from the fittings in Fig. 2C and tors. In multicomponent superconductors, the
fig. S5 is used to fit the profiles of the 0.2 components are electrically charged. The un-
fractional vortices in the same region. avoidable electromagnetic intercomponent in-
Similar temperature-dependent penetra- 0 teraction tends to confine fractional vortices
tion depth is obtained by the scanning 0.8 0.85 0.9 0.95 1.0 into integer flux vortices in the bulk of a multi-
SQUID susceptibility measurements T/Tc band superconductor (26). This is the principal
on region 1 (fig. S1). The error bars in B difference compared with multicomponent
the susceptibility measurements 5 superfluids where vortices carry no magnetic
represent the uncertainty of the SQUID Susceptibility at region 1 flux and no such electromagnetic confine-
height, ~±0.1 mm. The Tc ~11.2 K is 4 region 1 ment of vortices exists. Similarly, interband
obtained from the scanning SQUID region 2 interactions, such as interband Josephson
[µm]

susceptometry (fig. S1). 3 region 3 coupling, tend to lock phases in different bands,

p
making one-flux-quanta composite objects
Conventional vortex the energetically cheapest type of vortex
2
(26). Hence, although fractional vortices are
theoretically possible in multiband mate-
1 rials, they are more energetically expensive
(26). Nonetheless, a minority fraction of frac-
0

g
tional vortices are theoretically allowed to
0.8 0.85 0.9 0.95 1.0
form in the bulk of a superconductor owing
T/Tc to fluctuations, quenches, or pinning, akin
to the effect of remanent vorticity in super-
much weaker anisotropy of the upper critical previous experiments demonstrated that fluids. However, fractional vortices have not

fields in our sample compared with strongly
layered systems (19). Moreover, some of the
fractional vortex–like objects appeared with-
out any other fractional vortex–like objects
within a 50-mm area (figs. S2 and S3). Finally,
the temperature dependence of the fractional
vortices flux is inconsistent with the pancake
vortex in layered systems (Fig. 2B). The ob-
served robustness and mobility established
that the observed objects represented quan-
Ba1−xKxFe2As2 spontaneously breaks the time-
reversal symmetry (23–25). Systems with mul-
tiple broken symmetries are described by
multicomponent order parameters. This sug-
gests that, in principle, there are necessary
degrees of freedom to support several types
of vortices associated with phase windings in
different components of the order parameter.
A multiband superconductor with bands la-
beled by a band index
j is
described by multi-
y
been previously observed in multiband super-
conductors such as MgB2 (28–30) and pnictides
(31–33) nor in multiband materials where
time-reversal symmetry breaking was reported
(34–37).
This raises the question of what is special
about Ba1−xKxFe2As2 at x = 0.77. First, we note
that experiments indicate that Ba1−xKxFe2As2
is a so-called s + is (or similar) superconductor
(23–25, 38). This is a multiband state where

tum vortices.
To ensure that the local variation of mag-
netic field penetration depth does not play a
large role, we created integer and fractional
vortices at the same positions during different
y g
ple complex fields yj ¼
yj
eiqj. In such a case, in the time-reversal symmetry is broken. The pre-
the simplest model, the current will be given by a cise microscopic model for superconducting
sum of contributions from different bands J ¼ state in this material is not known. However,
X

2
ðeℏ=i2mÞðyj ∇yj  yj ∇yj Þ  ð2e2=mcÞ
yj
A , the calculations in the framework of the mini-
j mal phenomenological Ginzburg-Landau model
where A is the vector potential, and m is the

cooling cycles. Figure S7 demonstrates that the
two types of vortices have similar localization
but very different magnetic field amplitude.
We also observed homogeneous susceptibil-
ity and its monotonic increase at low temper-
ature over the whole scan area (fig. S6), which
ruled out the local variation of magnetic field
penetration depth scenario. We numerically
estimated the penetration depth from the
susceptibility at region 1 (20), which had a sim-
ilar temperature dependence to that obtained
from fitting vortex field (Fig. 3B). We did not
observe any magnetic defects above Tc.
Discussion
The material Ba1−xKxFe2As2 has a multiband
electronic structure (21, 22). Furthermore,
mass parameter. It was discussed at the level
of effective fields theory that vortex excitations
are possible with phase winding in only one of
the bands (26); the above expression can then
be used to calculate the magnetic flux enclosed
in the resulting vortex. This can be done by
integrating the vector potential over a path s
located far away from vortex where J = 0.
That gives the magnetic fluxX of
a
vortex: F ¼
∮X s A·dl


¼ ∮s∇q1·dl ðℏc=eÞjy1j2= j
yj
2 ¼ F0jy1j2=


2
j yj . The flux is thus no longer a function
of just fundamental physical constants; in-
stead, it depends on microscopic details and
temperature.


This is because the ratios of the
fields
yj
associated with superconducting
components in different bands have, in general,
,
suggested that, while it is not a necessary
ingredient, under certain conditions, the broken
time-reversal symmetry in a multiband system
helps the formation of mobile fractional vorti-
ces in individual bands (39). Second, the dop-
ing x = 0.77, where we could create fractional
vortices on demand, corresponds to the spe-
cial point where the critical temperature of
the time-reversal symmetry breaking is max-
imal (24, 25). This maximum suggests that
the components are well developed, and hence
vortices carry a substantial fraction of mag-
netic flux quantum near the temperature of
the superconducting phase transition at x =
0.77. By contrast, using the same simple ap-
proach, we could not observe fractional vorti-
ces in the Ba1−xKxFe2As2 samples with different

Iguchi et al., Science 380, 1244–1247 (2023) 23 June 2023 3 of 4

RES EARCH | R E S E A R C H A R T I C L E
Fig. 4. Mobility of the fractional vortex.
(A and B) The fractional vortex (FV) moves
while (A) cooling and (B) heating. This motion
happens randomly. (C and D) Scanning
SQUID susceptometer manipulates (C) the
FV and (D) both the FV and the conventional
vortex (CV) by applying a repulsive local
field. The schematic image of the field coil
indicates the location where the local field is
applied. Note that the peak fluxes of FV
in (B) and (C) at 11 K are ~13 mF0, from
which a fraction of ~0.3 is obtained. Vortices
were created by local field [(A) to (C)]
6 mA and (D) uniform field 3.5 mG cooling
to 10.5 K from 25 K. The “After 3rd”
image has two scans combined, separated
by a vertical black line.
doping levels (table S1). However, we cannot
exclude that fractional vortices can be cre-
ated at other doping levels by using different
approaches. Also, we cannot exclude that at
different doping levels there are fractional
vortices that carry too small a fraction of flux
quantum to be detected or too large a fraction
of the flux quantum to be distinguished from
single-quantum vortices.
We demonstrated the existence of stable
vortices with phase winding only in one of the
bands in a numerical solution of the three-band
Bogoliubov–de Gennes model. The solutions are
given in (40), using techniques presented in (41).
Although our solutions provide a fully micro-
scopic demonstration of the principle of stable
fractional vortices, further theoretical studies are
needed to connect the observation of fractional
vortices with concrete microscopic models of
Ba1−xKxFe2As2. Of special interest is the tem-
perature dependence of the enclosed fraction
of flux quantum, which contains information
that allows one to constrain microscopic mod-
els of pairing.
Our observations demonstrate that vortices
carrying a temperature-dependent fraction of
the flux quantum are possible in supercon-
ductors. Dynamics, control and manipulation,
the precise configuration of their magnetic
field, and the mechanisms of pinning of these
objects are promising directions that will give
insights into both the basic questions of super-
conductivity and the nature of superconduc-
tivity in Ba1−xKxFe2As2. They may also enable
the use of these objects in fluxonics-based cryo-
genic computing (42, 43).
Iguchi et al., Science 380, 1244–1247 (2023) 23 June 2023
RE FERENCES AND NOTES
1. L. Onsager, Nuovo Cim. 6 (suppl. 2), 279–287 (1949).
2. F. London, Phys. Rev. 74, 562–573 (1948).
3. F. London, Superfluids, Volume I: Macroscopic Theory of
Superconductivity (Dover Publications, ed. 2, 1960)
4. L. Onsager, Phys. Rev. Lett. 7, 50 (1961).
5. A. A. Abrikosov, Sov. Phys. JETP 5, 1174 (1957).
6. A. M. Chang et al., Appl. Phys. Lett. 61, 1974–1976 (1992).
7. P. L. Gammel et al., Phys. Rev. Lett. 59, 2592–2595 (1987).
8. H. F. Hess, R. B. Robinson, R. C. Dynes, J. M. Valles Jr.,
J. V. Waszczak, Phys. Rev. Lett. 62, 214–216 (1989).
9. J. R. Kirtley et al., Phys. Rev. Lett. 76, 1336–1339 (1996).
10. J. Jang et al., Science 331, 186–188 (2011).
11. Y. Li, X. Xu, M.-H. Lee, M.-W. Chu, C. L. Chien, Science 366,
238–241 (2019).
12. L. Luan et al., Phys. Rev. B 79, 214530 (2009).
13. H. Bluhm, N. C. Koshnick, M. E. Huber, K. A. Moler, Phys. Rev.
Lett. 97, 237002 (2006).
14. K. Kihou et al., J. Phys. Soc. Jpn. 85, 034718 (2016).
15. J. R. Kirtley et al., Rev. Sci. Instrum. 87, 093702 (2016).
16. J. R. Kirtley et al., Supercond. Sci. Technol. 29, 124001 (2016).
17. J. R. Kirtley, C. C. Tsuei, K. A. Moler, Science 285, 1373–1375
(1999).
18. E. Babaev, J. Jäykkä, M. Speight, Phys. Rev. Lett. 103, 237002 (2009).
19. Y. Liu et al., Phys. Rev. B 89, 134504 (2014).
20. J. R. Kirtley et al., Phys. Rev. B 85, 224518 (2012).
21. S. Maiti, A. V. Chubukov, Phys. Rev. B 87, 144511 (2013).
22. J. Böker, P. A. Volkov, K. B. Efetov, I. Eremin, Phys. Rev. B 96,
014517 (2017).
23. V. Grinenko et al., Phys. Rev. B 95, 214511 (2017).
24. V. Grinenko et al., Nat. Phys. 16, 789–794 (2020).
25. V. Grinenko et al., Nat. Phys. 17, 1254–1259 (2021).
26. E. Babaev, Phys. Rev. Lett. 89, 067001 (2002).
27. J. Garaud, M. Silaev, E. Babaev, Physica C 533, 63–73 (2017).
28. M. R. Eskildsen et al., Phys. Rev. Lett. 89, 187003 (2002).
29. T. Nishio et al., Phys. Rev. B 81, 020506(R) (2010).
30. J. Gutierrez et al., Phys. Rev. B 85, 094511 (2012).
31. L. J. Li, T. Nishio, Z. A. Xu, V. V. Moshchalkov, Phys. Rev. B 83,
224522 (2011).
32. B. Kalisky et al., Phys. Rev. B 83, 064511 (2011).
33. I. P. Zhang et al., Phys. Rev. B 100, 024514 (2019).
34. P. G. Björnsson, Y. Maeno, M. E. Huber, K. A. Moler, Phys. Rev. B
72, 012504 (2005).
35. J. R. Kirtley et al., Phys. Rev. B 76, 014526 (2007).
36. C. W. Hicks et al., Phys. Rev. B 81, 214501 (2010).
37. Y. Iguchi et al., Phys. Rev. B 103, L220503 (2021).
38. V. L. Vadimov, M. A. Silaev, Phys. Rev. B 98, 104504 (2018).
39. J. Garaud, E. Babaev, Phys. Rev. Lett. 112, 017003 (2014).
p
g
40. See supplementary materials.
41. A. Benfenati, M. Barkman, E. Babaev, Phys. Rev. B 107, 094503
(2023).
42. K. Miyahara, M. Mukaida, K. Hohkawa, Appl. Phys. Lett. 47,
y
754–756 (1985).
43. T. Golod, A. Iovan, V. M. Krasnov, Nat. Commun. 6, 8628 (2015).
44. Y. Iguchi et al., Data for “Observation of superconducting
vortices carrying a temperature-dependent fraction of the flux
quantum,” version 1, Zenodo (2023); https://doi.org/10.5281/
zenodo.7644830.
AC KNOWLED GME NTS
The authors thank N. Nandi and J. Garaud for fruitful discussions.
Funding: This work was primarily supported by the Department of
Energy, Office of Science, Basic Energy Sciences, Materials Sciences
and Engineering Division, under contract no. DE- AC02-76SF00515.
E.B. was supported by Swedish Research Council grants 2016-06122
y g
and 2022-04763 and by the Wallenberg Initiative Materials Science
for Sustainability (WISE) funded by the Knut and Alice Wallenberg
Foundation. Author contributions: Y.I. carried out the scanning
SQUID microscopy, analyzed data, and wrote the manuscript. R.A.S.
carried out the scanning SQUID microscopy. K.K. and C.-H.L.
synthesized the crystals. M.B. and A.L.B. wrote BdG code and carried out
and analyzed the BdG calculations. V.G. supervised the project and
selected and characterized the samples. E.B. conceptualized, contributed
,
to writing, and supervised the project. K.A.M. supervised the project. All
the authors discussed the results and implications and commented
on the manuscript. Competing interests: The authors declare no
competing financial interests. Data and materials availability: All data
needed to evaluate the conclusions in the paper are present in the
paper and/or the supplementary materials and available in Zenodo
(44). License information: Copyright © 2023 the authors, some
rights reserved; exclusive licensee American Association for the
Advancement of Science. No claim to original US government works.
https://www.science.org/about/science-licenses-journal-article-reuse
SUPPLEMENTARY MATERIALS
science.org/doi/10.1126/science.abp9979
Materials and Methods
Figs. S1 to S11
Table S1
References
Submitted 10 March 2022; resubmitted 26 August 2022
Accepted 15 May 2023
Published online 1 June 2023
10.1126/science.abp9979
4 of 4
RES EARCH

POLYMERS degree of polymerization of the primary chains

is controlled by the chain-transfer agent (CTA).
Facile mechanochemical cycloreversion of polymer The primary chain in this system is defined
as the polyacrylate backbone formed through
cross-linkers enhances tear resistance radical polymerization; for example, polymeriz-
ing a pregel solution without cross-linker (C)
Shu Wang1,2†, Yixin Hu1,2, Tatiana B. Kouznetsova1,2, Liel Sapir1,3, Danyang Chen1,3, directly yields primary chains (linear polyacrylate)
Abraham Herzog-Arbeitman1,4, Jeremiah A. Johnson1,4*, as a melt with no notable mechanical integ-
Michael Rubinstein1,2,3,5,6*, Stephen L. Craig1,2* rity. Polymerizing the pregel solution with bis-
acrylate cross-linkers C1 and C2 separately,
The mechanical properties of covalent polymer networks often arise from the permanent end-linking however, forms two percolated elastomer net-
or cross-linking of polymer strands, and molecular linkers that break more easily would likely produce works E1 and E2 because the primary chains
materials that require less energy to tear. We report that cyclobutane-based mechanophore cross-linkers in E1 and E2 are held together by cross-
that break through force-triggered cycloreversion lead to networks that are up to nine times as tough linkers C1 and C2. Cross-linker C1 is a cis-diaryl
as conventional analogs. The response is attributed to a combination of long, strong primary polymer substituted cyclobutane-based mechanophore
strands and cross-linker scission forces that are approximately fivefold smaller than control cross-linkers that reacts under tension by means of a force-
at the same timescales. The enhanced toughness comes without the hysteresis associated with coupled [2 + 2] cycloreversion to form two
noncovalent cross-linking, and it is observed in two different acrylate elastomers, in fatigue as well cinnamates (Fig. 1D) (8), whereas cross-linker
as constant displacement rate tension, and in a gel as well as elastomers. C2 consists of common hydrocarbon and
carbon-oxygen bonds, which are mechano-
chemically strong (9, 10). The force-coupled

he lifetime and usefulness of covalent
rubbery polymer networks are deter-
mined by their ability to stretch repeatedly
without breaking (1, 2). At sufficiently high
strains, the network breaks by forming a
crack that subsequently propagates through
mechanophore holds the network together
through side-chain cross-linking in randomly
cross-linked networks (Fig. 1B). We find that the
material properties again depend strongly on
the force-coupled reactivity of the mechanophore.
Side-chain cross-linked networks with
p
kinetics of C1 cycloreversion have been char-
acterized by single-molecule force spectros-
copy (fig. S1). The lifetime-force relationship
of C1 and common C-C bond [simulation
data by Beyer (10)] are shown in Fig. 1E. The
force required to achieve lifetimes relevant to

the material to the point of failure. Tearing at
the macroscopic level is resisted at the mol-
ecular level by polymer chains within the
network that need to break for the crack to
propagate (3). The scission of covalent chains
cyclobutane mechanophores
Our approach is shown in Fig. 1C. Side-chain
cross-linked networks with long polyacrylate
backbones (primary chain, light gray chains
g
material tearing (between microseconds and
seconds) is roughly a factor of five lower for
C1 than what is expected for C2 and the other
molecular components of the networks.
We start with a stoichiometry of [M]:[C]:

y
occurs through a chemical reaction (4)—typically in Fig. 1, B and C) are formed through rev- [CTA]:[PI] = 1:1/50:1/1200:1/2000. The reac-
homolytic bond scission (5) that is accelerated ersible addition–fragmentation chain-transfer tivity of alkyl acrylates is known to be largely
by the tension in overstretched chains at the (RAFT) polymerization of 2-methoxyethyl acrylate independent of the character of the alkyl group
propagating crack front. It is possible, however, monomer (M). The polymerization is initiated (11); thus, polymerizing the same pregel solu-
to design and incorporate a small fraction of by the photoinitiator (PI), and the average tion with C1 and C2 is expected to lead to
mechanically scissile functional groups (mechano-
phores) whose reactivity dominates the chain
scission events. In such systems, the mech-
anical properties of the network might be
expected to reflect the force-coupled reactivity

y g
of the mechanophores (6). When mechano-
phores are embedded into the middle of each
elastically active network strand (Fig. 1A), the
molecular and material properties are correlated
in an intuitively satisfying manner: Mechano-

,
chemical reactions that require less force to
break on a given timescale result in weaker
polymer networks (6, 7). We were curious to
see whether this relationship persists when the

1
NSF Center for the Chemistry of Molecularly Optimized
Networks, Duke University, Durham, NC, USA. 2Department
of Chemistry, Duke University, Durham, NC, USA. 3Thomas
Lord Department of Mechanical Engineering and Materials
Science, Duke University, Durham, NC, USA. 4Department of
Chemistry, Massachusetts Institute of Technology (MIT),
Boston, MA, USA. 5Departments of Biomedical Engineering
and Physics, Duke University, Durham, NC, USA. 6Institute
for Chemical Reaction Design and Discovery, Hokkaido
University, Sapporo 001-0021, Japan. Fig. 1. Networks design and the mechanochemical reactivities of cross-linkers. (A and B) Mechanophores
*Corresponding author. Email: jaj2109@mit.edu (J.A.J.); embedded into the middle of chains (A) and as the side-chain cross-linkers of primary chains (B). (C) General
michael.rubinstein@duke.edu (M.R.); stephen.craig@duke.edu procedure for RAFT network preparation. CN, cyano group; OH, hydroxyl group. (D) Force-triggered cycloreversion
(S.L.C.)
†Present address: Department of Mechanical Engineering, MIT, of C1. (E) Lifetime as a function of breaking force for C1 and for a common C-C bond. Dashed lines are shown
Boston, MA, USA. as guides.

Wang et al., Science 380, 1248–1252 (2023) 23 June 2023 1 of 5

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

Fig. 2. Mechanical characterizations for elastomers. (A to D) Frequency sweeps shear geometry (E) and tearing energies (F) for NW-1 and NW-2 adapted from previous

(×3 samples each) (strain = 0.5%) (A), uniaxial extension stress-strain curves of
unnotched (×3 samples each) (B) and notched (×5 samples each) (C) samples in
pure shear geometry, and tearing energies (D) for E1 and E2; P = 6.5 × 10−6, t test
assuming equal variance. (E and F) Uniaxial tensile test of notched samples in pure
p
work (6). NW, network prepared by end-linking tetra-arm polyethylene glycol with
bi-functional linkers. (G and H) Crack growth per cycle as a function of energy release
rate (×3 samples each) for E1 (G) and E2 (H). Shaded areas indicate the range of
corresponding tearing energies shown in (D). Error bars are standard deviations.

effectively identical networks that differ only
in the mechanically coupled reactivity of the
cross-linker. This expectation was verified
through several characterizations: (i) Shear
moduli of E1 and E2 (blue and red in Fig. 2A,
films in a pure shear geometry (1). The stress-
strain curves across five notched samples for
each elastomer in Fig. 2C show that E1 and E2
have very different critical strains for crack
propagation, and Г of E1 (113.0 ± 9.7 J m−2) is
g
are prepared as organogels (Г = 62.3 ± 6.4
versus 18.5 ± 2.3 J m−2) (fig. S8) and when
elastomers are made from ethyl acrylate
monomers (Г = 540 ± 84 versus 172 ± 25 J m−2)
(fig. S9). See the supplementary materials for

respectively) were characterized by small-
amplitude oscillatory rheology, and both
exhibit similar storage moduli G′ and loss
moduli G″ across frequencies of 0.1 to 100 Hz.
Their storage moduli G′ are independent of
frequency from 0.1 to 10 Hz and are well above
G″, which indicates that E1 and E2 are well-
formed elastic networks. The average G′ of
E1 and E2 at 0.1 Hz are statistically the same
(P = 0.83, t test assuming equal variance):
roughly an order of magnitude higher than Г
of E2 (11.5 ± 3.7 J m−2) (Fig. 2D). The impact
of cyclobutane reactivity on toughness is almost
the direct opposite of that observed in an end-
linked system (Fig. 1A), in which embedding
similar mechanophores into the middle of
chains leads to eightfold lower tearing energy
compared with the nonmechanophore control
(Fig. 2, E and F) (6). Here, stitching polymer
chains together through the much mechani-
y
details.
The results raise the important question of
why the effect of introducing the same scission
reaction in side-chain cross-linked networks is
almost the exact opposite of what is observed
in the end-linked networks of Fig. 1A. This
toughening effect is not due to simple energy
dissipation in the bulk because both G′ and G″
are unchanged by the cross-linkers. We hy-
pothesized that the different response to the

336 ± 35 and 330 ± 14 kPa, respectively, and
well above the modulus due to entanglement
Ge ≈ 111 ± 6 kPa (fig. S3); (ii) Young’s moduli
obtained from uniaxial stress-strain curves
(Fig. 2B) are indistinguishable; (iii) similar sol
cally weaker reactant leads to a much tougher
material.
The enhanced toughness brought about by
replacing C2 with C1 is not limited to net-
works made from a 2-methoxyethyl acrylate
y g
embedded reactivity originates from an import-
ant topological difference. In the end-linked
networks of Fig. 1A (6), every load-bearing
chain has a mechanophore, so increasing the
reactivity of the mechanophore increases every

fractions (fig. S4A); and (iv) similar equilibrium
swelling ratios (fig. S4B). These data suggest
that E1 and E2 are statistically identical in
terms of network connectivity. Furthermore,
the glass transition temperatures Tg of E1 and
E2 are similar and well below room temper-
ature (~−30°C) (fig. S5).
Although these two elastomers have similar
network connectivity, we noticed that their
unnotched films break very differently when
stretched by dynamic mechanical analysis (Fig.
2B). The network E1 made with weaker cross-
linkers is noticeably more difficult to tear than
E2 made with stronger cross-linkers. To quan-
titatively confirm their difference, the tearing
energies Г of E1 and E2 were characterized by
using the Rivlin-Thomas method on notched
comonomer, to the elastomer state, or to the
specific characterization test, although the
magnitude of the effect does change. For exam-
ple, the same trend observed in tearing is also
observed in fatigue testing, which minimizes
contributions to the tearing resistance from
the energy that is dissipated rather than stored
elastically during network stretching (hysteresis)
by slowing down the crack growth rate with
cyclic loading (12, 13). As shown in Fig. 2,
G and H, both E1 and E2 have fatigue thresholds
below their tearing energies (shaded areas),
which is common for elastomers (3, 12). How-
ever, the fatigue threshold of E1 (52.0 J m−2) is
still much larger than that of E2 (8.0 J m−2).
Moreover, a similar toughening effect also exists
when the 2-methoxyethyl acrylate networks
,
scission probability to the same extent. Incor-
porating the mechanophore as the side-chain
cross-linker, however, only weakens the cross-
linker and not the primary chain.
Toughening mechanism of E1
The consequences of this topology are pro-
posed in Fig. 3A. Preferential scission through
the cycloreversion of intra/intermolecular
C1, rather than primary chain scission, length-
ens rather than removes the bridging chain
at the crack interface, which gives rise to the
increase in tearing energy. There is no such
preferential cross-linker reactivity in E2 be-
cause the cross-linkers are similarly strong
as the primary chain (Fig. 3B). A molecular
dynamics simulation of network fracture under

Wang et al., Science 380, 1248–1252 (2023) 23 June 2023 2 of 5

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

p
Fig. 3. Toughening mechanism of mechanochemically weak cross-linkers. moduli (×3 samples each) (C) and tearing energies (×5 samples each) (D)
(A) Mechanochemically weak cross-linkers break preferentially before mechan- of E1 and E2 with different Np. (E) Storage moduli of E1 and E2 with different
ically strong primary chains break in E1, which increases the tortuosity of the cross-linking densities. (F) Log-log plot of G versus G′ of E1 and E2 with different

g
crack path. (B) Cross-linkers that are similarly strong as the primary chains in E2 cross-linking densities. Dashed lines are power-law fittings. Error bars are
do not break preferentially. Cartoons are schematic only. (C and D) Storage standard deviations.

uniaxial stretching provides a consistent pic- dissociation to be studied in the absence of elastomer E1 is only slightly tougher than E2,

ture: (i) At high strain, mechanochemically
weak cross-linkers in E1 break exclusively (up
to a strain of 6) while leaving the primary
chain bonds intact (fig. S18), and (ii) relevant
elastically active strands between cross-linkers,
on average, become much longer per bond
scission event in E1 compared with that in
E2 (fig. S22).
Random network cross-linking comprises a
complex mixture of intramolecular and inter-
other bond exchange processes.
On the basis of our picture in Fig. 3A, the
contour length of the primary chain would
influence this toughening effect, and so we
used the synthetic control afforded by RAFT
to vary the degree of polymerization of the
primary chains (Np) (table S2) while keeping
the extent of cross-linking constant. The Np
of different networks are verified experimen-
tally by comparing primary chains synthesized
y
but as the primary chains lengthen, the tear-
ing energies of E1 increase more than four-
fold for Np ≈ 2000. The unusual dependence
of Г on Np in E1 is consistent with our model;
the longer primary chains require a more tor-
tuous path.
The reactivity-enabled toughening afforded
by C1 provides a mechanism to mitigate an
otherwise intrinsic tradeoff in polymer net-
work optimization, namely, the inverse corre-

molecular junctions and loops (14). The pro-
grammed cycloreversion of intramolecular C1
releases hidden length in a manner that is
similar to noncovalent domain unfolding (15, 16)
and covalent reactive strand extension (17),
whereas that of intermolecular C1 deviates
without cross-linker to the primary chains
obtained by degrading the networks that were
prepared with chemically degradable cross-
linker C4 under the same conditions (see
fig. S12 and detailed methods in the supple-
y g
lation of modulus and toughness. At fixed Np ≈
1400, increasing the cross-linker content ([C]:
[M] = 1:200, 1:100, 1:50) leads to high modulus
materials. As expected, this increase in modu-
lus is accompanied by a significant drop of Г in
the control elastomer E2, with a scaling exponent

the crack and increases pathway tortuosity.
The effect of C1 is reminiscent of the toughen-
ing effects observed in most dynamic side-
chain cross-linked network systems (18–21).
The dynamic nature of the cross-linking
makes networks formed by such “sacrificial
bonds” prone to greater hysteresis and viscous
energy dissipation during use. Although the
magnitude of the overall toughening observed
from the use of C1 is smaller than that has
been reported from the addition of weak,
dynamic cross-linking (18, 21), it comes
with negligible hysteresis (fig. S15) or con-
tribution to loss modulus (Fig. 2A). It also
provides a model system of fixed covalent
structure that allows force-coupled cross-linker
mentary materials).
Elastomers of differing Np were prepared
with either C1 or C2, and the moduli and
swelling are again indistinguishable between
mechanophore and control networks for all
Np (Fig. 3C and fig. S4). The moduli increase
modestly with Np, as expected because of the
decrease in elastically inactive dangling chain
ends (22). The tearing energies of E2 do not
change much with Np (Fig. 3D), as is typical of
conventionally cross-linked elastomers, whose
tearing energy is dominated by the length of
the chain between cross-links (3), which is kept
almost constant here. The tearing energies
of E1, however, depend substantially on the
contour length of the primary chain (Fig. 3D).
When the primary chains are short (Np ≈ 350),
,
of −1.7 that is similar to that reported in other
side-chain cross-linked systems (Fig. 3F) (23, 24).
The loss of toughness is attributed to the
decreased chain length between cross-linkers
(3). By comparison, the reactivity engineered
into E1 enables the elastomer to acquire higher
stiffness without sacrificing the same extent of
tear resistance (scaling exponent of −0.7) because
the loss in toughness due to higher cross-link
density is offset by the enhanced toughness
afforded by additional mechanophores.
The decrease of Г for E1 with higher cross-
linking (Fig. 3F) suggests that the primary chains
cannot be completely pulled out before they
break, because if tearing energy is dominated
by complete chain pull-out (breaking all the

Wang et al., Science 380, 1248–1252 (2023) 23 June 2023 3 of 5

RES EARCH | R E S E A R C H A R T I C L E
C1 at least along one direction), the tearing
energy would increase while the modulus in-
creases because there are more cross-linkers to
break to dissipate energy. This incomplete pri-
mary chain pull-out might originate from en-
tanglement lockup under higher tension (25).
The locked entanglements act like strong cross-
linker C2, which force the primary chain to
break instead of being pulled out. The length
scale between two locked entanglements seems
to be cross-linking density dependent. This hy-
pothesis of incomplete chain pull-out suggests
that the tearing energy of E1 would eventually
saturate if Np keeps increasing while keeping
cross-linking density constant because the ef-
fective contour length between locked entan-
glements would limit the toughening effect (25).
The leveling off of tearing energy for E1 ob-
served in Fig. 3D also supports this hypothesis.
Effect of cross-linker mechanochemical
reactivities on Г
We wondered how the reactivity of the cross-
linker relative to the primary chain influences
the effect. We synthesized another mechano-
phore cross-linker C3 of intermediate strength
(fig. S10; relevant force at break is roughly half
that of conventional polymer components
and just over twice that of C1). E3 results in
only a very modest toughening effect (factor
of <2 versus >9 for otherwise identical net-
works; fig. S10), which is likely because the
larger mechanochemical strength of C3 com-
pared to C1 suppresses the lengthening effi-
ciency of bridging chains before primary chains
break. This result suggests that further decrease
in the mechanochemical strength of the cross-
linker while other components remain un-
changed might potentially lead to tougher
networks, yet such a cross-linker could suffer
from poor thermal stability, which risks the
physical integrity of the networks. The extreme
case is that the cross-linker has no strength (i.e.,
no cross-linker) or very low cross-linking den-
sity. Although, depending on the loading rate,
such highly entangled systems could achieve
high tearing energy through nonspecific near-
crack dissipation, their stiffness is no longer
tunable with cross-linkers (23, 24).
In addition, we tested a cross-linker that re-
produces the mechanochemical reactivity of
C1 but leads to molecular extension rather
than scission. We synthesized cross-linker C5
(synthetic details provided in the supplemen-
tary materials), in which initial cyclobutane
scission analogous to that found in C1 leads
to the release of local stored length, but com-
plete scission of the cross-linker requires sub-
sequent homolytic bond scission in a manner
analogous to C2. The cross-linker C5 was in-
corporated into elastomer E5 by following the
same procedure as used to make E1 and E2.
Consistent with the models proposed above,
E5 displays at best a very modest toughening
Wang et al., Science 380, 1248–1252 (2023) 23 June 2023
effect compared to E2 (fig. S11; the difference
in tearing energies is within the uncertainty of
the measurement error). This result further
supports the physical picture proposed in Fig.
3A because C5 is mechanochemically strong
after cycloreversion, which acts effectively as a
strong cross-linker that is reminiscent of C2.
Conclusions
When considered in combination with previ-
ous reports of quantitative relationships be-
tween mechanophore activity and material
properties (6, 17), the current work accentuates
how the position (along the polymer chain
versus at the cross-links), type (scissile versus
nonscissile), and mechanochemical reactivity
of mechanophores work in combination to
dictate the mechanical properties of the net-
works in which they are incorporated. When
mechanophores are incorporated along every
stress-bearing polymer strand within a network,
there is a direct correlation between mechano-
chemical reactivity and network toughness—
the easier it is to break the mechanophore, the
easier it is to tear the network (6). Replacing
strands of scissile mechanophores with anal-
ogous strands of nonscissile mechanophores,
however, leads to a toughening effect (17) be-
cause the latter release “hidden length” rather
than breaking, and the enhanced properties
at the single-strand level are translated to the
macroscopic materials. When mechanophores
are incorporated as otherwise conventional side-
chain cross-linkers, however, the responses are
opposite. Mechanochemically more reactive
(weaker) mechanophores lead to tougher rather
than more brittle networks (Fig. 2D) and scis-
sile mechanophores outperform their non-
scissile analogs (fig. S11) because the former
can share the load along the primary chains
through cycloreversion, whereas the latter can-
not because they are as strong as the primary
chain after activation.
From our observations, the macroscopic
properties of polymer networks can be di-
rectly correlated with molecular reactivity and
mechanisms. Fracture in cross-linked networks
is productively viewed as a sort of “molecular
composite,” and that picture has implications
for the use of “weak” bonds to toughen covalent
polymer networks. Typically, toughening a net-
work through weak cross-linkers involves the
use of dynamic interactions, such as ionic bond-
ing (18), hydrogen bonding (19), the reversible
formation of stable radicals (20, 26), or dynamic
covalent bonding (21). Without introducing
nondynamic covalent cross-linkers, all of these
systems would eventually lead to a fluid net-
work structure and added, potentially undesir-
able viscous dissipation on longer timescales or
at higher temperatures. The capacity to recom-
bine and/or exchange during or after the scis-
sion of dynamic bonds also complicates their
toughening mechanism even in hybrid net-
works that use partial covalent cross-linking to
provide long-term shape persistence (18, 27, 28).
No such scrambling or reformation is accessible
in E1 because C1 decouples its labile mechano-
chemical reactivity from thermal reversibility.
These results demonstrate that preferential
bond scission at the propagating crack front
alone is sufficient to provide a substantive
toughening effect. We note that pathway tor-
tuosity at the macroscale, for example, in
composites, is an established toughening mech-
anism (29), and the results observed here in-
dicate that molecular analogs of that behavior
likely contribute to the more complex toughen-
ing mechanisms at play in reversible networks.
For both those dynamic networks and the static
covalent networks demonstrated here, the pri-
mary chain length effect offers a clear design
principle for optimizing reactivity-enabled tough-
ening without losing stiffness.
We acknowledge that greater toughening
p
effects can be obtained through other mecha-
nisms, such as dynamic bonding and entan-
glements. The advantage of toughening through
the programmed reactivity of covalent cross-
linkers is that the properties of the mechano-
phore network are indistinguishable from those
g
of the conventional network, with the excep-
tion of the preferential scission behavior that
occurs only when and where necessary to in-
hibit material tearing and fatigue; even the
primary chain effect has a very modest impact
y
on low-strain mechanical properties relative
to the impact on toughness. The usefulness of
C1 shows that these gains can be realized with
mechanophores of good thermal stability (8)
(fig. S16; cyclobutane structure is stable at 150°C
for 90 hours), which offers the opportunity to
further optimize toughness through the judi-
cious design of the scissile reaction mechanism.
y g
REFERENCES AND NOTES
1. R. S. Rivlin, A. G. Thomas, J. Polym. Sci. 10, 291–318 (1953).
2. C. Creton, M. Ciccotti, Rep. Prog. Phys. 79, 046601 (2016).
3. G. J. Lake, A. G. Thomas, Proc. R. Soc. London A 300, 108–119
(1967).
4. S. Wang, S. Panyukov, M. Rubinstein, S. L. Craig,
Macromolecules 52, 2772–2777 (2019).
,
5. T. Matsuda, R. Kawakami, R. Namba, T. Nakajima, J. P. Gong,
Science 363, 504–508 (2019).
6. S. Wang et al., J. Am. Chem. Soc. 143, 3714–3718 (2021).
7. B. H. Bowser et al., J. Am. Chem. Soc. 143, 5269–5276 (2021).
8. H. Zhang et al., Nat. Commun. 8, 1147 (2017).
9. M. Grandbois, M. Beyer, M. Rief, H. Clausen-Schaumann,
H. E. Gaub, Science 283, 1727–1730 (1999).
10. M. K. Beyer, J. Chem. Phys. 112, 7307–7312 (2000).
11. A. P. Haehnel et al., Macromolecules 47, 3483–3496
(2014).
12. G. J. Lake, P. B. Lindley, J. Appl. Polym. Sci. 9, 1233–1251
(1965).
13. J. Tang, J. Li, J. J. Vlassak, Z. Suo, Extreme Mech. Lett. 10,
24–31 (2017).
14. J. Wang et al., Chem. Sci. 10, 5332–5337 (2019).
15. J. Chung, A. M. Kushner, A. C. Weisman, Z. Guan, Nat. Mater.
13, 1055–1062 (2014).
16. J. Fang et al., Nat. Commun. 4, 2974 (2013).
17. Z. Wang et al., Science 374, 193–196 (2021).
18. J. Y. Sun et al., Nature 489, 133–136 (2012).
19. J. A. Neal, D. Mozhdehi, Z. Guan, J. Am. Chem. Soc. 137,
4846–4850 (2015).
4 of 5
RES EARCH | R E S E A R C H A R T I C L E
20. H. Sakai et al., ACS Macro Lett. 9, 1108–1113 (2020).
21. M. Du et al., Nat. Commun. 13, 3231 (2022).
22. R. Henkel, P. Vana, Macromol. Chem. Phys. 215, 182–189
(2014).
23. J. Kim, G. Zhang, M. Shi, Z. Suo, Science 374, 212–216
(2021).
24. D. Zheng, S. Lin, J. Ni, X. Zhao, Extreme Mech. Lett. 51, 101608
(2022).
25. Y. Zheng, M. Tsige, S. Q. Wang, Macromol. Rapid Commun. 44,
2200159 (2023).
26. T. Watabe, D. Aoki, H. Otsuka, Macromolecules 55, 5795–5802
(2022).
27. G. Miquelard-Garnier, D. Hourdet, C. Creton, Polymer 50,
481–490 (2009).
28. R. Long, K. Mayumi, C. Creton, T. Narita, C. Y. Hui,
Macromolecules 47, 7243–7250 (2014).
29. R. O. Ritchie, Int. J. Fract. 100, 55–83 (1999).
Wang et al., Science 380, 1248–1252 (2023) 23 June 2023
30. S. Wang et al., Data from: Facile Mechanochemical Cycloreversion of All data are available in the main text or the supplementary materials.
Polymer Cross-linkers Enhances Tear Resistance, Duke Research All data are uploaded to the Duke Research Data Repository (30).
Data Repository (2023); https://doi.org/10.7924/r43r1215n. License information: Copyright © 2023 the authors, some rights
reserved; exclusive licensee American Association for the
ACKN OWLED GMEN TS Advancement of Science. No claim to original US government works.
We thank B. D. Olsen for helpful discussion and the K. A. Gall https://www.science.org/about/science-licenses-journal-article-reuse
laboratory for access to the fatigue test instrument. Funding: This SUPPLEMENTARY MATERIALS
work was supported by the NSF Center for the Chemistry of Molecularly
science.org/doi/10.1126/science.adg3229
Optimized Networks (MONET) (CHE-2116298 to J.A.J., M.R., and S.L.C.).
Materials and Methods
Author contributions: Conceptualization: S.W., A.H.-A., J.A.J., M.R.,
Supplementary Text
and S.L.C. Methodology: S.W., Y.H., T.B.K., L.S., D.C., A.H.-A., J.A.J., M.R.,
Figs. S1 to S52
and S.L.C. Investigation: S.W., Y.H., T.B.K., L.S., and D.C. Funding
Tables S1 and S2
acquisition: J.A.J., M.R., and S.L.C. Project administration: J.A.J., M.R.,
References (31–51)
and S.L.C. Supervision: J.A.J., M.R., and S.L.C. Writing – original draft:
Movie S1
S.W. and S.L.C. Writing – review & editing: S.W., Y.H., T.B.K., L.S.,
D.C., A.H.-A., J.A.J., M.R., and S.L.C. Competing interests: The authors Submitted 15 December 2022; accepted 16 May 2023
declare no competing interests. Data and materials availability: 10.1126/science.adg3229
p
g
y
y g
,
5 of 5
RES EARCH

BIOMATERIALS formance of the hinge under much larger loads,

we further increased the load stepwise to about
Deformable hard tissue with high fatigue resistance 100% overload in the cyclic tests (fig. S6). Its
function is still largely preserved, although the
in the hinge of bivalve Cristaria plicata resilience of the hinge decreases under such
conditions.
Xiang-Sen Meng1†, Li-Chuan Zhou2,3†, Lei Liu1†, Yin-Bo Zhu2, Yu-Feng Meng1, Dong-Chang Zheng2, Reconstructed x-ray microcomputed tomo-
Bo Yang1, Qi-Zhi Rao4, Li-Bo Mao1*, Heng-An Wu2*, Shu-Hong Yu1,5* graphic (mCT) images reveal the inhomoge-
neous electron density distribution in the hinge,
The hinge of bivalve shells can sustain hundreds of thousands of repeating opening-and-closing valve which is surrounded by a porous tissue and two
motions throughout their lifetime. We studied the hierarchical design of the mineralized tissue in the slab areas (Fig. 1B, ii, and fig. S1F). Given the
hinge of the bivalve Cristaria plicata, which endows the tissue with deformability and fatigue resistance differences in both optical and mCT images
and consequently underlies the repeating motion capability. This folding fan–shaped tissue consists of (Fig. 1B), the hinge area can be divided into two
radially aligned, brittle aragonite nanowires embedded in a resilient matrix and can translate external distinct regions: a folding fan–shaped region
radial loads to circumferential deformation. The hard-soft complex microstructure can suppress stress (FFR) and an outer ligament (OL). Elemental
concentration within the tissue. Coherent nanotwin boundaries along the longitudinal direction of the mapping images reveal a sharp increase of cal-
nanowires increase their resistance to bending fracture. The unusual biomineral, which exploits the cium concentration from the OL to the FFR
inherent properties of each component through multiscale structural design, provides insights into (Fig. 1D), which is consistent with the elastic
the evolution of antifatigue structural materials. modulus and hardness variations of the re-
gions in the dried hinge and the density dif-
ference in the mCT image (Fig. 1E and fig. S7)

rittle materials are extensively used as
structural or functional components in
various fields such as aerospace, tissue
engineering, and electronics (1–3). How-
ever, artificial brittle materials are sen-
sitive to microcracks and imperceptible defects
rigid biominerals such as nacre are not effective
for the design of such flexible materials. Besides
the lack of flexibility, the fatigue tolerance of
these models strongly relies on the rising R-curve
behavior during crack propagation (11), yet the
crack extension can cause irreversible impacts
on the device performances (16, 17). This indi-
p
(22). This correlation can be explained by cal-
cium carbonate in the form of aragonite being
the only mineral phase in the two hinge re-
gions that can be observed with powder x-ray
diffraction (fig. S8) (23).
We monitored the two hinge regions to un-

and thus suffer from the risk of cumulative
fatigue damage caused by the prolonged cyclic
loading, which can eventually cause catastrophic
failure (4–7). In this respect, living organisms
produce rigid biominerals such as nacre (8),
cates that the protection of the brittle compo-
nents in such devices against fatigue should
not be overdependent on the mechanisms that
only come into effect in the crack wake (2, 11).
g
ravel their roles during the ROAC motions
(Fig. 2A and movie S4). During closing, the
FFR region is pushed by the rotating slab areas
that are bound to the valves. The two sides of
the FFR then rotate, and the whole FFR re-

bone (9), and tooth (10) in which the fatigue
tolerance is greatly improved but the native
high strength and rigidity of the minerals are
retained. The fatigue damage in these biomin-
erals, such as crack propagation, can be shielded
by extrinsic mechanisms, including crack bridg-
ing, deflection, and branching (8–11).
Although these biominerals have inspired
fatigue-tolerant artificial materials such as
tough biomimetic ceramics (12, 13), more and
Macrostructures and mechanical performance
We report the antifatigue design of the de-
formable calcareous tissue in the hinge of
the bivalve shell Cristaria plicata (Fig. 1A, i,
and fig. S1A) (18). This tissue exhibits both
high deformability and exceptional fatigue
resistance during the hundreds of thousands
of repeating opening-and-closing (ROAC) mo-
tions of the shell valves. The hinge is located
y
gion simultaneously bends and deforms in the
circumferential direction, which is accompa-
nied with the stretch of the OL (Fig. 2A). The
FFR undertakes a minor radial deformation
and provides robust radial support to fix the
OL to ensure effective OL circumferential stretch.
We validated this behavior by means of finite
element analysis (FEA) during the closing
process (Fig. 2B, fig. S9, and movie S5), and
the behavior reverses during opening. There-

diverse design principles have to be uncovered
to extend the scope of artificial materials. For
example, materials with high flexibility have
received intense focus owing to the develop-
ment of foldable and wearable devices (14, 15).
at the dorsal edge of the bivalve shell, by which
the two valves are joined together, and func-
tions as the axis of the ROAC motions (Fig. 1,
A, ii, and B, i; fig. S1; and movie S1) (19). It un-
dergoes large deformation during the clos-
y g
fore, the structure of the FFR, which is sim-
ilar to that of traditional arch structures, can
effectively translate the external radial load
to the circumferential deformation. We also
calculated the stress distributions in the FEA

Existing antifatigue models derived from very
1
Department of Chemistry, Institute of Biomimetic Materials
and Chemistry, Anhui Engineering Laboratory of Biomimetic
Materials, New Cornerstone Science Laboratory, Division of
Nanomaterials and Chemistry, Hefei National Research
Center for Physical Sciences at the Microscale, University of
Science and Technology of China, Hefei 230026, China.
2
CAS Key Laboratory of Mechanical Behavior and Design of
Materials, Department of Modern Mechanics, CAS Center
for Excellence in Complex System Mechanics, University of
Science and Technology of China, Hefei 230027, China.
3
School of Mechanical Engineering, Hefei University of
Technology, Hefei 230009, China. 4Anhui Shuyan Intelligent
Technologies Co., Wuhu 241200, China. 5Institute of
Innovative Materials, Department of Materials Science and
Engineering, Department of Chemistry, Southern University
of Science and Technology, Shenzhen 518055, China.
*Corresponding author. Email: maolb@ustc.edu.cn (L.-B.M.);
wuha@ustc.edu.cn (H.-A.W); shyu@ustc.edu.cn (S.-H.Y.)
†These authors contributed equally to this work.
ing process of two rigid valves driven by the
adductor muscles and provides the driving
force for the spontaneous valve opening by
releasing the stored elastic potential energy
(movie S2) (20).
To verify the mechanical and antifatigue
performance of the hinge, we performed cyclic
loading tests that simulated the ROAC mo-
tions on freshly prepared C. plicata samples
(fig. S2A and movie S3) (21). No obvious fatigue
failure was observed even after 1,500,000 cycles
under natural working conditions (Fig. 1C and
figs. S3 and S4A). We also imposed a larger
load (about 20% overload) on the edge-trimmed
samples (fig. S2), and the hinge exhibited
similar behavior in this case (Fig. 1C and figs. S4B
and S5). To investigate the antifatigue per-
,
model along both circumferential and radial
directions at the open and closed states, re-
spectively (Fig. 2, C to E, and supplementary
text 1). The circumferential tensile stress in
the hinge is mainly borne by the OL (Fig. 2D
and fig. S10), which agrees with the mea-
sured result that the stress along the circum-
ferential direction is much higher in the OL
than that in the FFR (Fig. 2, F and G, and
fig. S10).
In addition, given that the OL and the out-
most edge of the FFR are bound together,
the deformations are supposed to be similar
for both when stretched. The elastic strain
energy released by the FFR and the OL was
derived by integrating the products of the
energy density of each element by its volume

Meng et al., Science 380, 1252–1257 (2023) 23 June 2023 1 of 6

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

A z B z z C 150
i i MaxNWC
x y y MinNWC
Max120%
120

Relative load (%)

Min120%

90

ii ii 60

30

1 10,000 100,000 1,000,000

Hinge
Number of cycles
SA FFR OL PT Valve

E 60
D Ca O 3.0

Modulus (GPa)
Hardness (GPa)
40
2.0

p
20
1.0

0.0 0
4
OL FFR SA 64

Modulus (GPa)
g
Relative intensity

Hardness (GPa)

3
48
Ca
2 32
O 1 16

y
0 0
OL FFR SA
0 600 1200 1800 2400 0 150 300 450 600
Distance (µm) Distance (µm)

Fig. 1. Structural features and mechanical properties of the hinge. (A) (i)
Optical image of C. plicata and (ii) sectional photograph of the shell at the cutting
plane (dotted rectangle) shown in (i). The yellow dash-dotted line in (i) represents
the hinge axis as well as the axis of the ROAC motions. This axis is indicated
with the circled dot in (ii). Scale bars, 2 cm. (B) (i) Optical image and (ii)
lighter gray color than that of the nacreous part of the valves, indicating a higher
electron density. Scale bars, 1 cm. (C) Relative maximum and minimum loads in
the fatigue tests under the natural working condition (NWC) and overloading
condition (about 20% overload) (supplementary materials, materials and methods).
(D) Elemental maps and line scans showing the calcium (Ca) and the oxygen (O)

reconstructed three-dimensional mCT image of a polished hinge sample viewed in the
longitudinal direction [(A), ii, dashed box]. The different colors in (ii) indicate the
distinct regions in the hinge: FFR, blue; OL, saddle brown; slab areas (SAs), sienna;
porous tissue (PT), light cyan; and valve, pale yellow. The SAs exhibit a slightly
y g
distributions in the yellow square in (B), in which the FFR, the OL, and the SAs
are included. Scale bars, 500 mm. (E) Maps of the elastic modulus and hardness of
the yellow square in (B) and the corresponding line scans from the OL to the FFR
and then the SA of a dried sample. Scale bars, 200 mm.

in the numerical model. The results suggest
that the OL stores most of the elastic strain
energy during closing, which is then released
to sustain valve opening (supplementary text
1 and fig. S11). Compressive tests show that
the FFR is highly anisotropic (Fig. 2, G and
H). In the radial direction, the whole FFR is
under compressive stress (Fig. 2E), which is
associated with the radial support of the FFR
to the OL. Nevertheless, the deformation of
the FFR in this direction is limited by the space
of the entire hinge area (Fig. 2, A and B). Com-
pensatorily, the FFR has a much larger tangent
modulus along the radial direction compared
with that along the circumferential direction
(Fig. 2, G and H). Therefore, it can translate
sufficient radial load to support the OL within
a very small radial deformation. By contrast,
the FFR can deform much easier circumfer-
entially, which allows the FFR to adapt itself
to the hinge deformation. These observations
reveal that as a dense and relatively rigid cal-
careous tissue (figs. S8 and S12 and tables S1
and S2), the FFR endures a large radial load to
support the OL while undertaking a large cir-
cumferential deformation during the ROAC mo-
tions of the valves without fatigue failure (more
than 1,500,000 cycles). This distinguishes the
FFR from other biominerals such as nacre, cor-
tical bone, and human enamel (table S3). Once
produced, the acellular FFR begins to func-
tion during the ROAC motions. Yet it neither
comes in contact with cells, like the nacreous
inner layer of the valves does, nor contains any
,
cells inside and thus cannot be repaired after
being damaged, suggesting that the FFR has
to be inherently robust (fig. S13) (24, 25).
Microstructures and crystallographic features
To understand the mechanisms of the mechan-
ical function and antifatigue performance of
the FFR, we analyzed its microstructures and
crystallographic features. The fracture surface
of the FFR exhibits a concentrically laminated
structure (Fig. 1B, i, and fig. S14). Each layer
consists of tightly aligned and stacked, long
nanowires with diameters of about 100 to
200 nm (Fig. 3A). Detailed measurements re-
veal that the diameters increase gradually
from the inner part of one nanowire layer to
the outer part (fig. S15). Such laminated structure

Meng et al., Science 380, 1252–1257 (2023) 23 June 2023 2 of 6

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

Fig. 2. Transmission relationship of the ROAC motions. (A) Optical images
and (B) FEA models of the hinge in an open state and a closed state. The contours of
the FFR and the OL at the two states are compared. Yellow, open state; red, closed
state. Scale bar, 500 mm. (C to E) FEA of the stress distributions in (C) the hinge
p
g
y
along (D) the circumferential and (E) radial directions. The red arrows indicate the
directions of the stresses. (F) Tensile tests of the OL and (G and H) compressive tests
of the FFR (freshly prepared wet sample) along the (G) circumferential and (H) radial
directions, showing the different mechanical performances of these regions.

and varying nanowire diameters can help the phological and the h002i crystallographic ori- that is parallel to the nanowires, which is the
FFR accommodate the nanowire space-filling entations of the aragonite nanowires match radial direction in the FFR, the RVE can trans-

pattern.
Additionally, the cross section of each nano-
wire exhibits a pseudo-hexagonal shape (Fig.
3B) (26). We decalcified an FFR sample with
an ethylenediaminetetraacetic acid disodium
each other. This preferred h002i growth orien-
tation of the nanowires is also the fastest growth
direction during abiotic aragonite crystallization
(29, 30). Although biological control can substan-
tially change the mineralization process, the
y g
fer a large load with a limited deformation (fig.
S17, B and D). We also found that the RVE with
aragonite nanowires uniformly oriented in
the h002i direction exhibits a higher stress level
than that oriented in the h100i=h010i at the

salt dihydrate solution to remove the aragonite
minerals (27). The remnant shows a typical
honeycomb-like structure, indicating that the
aragonite nanowires are embedded in a con-
tinuous organic matrix (Fig. 3, C and D; fig.
S12; and table S2). The nanowire orientations
at several sampling positions in the fracture
surface of the FFR suggest that they are ra-
dially oriented in the whole region, which are
similar to the ribs in a folding fan (Fig. 3I and
fig. S16). Micro x-ray diffraction mapping of
the same surface reveals that the aragonite
nanowires uniformly grow along the h002i
crystallographic direction (Fig. 3J) (28), which
agrees with the high-resolution transmission
electron microscopy (HRTEM) analysis (Fig. 3,
E to H). These results indicate that the mor-
crystallographic anisotropy in growth rate can
contribute to the biominerals formation with
highly anisotropic morphologies, leading to a
more cost-effective biomineralization process
(31, 32).
Origins of deformability and fatigue resistance
To study the role of the morphological and
crystallographic orientations of the aragonite
nanowires in functionality, we extracted a rep-
resentative volume element (RVE) from the
FFR and simulated its response to loads along
different directions (fig. S17). In the direction
that is normal to the nanowires, which is the
circumferential direction in the FFR, the RVE
can easily deform because of its relatively low
modulus (fig. S17, A and C). In the direction
,
same strain (fig. S17D), indicating that the
former can transfer larger radial loads. Given
that the h002i crystallographic direction has
the fastest growth rate during the aragonite
crystallization (30), the above analyses suggest
an elegant consistency of the mechanically
favored orientation, the thermodynamically
favored orientation, and the actual nanowire
crystal growth direction. Consequently, although
the FFR is fabricated in a cost-effective way con-
cerning the nanowire growth direction, the
microstructures and crystallographic features
of the FFR are also optimized for the sake of
good circumferential deformability and radial
load translation capability, both of which are
the cornerstones of the long-term functioning
performance of the FFR (Fig. 2).

Meng et al., Science 380, 1252–1257 (2023) 23 June 2023 3 of 6

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

Fig. 3. Microstructural and crystallographic features of the aragonite nano-
wires in the FFR. (A) Longitudinal and (B) cross-sectional fracture surfaces of the
FFR showing the oriented nanowires with a pseudo-hexagonal shape. Scale bars,
(A) 1 mm and (B) 400 nm. (C) Longitudinal and (D) cross-sectional fracture surfaces
p
g
y
5 nm and (inset) 2 nm−1. (G) TEM and (H) HRTEM images of an ultrathin longitudinal
section of the FFR. The aragonite crystals grow along the h002i direction. Scale bars,
(G) 200 nm and (H) 3 nm. (I) Morphological orientation of nanowires in the FFR
(indicated with the white fan shape). The images in the region were obtained by

of the decalcified FFR showing the honeycomb structure. Scale bars, 400 nm.
(E) TEM image of a transverse ultrathin section of the FFR. The yellow arrows indicate
the twin boundaries in the nanowires. Not all the boundaries are visible because
of the tilting angles. Scale bar, 200 nm. (F) HRTEM image of an aragonite nanowire
showing the twin boundary. (Inset) Selected area electron diffraction image of the
twin structure revealing that the twin boundary is the (110) plane. Scale bars,
y g
means of fast Fourier transform (FFT) of the scanning electron microscopy (SEM)
images (for example, the FFT image in the box is transformed from the SEM
image in the dashed box). Scale bar, 1 mm. (J) Crystallographic orientation h002i
(black lines) of the aragonite nanowires in the FFR, which is acquired from
the crystal diffractogram. The color of each square indicates the deflection angle of
h002i orientation against the mirror plane of the FFR.

Because the morphological and crystallo-
graphic orientations of these ordered nano-
wires are the same, the internal stress in the
FFR can be deduced from the stress-induced
lattice distortion of the aragonite crystals (33).
We therefore investigated the stress states of
some microdomains in the FFR by means of
high-resolution synchrotron x-ray diffraction.
The cell parameter c of the aragonite nano-
wires at the closed state decreases compared
with that at the open state (Fig. 4A and fig. S18),
whereas the changes of parameters a and b
are much smaller. Because the nanowires grow
along the h002i direction, the result indicates
that each aragonite nanowire experiences a
large radial load along the longitudinal direc-
tion. However, although the FFR endures a
larger deformation in the circumferential di-
rection, its elastic modulus along this direction
is smaller, and it is the organic matrix that
bears most of the deformation (fig. S17). This
explains the much smaller changes of a and b
as well as the small circumferential loads on
the nanowires. The result is thus in good agree-
ment with the properties of the RVE.
The ROAC motions can be implemented hun-
dreds of thousands of times during the course
of the shell’s lifetime, and thus the FFR that
exerts distinct functionalities in different direc-
tions also has to be fatigue resistant (Fig. 1C and
fig. S4). Because the aragonite nanowires are ,
extremely long and brittle (34), they should
have broken off easily as the FFR undergoes
the radial load from the OL. In this respect, the
organic matrix, which is resistant to fracture
and wraps around each nanowire, can prevent
the fragile nanowire from bending and break-
ing (fig. S19). We evaluated the stress states of
the nanowires during the ROAC motions to see
how the nanowires can sustain the motions. Be-
cause the FFR bears a circumferential deforma-
tion, lateral compressive stress or tensile stress
should thus be applied to the nanowires.
Atomic-force microscopy (AFM) observation
of the FFR fracture surface in liquid reveals,

Meng et al., Science 380, 1252–1257 (2023) 23 June 2023 4 of 6

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

Fig. 4. In situ stress state analyses of the nano- A B C

wires. (A) High-resolution synchrotron x-ray diffraction Δa (%) Δb (%) Δc (%)
dc do 120 nm

-0.3
-0.2
-0.1

-0.3
-0.2
-0.1

-0.3
-0.2
-0.1
0.0

0.0

0.0
analysis of the lattice constant variation (from open to Up

0
3
2
1

0
3
2
1

3
2
1
closed) of the aragonite crystals. (B) AFM images

Up
obtained at “up,” “middle,” and “down” positions in the -110 nm
Up

Sampling position
FFR (fig. S22B) at the open and closed states of the
valves. Scale bars, 500 nm. (C) Shift of relative 340 nm Middle

Middle
position (Dd) of two adjacent nanowires. dc do
Middle
-370 nm

230 nm
Down

Down
Down dc do
40 50 60 70
-300 nm
Δd (nm)

Deformability

I
10 µm - 10 mm

II

p
Folding fan shape
II

g
100 nm - 10 µm

Hard fibers and soft matrix

Weak interfaces III

y
Radially aligned
nanowires

III

Bicontinuous phase
Grain refinement networks
10 - 100 nm

Nanorod Microcracks and Inclusions

remineralization and ligament bridges IV

y g
Protein

Hard-soft combination

IV

,
Cross-linked networks
<10 nm

and crystalline domains

Nanorod TB GB

Chemical and Sliding and Grain boundary

structural gradients sacrificial bonding and twin boundary Twin boundary

Enamel Cortical bone Metal FFR Elastomer

Fig. 5. Antifatigue design of typical biological and artificial structural
materials. The antifatigue performance of the structural materials with diverse
stiffness relies on their specific structures. Enamel: Chemical and structural
gradients inhibit the fatigue crack initiation and propagation; damages can be
repaired by the hydroxyapatite nanorod remineralization (10, 39). Cortical bone:
Sacrificial bond rupture and collagen fibril sliding impede the crack initiation;
microcrack, ligament bridges, and weak interface induce the crack deflection and
slows the crack propagation; remodeling substitutes old or damaged bone with
new bone tissue (9, 40). Metal: Twin boundaries (TB) and grain boundaries (GB)
suppress dislocation activities; nano inclusions retard microcrack propagation (41).
FFR: External load is translated by the folding fan–like structure (I) to
circumferential deformation without inducing stress concentration; the hard-soft
complex structure consisting of radially aligned nanowires (II) deforms
compatibly to sustain the circumferential deformation, while the soft matrix (III)
undertakes most of the volume compression; twin boundaries (IV) along the
length direction of the nanowires provide a bottom-level guarantee to the
aragonite nanowires against bending fracture. Elastomer: Cross-linked networks
and crystalline domains raise the threshold of energy release rate; bicontinuous
phase networks and a hard fiber and soft matrix complex enhance the crack
deceleration and blunting (42).

Meng et al., Science 380, 1252–1257 (2023) 23 June 2023 5 of 6

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

through comparison of the relative positions of
adjacent nanowires, the nanowires’ axial sliding
indicating a shear force around the nanowires as
well as the lateral stresses (Fig. 4, B and C, and
fig. S20) (35). We thus simulated the three stress
states—compression, tension, and shear—in a
two-dimensional numerical model (fig. S21). The
FEA results suggest that it is the matrix that
bears most of the compressive and shear strains
in all the states; no stress concentration is
found in the nanowires or the organic matrix.
Therefore, the deformation compatibility of
this hard-soft complex structure can effectively
reduce the possibility of brittle rupture of the
aragonite nanowires, by which the fatigue
damage of the FFR can be suppressed.
Aragonite belongs to the Pmcn space group
in which (110) twinning planes can easily de-
velop in aragonite crystals. This makes these
orthorhombic crystals exhibit a pseudohex-
agonal appearance (Fig. 3, B and E) (36). As
inherent crystallographic characteristics of
aragonite for the biomineralization of the
FFR; this tactic is rarely used in the fabrication
of artificial structural materials. Although there
is much to learn from this biomineral, we fab-
ricated a proof-of-concept glass-polymer com-
posite with the FFR-like microstructure as a
primitive demonstration (fig. S22). The com-
posite exhibits anisotropic mechanical behav-
iors and good fatigue resistance similar to
those of the FFR (supplementary text 2). This
biomineral provides a multiscale model for de-
signing structural materials that contain brit-
tle components, such as ceramics, yet require
both deformability and fatigue resistance.
RE FERENCES AND NOTES
1. D. Gu et al., Science 372, eabg1487 (2021).
2. S. Bhumiratana et al., Sci. Transl. Med. 8, 343ra83 (2016).
3. S. Park, H. Park, S. Seong, Y. Chung, Sci. Rep. 10, 7660 (2020).
4. K. J. Miller, Mater. Sci. Technol. 9, 453–462 (1993).
5. C. E. Renshaw, E. M. Schulson, Nature 412, 897–900 (2001).
6. J. Schijve, Int. J. Fatigue 25, 679–702 (2003).
36. W. L. Bragg, Proc. R. Soc. A Contain. Pap. Math. Phys. Character
105, 16–39 (1924).
37. A. J. Cao, Y. G. Wei, S. X. Mao, JOM 60, 85–88 (2008).
38. Y. Tian et al., Nature 493, 385–388 (2013).
39. K. A. DeRocher et al., Nature 583, 66–71 (2020).
40. J. Li, H. Gong, Acta Mech. Sin. 37, 516–526 (2020).
41. D. Rozumek, Metals (Basel) 11, 1957 (2021).
42. Y. Huang et al., Soft Matter 18, 5153–5165 (2022).
43. S.-H. Yu et al., Deformable hard tissue with high fatigue
resistance in the hinge of bivalve Cristaria plicata. Dryad (2023);
https://doi.org/10.5061/dryad.69p8cz95j.
AC KNOWLED GME NTS
The authors thank P. Fratzl from Max Planck Institute for helpful
comments and discussions on this work and the manuscript. The
authors also thank Z.-L. Zhu for providing micro x-ray diffraction
mapping; Y.-G. Gu for fatigue tests; J. Tian and S.-Q. Fu for
providing scanning electron microscopy tests and element
mapping; T.-W. Li and Y.-R. Wang for providing high-resolution
transmission electron microscopy analyses; Z. He, S.-C. Zhang,
and J. Pang for AFM tests; Y.-T. Wei (Bruker) and J.-P. Wang
(Bruker) for supporting nanoindentation tests; G.-Y. Gao for
powder x-ray diffraction tests; W. Wen for high-resolution x-ray
diffraction tests; and B.-H. Zhan and R.-D. Wang (Beijing Institute
of Fashion Technology) and C.-X. Yu (Anhui University) for their
assistance in visualizing this work. This research used Beamline
BL14W of the Shanghai Synchrotron Radiation Facility (SSRF)

the aragonite nanowires in the FFR grow along
the h002i direction, abundant twin boundaries
are found in parallel with the axial direction of
the nanowires (Fig. 3, E and F). Such twin
boundaries can improve the nanowire re-
sistance to bending fracture (37, 38). There-
p
7. J. J. Kruzic, Science 325, 156–158 (2009). for high-resolution synchrotron x-ray diffraction tests. The authors
also thank Shiyanjia Lab (www.shiyanjia.com) for hematoxylin and
8. Y. Shao, H.-P. Zhao, X.-Q. Feng, H. Gao, J. Mech. Phys. Solids
eosin staining. This work was partially carried out at the USTC
60, 1400–1419 (2012).
9. M. E. Launey, M. J. Buehler, R. O. Ritchie, Annu. Rev. Mater. Center for Micro and Nanoscale Research and Fabrication.
Res. 40, 25–53 (2010). Funding: This work was supported by financial support from the
10. D. Bajaj, A. Nazari, N. Eidelman, D. D. Arola, Biomaterials 29, National Key Research and Development Program of China
(grants 2021YFA0715700 and 2018YFE0202201) and the National
4847–4854 (2008).
11. R. O. Ritchie, Nat. Mater. 10, 817–822 (2011). Natural Science Foundation of China (grants 21701161, 22293044,

fore, in addition to the higher stress transfer
capability of the h002i orientation of the arag-
onite nanowires, this orientation is also asso-
ciated with the nanotwin boundary formation
that plays a key role in preventing the fracture
g
12. Z. Ma et al., Sci. Rep. 6, 22156 (2016). 12232016, and 12172346). Y.-B.Z. acknowledges funding support
from the Youth Innovation Promotion Association CAS (2022465).
13. T. Magrini et al., Nat. Commun. 10, 2794 (2019).
This work has been supported by New Cornerstone Science
14. M. J. Kim et al., Sci. Adv. 7, eabg8169 (2021).
Foundation. Author contributions: S.-H.Y. and L.-B.M. conceived
15. C. Xu, Y. Song, M. Han, H. Zhang, Microsyst. Nanoeng. 7, 25
(2021). the idea and designed the experiments. S.-H.Y., L.-B.M., X.-S.M.,
L.-C.Z., and L.L. wrote and edited the paper. X.-S.M., Y.-F.M., B.Y.,

of the nanowires under cyclic stress states.
Conclusions and outlook
We have revealed the hierarchical structure
design of the FFR in the hinge of C. plicata,
which spans from the macroscale level down
to the lattice level. This design is not a simple
accumulation of isolated antifatigue mecha-
nisms; rather, each aspect works synergisti-
cally (Fig. 5). The notable deformability and
y
16. E. H. Cho et al., Nanoscale 10, 628–638 (2018).
17. G. Zan et al., Matter 4, 3232–3247 (2021). and Q.-Z.R. performed the experiments and analyzed the data.
L.-C.Z., Y.-B.Z., and H.-A.W. performed theoretical analyses. Y.-B.Z.
18. E. Savazzi, Y. A. O. Peiyi, Lethaia 25, 195–209 (1992).
19. S. A. Wainwright, Nature 224, 777–779 (1969). and D.-C.Z. provided valuable advice for mechanical analyses.
20. K. Ono, Y. Kikuch, K. Higashi, N. Tamiya, N. Yasuoka, J. Biomech. Y.-B.Z. and L.-B.M. contributed to antifatigue mechanisms. All
23, 307–312 (1990). authors discussed the results. Competing interests: All authors
declare that they have no competing interests. Data and
21. A. Fernández-Canteli, S. Blasón, B. Pyttel, M. Muniz-Calvente,
E. Castillo, Int. J. Fract. 223, 189–196 (2020). materials availability: Data are available in the manuscript or the
22. J. C. Weaver et al., Science 336, 1275–1280 (2012). supplementary materials or are deposited in Dryad (43). License
23. D. Labonte, A.-K. Lenz, M. L. Oyen, Acta Biomater. 57, 373–383 information: Copyright © 2023 the authors, some rights reserved;
exclusive licensee American Association for the Advancement of
(2017).
24. G. Bevelander, H. Nakahara, Calcif. Tissue Res. 4, 101–102 (1969). Science. No claim to original US government works. https://www.
25. C. Ituarte, D. G. Zelaya, J. Molluscan Stud. 81, 455–465 (2015). science.org/about/science-licenses-journal-article-reuse
26. M. E. Marsh, R. L. Sass, Science 208, 1262–1263 (1980).

y g
27. C. Picart, D. E. Discher, Nature 448, 879–880 (2007). SUPPLEMENTARY MATERIALS
load translation capability of the FFR orig- 28. O. Paris, Biointerphases 3, FB16 (2008).
inate from the hierarchical structures, which science.org/doi/10.1126/science.ade2038
29. A. G. Checa, T. Okamoto, J. Ramírez, Proc. Biol. Sci. 273, 1329–1337 (2006).
Materials and Methods
cannot be achieved by any specific mechanism 30. V. Schoeppler et al., Proc. Natl. Acad. Sci. U.S.A. 116,
Supplementary Text
20388–20397 (2019).
that acts at only a few length scales. The com- Figs. S1 to S23
31. P. U. P. A. Gilbert et al., J. Am. Chem. Soc. 130, 17519–17527 (2008).
bination of the functionality and fatigue re- Tables S1 to S3
32. F. Nudelman, Semin. Cell Dev. Biol. 46, 2–10 (2015).
References (44–53)

,
sistance of the FFR exemplifies how the service 33. B. Pokroy, J. P. Quintana, E. N. Caspi, A. Berner, E. Zolotoyabko,
MDAR Reproducibility Checklist
life of a material can be prolonged by exploit- Nat. Mater. 3, 900–902 (2004).
Movies S1 to S5
34. Z. Wu, Y.-W. Zhang, M. H. Jhon, H. Gao, D. J. Srolovitz, Nano Lett.
ing the intrinsic properties of each component. 12, 910–914 (2012). Submitted 1 August 2022; accepted 25 April 2023
Furthermore, C. plicata shell also exploits the 35. M. Frey et al., Adv. Sci. (Weinh.) 6, 1802190 (2019). 10.1126/science.ade2038

Meng et al., Science 380, 1252–1257 (2023) 23 June 2023 6 of 6

RES EARCH

CELLULAR IMMUNOLOGY We then used the saporin-puromycin assay

in a CRISPR-Cas9–based screen of 281 genes
Perforin-2 is a pore-forming effector of endocytic highly expressed in cDC1s compared with cDC2s
(fig. S3). We used a mix of Atto550-labeled and
escape in cross-presenting dendritic cells unlabeled saporin, gated on cells with similar
uptake efficiency, and sorted the cells into two
Pablo Rodríguez-Silvestre1, Marco Laub1, Patrycja A. Krawczyk1, Alexandra K. Davies2†, bins: purolow (saporin escape, translation arrest)
Julia P. Schessner2, Reejuana Parveen1, Benjamin J. Tuck1,3, William A. McEwan3, and purohigh (saporin retention, efficient trans-
Georg H. H. Borner2, Patrycja Kozik1* lation) (fig. S3C). In the absence of saporin, none
of the single guide RNAs (sgRNAs) affected the
During initiation of antiviral and antitumor T cell–mediated immune responses, dendritic cells (DCs) translation rate (fig. S3E and table S1). The
cross-present exogenous antigens on major histocompatibility complex (MHC) class I molecules. strongest hit in saporin-pulsed cells was Mpeg1
Cross-presentation relies on the unusual “leakiness” of endocytic compartments in DCs, whereby (Fig. 1C), with the four sgRNAs enriched in
internalized proteins escape into the cytosol for proteasome-mediated generation of MHC I–binding purolow versus purohigh populations (fig. S3F).
peptides. Given that type 1 conventional DCs excel at cross-presentation, we searched for cell type–
Perforin-2 is necessary for endocytic
specific effectors of endocytic escape. We devised an assay suitable for genetic screening and identified
escape in DCs
a pore-forming protein, perforin-2 (Mpeg1), as a dedicated effector exclusive to cross-presenting cells.
Perforin-2 was recruited to antigen-containing compartments, where it underwent maturation, releasing its Mpeg1 encodes perforin-2, a member of the
pore-forming domain. Mpeg1−/− mice failed to efficiently prime CD8+ T cells to cell-associated antigens, membrane attack complex (MAC) and perforin
revealing an important role for perforin-2 in cytosolic entry of antigens during cross-presentation. superfamily (MACPF) of pore-forming proteins
(23). Perforin-2 can form oligomeric pores on

T
he integrity of endosomal and lysosomal
membranes is critical to protect the cell
against extracellular pathogens and toxins,
and from the activity of lysosomal hydro-
lases. In dendritic cells (DCs), however,
Here, to identify DC-specific regulators
of endocytic escape, we developed a flow
cytometry–based assay to monitor escape in
individual cells and applied it in a CRISPR-
Cas9–based screen using cells specialized in
p
liposomes with an opening of at least 75 Å in
diameter (24–26) (Fig. 1D). It was initially pro-
posed that perforin-2 pores facilitate killing of
intravacuolar bacteria (27), but these results
were not replicated in a recent study (28). Here,
we explored whether perforin-2 can function

internalized proteins are delivered from endo-
cytic organelles into the cytosol, where they can
be proteolytically processed for presentation
on major histocompatibility complex (MHC)
class I molecules (1). The ability of DCs to present
cross-presentation, conventional DC1s (cDC1s).
A CRISPR-Cas9 screen identifies Mpeg1
(perforin-2) as a regulator of endocytic escape
To monitor endocytic escape, we used the
g
as an effector of endocytic escape.
We generated Mpeg1KO MutuDCs and con-
firmed protein depletion in sorted, sgRNA-
expressing blue fluorescent protein (BFP)
cells (fig. S4A). To account for the effect of

exogenous peptides on endogenous MHC-I is
termed cross-presentation. Cross-presentation
is critical for initiation of cytotoxic T cell (CTL)
responses to antigens not expressed in DCs
such as neoantigens or antigens from virally
infected cells (2–4).
Various mechanisms have been suggested
to facilitate endocytic escape and promote
cross-presentation (5–10). Early studies pro-
posed that escape is mediated by protein
28-kDa type I ribosome-inactivating protein
(RIP) saporin (5). Once in the cytosol, RIPs
arrest translation through depurination of the
sarcin-ricin loop in the 28S subunit of the
ribosome (15). In contrast to type II RIPs, such
as ricin, which comprise a domain that facil-
itates entry into the cytosol, cytosolic delivery
of type I RIPs is dependent on the cell-intrinsic
efficiency of endosome-to-cytosol transport (5).
To detect saporin-induced translation inhibi-
y
passage numbers on MutuDC behavior (20), we
cultured control cells expressing nontargeting
(NT) sgRNA in parallel with the knockout (KO)
line. We first demonstrated that disruption of
Mpeg1 protected the cells from the cytotoxic
effects of saporin and from death induced by
a different RIP, gelonin (fig. S5A). We also tested
the sensitivity of Mpeg1KO DCs to a glycopeptide
chemotherapeutic, bleomycin A2, which induces
DNA damage, but owing to its hydrophilicity

channels (such as Sec61) recruited from the
endoplasmic reticulum to endosomes (7, 11).
More recent data suggest that escape occurs
through unrepaired damage to membranes
(e.g., due to reactive oxygen species–driven
lipid peroxidation) (6, 9, 10). Both models
tion, we monitored incorporation of puromycin,
a structural analog of aminoacyl tRNAs, into
nascent polypeptides (16). We labeled intra-
cellular puromycylated polypeptides with a
fluorescent 12D10 antibody allowing for flow
y g
does not enter the cells efficiently (29). We
used automated imaging to monitor MutuDC
growth rate and found that loss of perforin-2
rendered the cells more resistant to bleomycin-
mediated cytotoxicity (fig. S5A). Notably, the
Mpeg1KO cells were not protected against the

imply that the unusual “leakiness” of endo-
cytic compartments in DCs might not rely on
any cell type–specific effectors, but on the
regulation of ubiquitously expressed proteins
through signaling (12, 13) and trafficking (14)
events specific to cross-presenting cells.
1
MRC Laboratory of Molecular Biology, Cambridge, UK.
2
Department of Proteomics and Signal Transduction,
Max Planck Institute of Biochemistry, Martinsried, Germany.
3 previously reported differences in escape effi-
UK Dementia Research Institute at the University of
Cambridge, Department of Clinical Neurosciences,
Cambridge, UK.
*Corresponding author. Email: pkozik@mrc-lmb.cam.ac.uk
†Present address: School of Biological Sciences, Faculty of Biology,
Medicine and Health, Manchester Academic Health Science Centre,
University of Manchester, Manchester, UK.
cytometry–based readout of translation effi-
ciency and thereby of endocytic escape (Fig. 1A).
cDC1s, a subset of DCs that excels at cross-
presentation in vivo (17), have been reported
to have the most efficient endocytic escape
pathway (17, 18). We thus developed the
saporin-puromycin assay using a murine cDC1-
like cell line, MutuDCs (13, 19, 20) (Fig. 1B and
fig. S1, A and B). We confirmed that saporin
escape is the rate-limiting step in the assay (fig. S1,
C and D) and that the assay recapitulates
ciency, namely, more efficient escape in cDC1s
compared with cDC2s (fig. S2, A and C) (18, 21)
and enhanced “leakiness” of endocytic com-
partments in cells from lupus-prone MRL/
MpJ-Faslpr/J mice (fig. S2, B and D) (22).
,
effects of poly(I:C), which induces cell death
via endosomal Toll-like receptor 3 (TLR3) (30),
or against membrane-permeable cycloheximide
(fig. S5A).
We next verified that the sensitivity of
perforin-2–expressing DCs to cytosolic toxins
is due to efficient escape rather than due to
efficient uptake. In the saporin-puromycin
assay with Atto550-labeled saporin, endocytic
escape was impaired in Mpeg1KO DCs, even
though they internalized similar amounts of
saporin compared to NT cells (Fig. 1E). The
differences in escape were also not due to
changes in DC activation, because neither NT
nor Mpeg1KO MutuDCs were activated by
saporin (fig. S5B). Finally, we rescued saporin

Rodríguez-Silvestre et al., Science 380, 1258–1265 (2023) 23 June 2023 1 of 8

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

Fig. 1. Saporin-puromycin assay to monitor endocytic escape in DCs.
(A and B) Saporin-puromycin assay. (A) Schematic representation. Saporin-
pulsed cells are labeled with puromycin to monitor translation rate. Puromycin
p
g
alongside structures of single-subunit (PDB ID: 6U2K) and the hexadecameric
perforin-2 pore (PDB ID: 6SB5). (E) Quantification of translation arrest in
Mpeg1KO and NT MutuDCs. Cells were pulsed with saporin (11:1 unlabeled:

incorporated into nascent peptides is detected with an aPuromycin Ab and
flow cytometry. If saporin is retained within the endosomes, translation remains
high. When saporin escapes into the cytosol, it depurinates ribosomes,
inducing translation arrest. (B) Representative flow cytometry plot. MutuDCs
were incubated with 0.5 mg/ml of saporin followed by 0.01 mg/ml puromycin
(purple histogram), with puromycin alone (yellow), or in media only (gray). Cells
in translation arrest are denoted by the purple, gate. See also fig. S1B.
(C) Volcano plots showing the sgRNA enrichment analysis for the saporin-
puromycin endocytic escape screen. Each of the dots represents one targeted
gene. Data represent the combined mean enrichment scores and the non-
y
Atto550-labeled saporin) for 2 hours, and translation was monitored by a
30-min puromycin chase. The x axis represents Atto550 mean fluorescence
intensity (MFI) normalized to the NT MutuDC Atto550 MFI at the highest
saporin concentration. Data represent mean and SEM of three independent
experiments; ns, not significant; **P < 0.01 using a multiple unpaired t test
(two-stage step-up, Benjamini, Krieger, and Yekutieli). Significance symbols in
the plot refer to the differences in proportion of cells in translation arrest.
Differences in saporin Atto550 MFI were not significant. See also fig. S4A.
(F) Mpeg1KO MutuDCs were reconstituted with the indicated Mpeg1 mutants
or with mScarlet only and used in the saporin-puromycin assay with a 2-hour

y g
adjusted P values from three independent experiments (Fisher’s method). See pulse with 0.1 mg/ml saporin. Data are representative of three independent
also fig. S3. (D) Schematic representation of the different perforin-2 domains experiments. See also fig. S4C.

import without affecting uptake by transducing Thus, perforin-2 pores mediate endocytic escape marin linked by a b-lactam ring, which is
Mpeg1KO cells with full-length sgRNA-resistant in cross-presenting DCs. cleaved when b-lactamase escapes into the
Mpeg1 (figs. S4C and S5C).

Next, we addressed whether the pore-forming
ability of perforin-2 is required for endocytic
escape. Perforin-2 forms pores by oligomeriza-
tion and unwinding of two helices in the
MACPF domain into b sheets (shown in red in
Fig. 1D) (24, 25). To test whether this conforma-
tional change is required for endocytic escape,
we generated two mutants: Mpeg1G212V/A213V,
with mutations in the conserved MACPF motif
(31), and Mpeg1K251C/G286C, with a disulfide bond
known to constrain one of the pore-forming
helices preventing pore formation in vitro (24).
Both Mpeg1G212V/A213V and Mpeg1K251C/G286C
were expressed at levels similar to those of
Mpeg1WT, but neither rescued saporin escape
in Mpeg1KO MutuDCs (Fig. 1F and fig. S4C).
Perforin-2 is sufficient for endocytic escape in
nonimmune cells
Because perforin-2 expression is restricted
to antigen-presenting cells (32, 33), we asked
whether it is sufficient for endocytic escape in
nonimmune cells. We generated human em-
bryonic kidney 293T (HEK293T) and HeLa
cells coexpressing murine Mpeg1 with mScarlet
or BFP, or expressing the fluorescent protein
alone, and used them in a range of escape
assays.
Ectopic expression of perforin-2 was suffi-
cient to promote saporin escape in HEK293Ts
(Fig. 2A). We also monitored escape of 34-kDa
b-lactamase using a cytosolic dye CCF4 (34, 19).
CCF4 consists of fluorescein and 7-hydroxycou-
,
cytosol, resulting in a shift in fluorescence
emission (Fig. 2B). Expression of perforin-2
in HeLa cells increased the frequency of cells
with cleaved CCF4 and thus the efficiency of
b-lactamase escape (Fig. 2B and fig. S6). We
then adopted a split luciferase-based assay
to monitor endocytic escape of microtubule-
associated protein, tau (35). We expressed the
large 18-kDa NanoLuc subunit (LgBiT) in the
cytosol of HEK293Ts and pulsed the cells with
oligomers of 42-kDa tau fused to the short
HiBiT peptide. Upon entry into the cytosol,
LgBiT binds HiBiT, resulting in catalytically
active luciferase and luminescence in the pres-
ence of the Nano-Glo(R) substrate. Again, in
perforin-2–expressing HEK293Ts, tau-HiBit
escaped more efficiently compared with the

Rodríguez-Silvestre et al., Science 380, 1258–1265 (2023) 23 June 2023 2 of 8

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

Fig. 2. Perforin-2 is sufficient for endocytic escape of cargo in
non-immune cells. (A) HEK293Ts and Mpeg1-complemented HEK293Ts were
pulsed with saporin (11:1 unlabeled:Atto550-labeled saporin) for 2 hours, and
translation was monitored by a 30-min puromycin chase. The x axis represents
Atto550 MFI normalized to the WT cells pulsed with 0.5 mg/ml saporin.
p
g
significant; ns, not significant; **P < 0.01; ***P < 0.001 using a multiple t test
(two-stage step-up, Benjamini, Krieger, and Yekutieli) For gating strategy, see
fig. S6. (C) To monitor the escape of tau oligomers, cells expressing NLS-eGFP-
LargeBiT were pulsed with Tau-HiBit oligomers. The escape of Tau-HiBiT into the
cytoplasm allows binding to the 18-kDA luciferase subunit, LgBiT. This results

Data represent mean and SEM of three independent experiments; ns, not
significant; ****P < 0.0001 using a multiple t test (Bonferroni-Dunn).
Significance symbols on the plot refer to the differences in cells in translation
arrest. Differences in saporin Atto550 MFI were not significant. (B) Cytosolic
escape of b-lactamase in cells loaded with the CCF4 results in CCF4 cleavage,
loss of fluorescence resonance energy transfer, and shift in emission
fluorescence. HeLa cells expressing either Mpeg1IRES-mScarlet or mScarlet only
were pulsed with b-lactamase for the indicated time. b-lactamase escape was
monitored by measuring the shift in fluorescence emission by flow cytometry.
Data represent mean and SEM of three independent experiments, ns, not
y
in reconstitution of catalytic activity and generation of luminescence. NLS-eGFP-
LargeBiT HEK293Ts expressing either Mpeg1IRES-BFP or BFP only were pulsed
with tau-HiBiT for the indicated time. Following substrate addition, luminescence
and cell viability were assessed. Relative luminescent units (RLUs) were then
normalized to viability per well. Data represent mean and SEM of three independent
experiments each with six technical replicates, ***P < 0.001 using a paired t test.
(D) HEK293Ts were plated in the presence or absence of bleomycin and cultured
in an Incucyte for 48 hours to monitor the growth rate. Data represent mean and
SEM of three independent experiments each with four wells per condition; ns, not
significant; *P < 0.5; **P < 0.01; ****P < 0.0001 using a multiple t test (Bonferroni-Dunn).

control line (Fig. 2C). Finally, we demonstrated
that expression of perforin-2 renders HeLa
cells more sensitive to bleomycin-mediated
toxicity (Fig. 2D). Thus, ectopic expression of
proteolytically released to facilitate endocytic
escape.
To investigate the proteolytic processing of
perforin-2, we analyzed a SILAC (stable isotope
y g
sistent with the hypothesis that the perforin-2
ectodomain is released from the TMD anchor.
The remaining perforin-2 peptides, p340-372
[within the epidermal growth factor (EGF)–

perforin-2 is sufficient to drive endocytic es-
cape in nonimmune cells.
Perforin-2 is proteolytically processed
in lysosomes
Given the cytotoxic potential of pore-forming
proteins, we asked how cDC1s regulate pore
formation and restrict it to antigen-containing
compartments (19, 35). Perforin-2 is the only
known mammalian pore-forming protein with
an additional transmembrane domain (TMD)
that is not involved in pore formation (Fig.
1D). The TMD has been proposed to act as
an anchor preventing damage to endogenous
membranes and orientating the pore toward
intravacuolar bacteria (24, 27). We hypothe-
sized that the ectodomain would have to be
labeling by amino acids in cell culture)–based
organellar mapping dataset generated previ-
ously (19). The maps were prepared by mass
spectrometry–based analysis of the fractionated
postnuclear supernatants from MutuDCs (fig.
S7A) (36). In the original maps, perforin-2 had
a profile similar to those of lysosomal proteins.
Here, instead of analyzing protein profiles, we
analyzed each tryptic peptide individually, only
including endo- or lysosomal proteins (Fig. 3, A
and B). Peptides derived from endosome- and
lysosome-resident proteins formed separate
clusters, as did peptides derived from trans-
membrane and luminal proteins in the lyso-
some cluster. Fifteen out of 17 perforin-2–
derived peptides clustered with soluble rather
than transmembrane lysosomal proteins, con-
,
like domain] and p629-635 (within the TMD-
proximal region), coclustered with endosomal
proteins (such as Vps35, Fig. 3A), suggesting
that they are present in the full-length protein
as it transits through endosomes but are absent
(cleaved) once perforin-2 reaches lysosomes.
We confirmed that perforin-2 was proteo-
lytically processed using antibodies against
the MACPF, P2, and the C-terminal tail (C-term)
(fig. S7, B and C). All three antibodies detected
full-length perforin-2 at 70 kDa. We also iden-
tified a 40-kDa MACPF and a 30-kDa P2 frag-
ments, consistent with the cleavage in the EGF
domain. The aC-term antibody detected only
the full-length protein, indicating that the tail
(and most likely the TMD) are rapidly degraded
after release of the ectodomain. Bafilomycin

Rodríguez-Silvestre et al., Science 380, 1258–1265 (2023) 23 June 2023 3 of 8

RES EARCH | R E S E A R C H A R T I C L E
Fig. 3. Perforin-2 undergoes proteolytic cleavage, releasing its pore-
forming domain into the organellar lumen. (A and B) (A) Principal-component
analysis of mass spectrometry–based organellar mapping of MutuDCs (19)
(see also fig. S7A). The maps were prepared from control MutuDCs and cells
treated with drugs that promote lysosomal leakiness, prazosin and tamoxifen.
Peptides derived from lysosomal and endosomal proteins are represented by
filled and empty circles, respectively. The different colors indicate localization
of the protein within the corresponding organelle. Perforin-2 peptides are
displayed as filled black circles regardless of their localization. (B) Mapping of the
different lysosomal and endosomal perforin-2 peptides detected by organellar
mass spectrometry onto the different perforin-2 domains and structure.
(C) Perforin-2 levels in NT MutuDCs treated with 0.5 mM BafA1 for 3 hours
were assessed by immunoblot under reducing conditions using the aMACPF
and aC-terminal tail antibodies. (D) Confocal microscopy images of MutuDCs
Rodríguez-Silvestre et al., Science 380, 1258–1265 (2023) 23 June 2023
p
g
y
y g
,
stained for perforin-2 with either aC-terminal tail or aMACPF antibodies (red),
Vps35 (green), and lysotracker (blue). Data represent two independent
experiments each with at least 80 cells, ***P < 0.0001 using a Kolmogorov-
Smirnov test. (E) BafA1- and CpG-induced changes in the abundance of
tryptic and semi-tryptic (cleaved) perforin-2 peptides. Control and treated
cells (1 mM BafA1 or 1 mM CpG) were analyzed by mass spectrometry, and
peptide intensities were normalized to the corresponding protein intensities.
Statistical analysis was performed with a two-sided student’s t test. P values
(y axis) and fold change in abundance (line thickness) in treated versus
control cells are shown. The amino acid position indicates the location of the
peptides along the different perforin-2 domains. (F) Differences in the
abundance of tryptic and semi-tryptic (cleaved) perforin-2 peptides
between control and AEPKO MutuDCs (fig. S9B) by full proteome mass
spectrometry. The analysis was performed as in (E).
4 of 8
RES EARCH | R E S E A R C H A R T I C L E

A1, a vacuolar-type ATPase (V-ATPase) inhib-

itor that interferes with lysosomal acidifi-
cation, accumulated full-length perforin-2,
consistent with the hypothesis that proteo-
lytic processing occurs in lysosomes (Fig. 3C).
Finally, confocal microscopy confirmed that
the aC-term antibody, which we predicted to
recognize the immature (full-length) perforin-2,
colocalized with Vps35 (Fig. 3D), and the
aMACPF antibody colocalized with lyso-
tracker, but not with Vps35. This result,
together with the organellar mapping data,
suggests that most of the protein is in lysosomes
at steady-state. Thus, full-length perforin-2
resides in (or transits through) endosomes,
and upon reaching low pH compartments,
undergoes maturation involving at least two
cleavage events.

Perforin-2 maturation is controlled by

asparagine endopeptidase (AEP)

p
We then asked whether perforin-2 maturation
might be regulated by antigen-associated sig-
nals. From a panel of TLR agonists, CpG (TLR9-
agonist) and Toxoplasma profilin (TLR11-agonist)
were most efficient in promoting perforin-2
proteolytic processing (fig. S8). To further

g
characterize the putative cleavage sites, we
performed comparative proteomics analysis of
tryptic and semi-tryptic peptides from CpG-,
BafA1-, and mock-treated cells (Fig. 3E). Semi-
tryptic peptides are considered to be derived

y
from proteins that were cleaved in the cell,
before sample processing. The semi-tryptic
peptide p340-349 (within the EGF domain)
was enriched in CpG-treated cells and depleted
in BafA1-treated cells, whereas tryptic peptides
near the TMD, p621-628 and p629-637, were
depleted in CpG-treated cells and enriched in
BafA1-treated cells.
Several of the nontryptic peptides in the
EGF region terminated on asparagine (fig.

y g
S9A), suggesting that perforin-2 processing
might be mediated by AEP (37). Indeed, the
cleavage pattern in the EGF domain was dif-
ferent in AEPKO MutuDCs (fig. S9B) compared
with control cells (see peptides p340-358 and

,
p340-351, Fig. 3F), suggesting that AEP medi-
ates perforin-2 maturation, but its activity
can be replaced by other enzymes [most likely
cathepsins (38)]. Immunoblot analysis con-
firmed that the EGF cleavage is less efficient
(albeit not completely abolished) in the absence
of AEP, resulting in accumulation of the 60-kDa
ectodomain (fig. S9C). We were also able to
Fig. 4. Perforin-2 undergoes pH-dependent maturation in antigen-containing phagosomes. reconstitute this cleavage in vitro using re-
(A) Schematic representation of the phagoFACS assay. Cells are pulsed with OVA-beads and allowed to combinant perforin-2 and AEP (fig. S9D).
internalize them. After an indicated chase period, noninternalized beads are labeled with an aOvalbumin In summary, perforin-2 maturation is con-
antibody. After cell homogenization, phagosomes are stained with antibodies against OVA coupled to trolled by at least two cleavage events, both oc-
an alternative fluorophore and phagosomal markers. (B) Mpeg1KO and NT MutuDCs were pulsed with OVA- curring at steady-state but stimulated by TLR
beads and chased for the indicated time in the presence of either BFA or BafA1. Isolated phagosomes signaling. The cleavage in the TMD-proximal
were stained with antibodies against Lamp1 and either the perforin-2 C-terminal tail (top panel) or MACPF CTT domain releases the ectodomain into the
domain (bottom panel). Data are representative of three independent experiments. See also fig. S11 for lysosomal lumen to orient the pore-forming
gating strategy, quantification, and additional plots. domain toward the endogenous membranes.

Rodríguez-Silvestre et al., Science 380, 1258–1265 (2023) 23 June 2023 5 of 8

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

Fig. 5. Antigen cross-priming is

impaired in vivo in the absence
of perforin-2. (A) Perforin-2
expression was assessed by intra-
cellular staining with an aPerforin-2
antibody and flow cytometry in
Mpeg1+/+ and Mpeg1−/− spleno-
cytes. cDC1s are defined as Lineage
(CD3, CD19, NK1.1)−, F4/80−,
CD11c, XCR1+; cDC2s as Lineage
(CD3, CD19, NK1.1)−, F4/80−,
CD11c, CD172a+; and pDCs as
Lineage (CD3, CD19, NK1.1)−,
F4/80-, CD11cintSiglecH+. For gat-
ing strategies, see fig. S13C.
(B) CD11c+ magnetically enriched
splenocytes from WT and Mpeg1−/−
mice were pulsed with saporin
for 2 hours, and translation was
monitored by a 30-min puromycin
chase. cDC1s are defined as CD11c+,

p
XCR1+, cDC2s are defined as
CD11c+, CD172a+, and pDCs are
defined as CD11cintSiglecH+. Dot
plots are representative of two
independent experiments. For gat-
ing strategies, see fig. S14E. (C and
D) Wild-type and Mpeg1−/− mice

g
were intravenously (i.v.) injected
with 0.5 × 106 cell trace violet
(CTV)–labeled magnetically purified
OT-I cells. One day later, mice were

y
injected i.v. with 1 × 106 UVC-
irradiated (240 mJ/cm2) 3T3 cells,
coated with 10 mg/ml ovalbumin
as antigen source and 0.25 mg/ml
Poly(I:C) as an adjuvant. Three
days later, OT-I proliferation was
assessed by flow cytometry (C).
OT-I are defined as Lineage (CD19,
F4/80, CD11c)−, CD3+, CD4−CD8+,
TCRvb5.1, 5.2+TCRva2+, CTV+. For

y g
gating strategy, see fig. S15A.
(D) Normalized OT-I counts 3 days
after intravenous antigen injection.
Each dot corresponds to an indi-
vidual mouse, with three to five

,
mice per group. For each experi-
ment, OT-I counts per 1 × 106 splenocytes were normalized to the average of WT controls. Data represent five independent experiments.
ns, not significant; **P < 0.01 using an unpaired t test.

The AEP-mediated cleavage in the EGF region,
which is not required for pore formation in vitro
(24, 39), may provide additional flexibility to
complete pore insertion in vivo or may serve
to inactivate the pores.
Perforin-2 undergoes maturation in
antigen-containing compartments
To test whether perforin-2 undergoes matura-
tion upon recruitment to antigen-containing
compartments or whether antigen-containing
compartments acquire mature perforin-2, we
analyzed perforin-2 processing in phagosomes.
We first confirmed that perforin-2 mediates
the escape of bead-conjugated saporin from
phagosomes into the cytosol (fig. S10). We then
followed perforin-2 maturation in individual
phagosomes by phagoFACS, a technique in which
DCs are pulsed with ovalbumin (OVA)–coated
beads and the resulting phagosomes are an-
alyzed by flow cytometry (Fig. 4A). Using
the aC-term antibody, we demonstrated that
perforin-2 was rapidly recruited to phagosomes,
reaching its highest levels within 30 min (Fig. 4B
and fig. S11, A and B). This rapid acquisition
suggests that perforin-2 is recruited before
phagosome-lysosome fusion (40). Indeed, bre-
feldin A (BFA), an inhibitor of ARF GTPases
that blocks protein trafficking through the
early secretory pathway, inhibited perforin-2
(but not Lamp1) recruitment, consistent with
the delivery of full-length perforin-2 to phago-
somes from the Golgi (Fig. 4B and fig. S11, B

Rodríguez-Silvestre et al., Science 380, 1258–1265 (2023) 23 June 2023 6 of 8

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

and C). In phagoFACS experiments with the
aMACPF antibody, perforin-2 acquisition was
also BFA dependent. However, the aMACPF
staining was restricted to Lamp1+ phagosomes
and increased over time, whereas the aC-term
signal was gradually lost (Fig. 4B and fig. S11, A
and B). Thus, during phagosome maturation,
the C-terminal part of perforin-2 is cleaved off
and degraded, whereas the ectodomain under-
goes a conformational change that exposes an
epitope recognized by the aMACPF antibody.
To confirm that perforin-2 maturation in
phagosomes is pH dependent, we first asked
whether MutuDC phagosomes acidify given
the conflicting reports on whether the pH in
DC phagosomes remains high (41–43) or de-
creases (44–46) over time. Using pHrodo-beads,
we found that phagosome acidification in
MutuDCs was abrupt and occurred on a time-
scale comparable to that of Lamp1 acquisition
(fig. S12). Inhibition of phagosomal acidifica-
subpopulation of cDC2s expressed perforin-2,
whereas in the lungs, a large fraction of cDC2s
was perforin-2 positive (fig. S14, B and C).
Consistent with perforin-2 expression patterns,
knocking out Mpeg1 decreased the efficiency
of endocytic escape in splenic cDC1s and pDCs,
but not in cDC2s (Fig. 5B).
We also asked whether perforin-2–mediated
escape delivers antigens for cross-presentation.
To assess the efficiency of antigen presentation
in vivo, we adoptively transferred OT-I T cells
(expressing a T cell receptor specific to H2-
Kb MHC-I with OVA257-264 peptide) and mo-
nitored OT-I proliferation after immunization.
Because cross-presentation of soluble OVA,
which can be processed by cell types other than
cDC1s (48, 50), was not significantly affected in
the Mpeg1−/− mice (fig. S15), we used dead cells
as an antigen source. Compared to wild-type
(WT) mice, Mpeg1−/− had fewer OT-I T cells
after immunization with ultraviolet C (UVC)–
the initiation of immune responses. The ability
to genetically manipulate perforin-2–mediated
endocytic escape provides a tool for exploring
the contribution of the escape pathway to
anticancer and antiviral immunity in vivo.
REFERENCES AND NOTES
1. M. J. Palmowski et al., J. Immunol. 177, 983–990 (2006).
2. J. Helft et al., J. Clin. Invest. 122, 4037–4047 (2012).
3. D. J. Theisen et al., Science 362, 694–699 (2018).
4. E. Giampazolias et al., Cell 184, 4016–4031.e22 (2021).
5. M. Kovacsovics-Bankowski, K. L. Rock, Science 267, 243–246 (1995).
6. J. Canton et al., Nat. Immunol. 22, 140–153 (2021).
7. M. Zehner et al., Immunity 42, 850–863 (2015).
8. M. Zehner et al., Proc. Natl. Acad. Sci. U.S.A. 108, 9933–9938
(2011).
9. I. Dingjan et al., Sci. Rep. 6, 22064 (2016).
10. S. C. Nalle et al., PLOS ONE 15, e0238484 (2020).
11. J. E. Grotzke et al., Proc. Natl. Acad. Sci. U.S.A. 114,
E5910–E5919 (2017).
12. Y. Ding et al., Nat. Immunol. 17, 1167–1175 (2016).
13. J. Canton et al., Nat. Immunol. 22, 140–153 (2021).
14. S. Zelenay et al., J. Clin. Invest. 122, 1615–1627 (2012).
15. K. Nielsen, R. S. Boston, Annu. Rev. Plant Physiol. Plant Mol. Biol.
52, 785–816 (2001).

tion with BafA1 interfered with both gradual
loss of the aC-term signal and acquisition of
the aMACPF signal (Fig. 4B and fig. S11B),
confirming that perforin-2 maturation is
regulated by pH.
The pHrodo experiments also revealed that
phagosomal pH was similar in NT and Mpeg1KO
p
irradiated, OVA-coated fibroblasts (Fig. 5C). 16. E. K. Schmidt, G. Clavarino, M. Ceppi, P. Pierre, Nat. Methods
Similarly, Mpeg1−/− bone marrow-derived cDC1s 6, 275–277 (2009).
17. K. Hildner et al., Science 322, 1097–1100 (2008).
[from FMS-like tyrosine kinase 3 ligand (Flt3 L) 18. M. L. Lin et al., Proc. Natl. Acad. Sci. U.S.A. 105, 3029–3034 (2008).
and granulocyte-macrophage colony-stimulating 19. P. Kozik et al., Cell Rep. 32, 107905 (2020).
factor (GM-CSF) cultures] displayed impaired 20. S. A. Fuertes Marraco et al., Front. Immunol. 3, 331 (2012).
21. M. Gros et al., Cell Rep. 40, 111205 (2022).
endocytic escape in the saporin assay and a 22. A. J. Monteith et al., Proc. Natl. Acad. Sci. U.S.A. 113,

MutuDC (fig. S12), suggesting that perforin-2
pores are either not permissive to protons, pore
formation is transient, or the cells can compen-
sate for proton loss. Furthermore, these data
reduced capacity to cross-present cell-associated
antigens (fig. S16). Thus, loss of perforin-2 leads 23.
to a defect in cross-presentation of cell-associated 24.
antigens in vitro and in vivo, suggesting that in 25.
cross-presenting cells, endocytic pores provide a 26.
g
E2142–E2151 (2016).
R. McCormack, L. R. de Armas, M. Shiratsuchi, J. E. Ramos,
E. R. Podack, J. Innate Immun. 5, 185–194 (2013).
T. Ni et al., Sci. Adv. 6, eaax8286 (2020).
S. S. Pang et al., Nat. Commun. 10, 4288 (2019).
X. Yu, T. Ni, G. Munson, P. Zhang, R. J. C. Gilbert, EMBO J. 41,

suggest that membrane integrity of phagosomes
is not compromised during perforin-2–mediated
escape, in line with the observation that phago-
somes containing OVA-beads do not recruit
galectin-3, a marker of damaged compartments
(6). Consistent with these data, OVA degra-
dation was not affected by knocking out Mpeg1,
confirming that perforin-2 does not drastically
alter the degradative potential of phagosomes
(fig. S11, D and E).
y
e111857 (2022).
route for cytosolic entry of antigens. 27. R. M. McCormack et al., eLife 4, e06508 (2015).
28. S. Ebrahimnezhaddarzi et al., Immunol. Cell Biol. 100, 529–546
Conclusions (2022).
29. M. J. Jaroszeski et al., Anticancer Drugs 11, 201–208 (2000).
Here, we uncovered a mechanism of endocytic
30. S. A. Fuertes Marraco et al., PLOS ONE 6, e20189 (2011).
escape that is governed by a cell type–specific 31. C. J. Rosado et al., Science 317, 1548–1551 (2007).
pore-forming effector protein, perforin-2. The 32. T. S. Heng et al., Nat. Immunol. 9, 1091–1094 (2008).
role of perforin-2 in the delivery of antigens for 33. J. Lonsdale et al., Nat. Genet. 45, 580–585 (2013).
34. I. Cebrian et al., Cell 147, 1355–1368 (2011).
cross-presentation suggests that the immune 35. B. J. Tuck et al., Cell Rep. 39, 110776 (2022).
system evolved to use two related pore-forming 36. D. N. Itzhak, S. Tyanova, J. Cox, G. H. Borner, eLife 5, e16950
effectors, perforin-1 and perforin-2, during diff- (2016).

Thus, perforin-2 undergoes pH-dependent
maturation in antigen-containing compartments,
and perforin-2–mediated endocytic escape of
antigens can occur while preserving the overall
integrity of the phagosomal membrane.
y g
37. J.-M. Chen, P. M. Dando, R. A. E. Stevens, M. Fortunato,
erent stages of adaptive immune responses. A. J. Barrett, Biochem. J. 335, 111–117 (1998).
Perforin-1, expressed by cytotoxic T cells, is a 38. S. E. Ewald et al., J. Exp. Med. 208, 643–651 (2011).
well-characterized effector used for the deliv- 39. F. Jiao et al., Nat. Commun. 13, 5039 (2022).
40. F. M. Cruz, J. D. Colbert, K. L. Rock, EMBO J. 39, e102020 (2020).
ery of granzymes into the cytosol of target cells 41. A. Savina et al., Immunity 30, 544–555 (2009).
(51). Our data suggest that perforin-2 delivers 42. A. R. Mantegazza et al., Blood 112, 4712–4722 (2008).

Perforin-2 is expressed in antigen-presenting
cells and facilitates cross-presentation
To test whether perforin-2 is involved in the
delivery of antigens for cross-presentation,
we generated Mpeg1−/− mice (fig. S13A and B).
Notably, knocking out Mpeg1 did not result in
any obvious disease phenotype or a change in
immune cell frequencies (fig. S13, C and D).
We used intracellular staining to confirm
that perforin-2 expression is restricted to splenic
endocytic contents into the cytosol of cross-
presenting DCs to enable generation of MHC-
I:peptide complexes and T cell priming.
We do not exclude the possibility that in
different contexts, membrane destabilization
may also result in antigen delivery for cross-
presentation (6, 9, 52). Membrane integrity is
regulated by a wide range of mechanisms
(53, 54), and different perturbations can result
in leakage of endosomal contents into the
cytosol under pathological conditions. Perforin
,
43. A. Savina et al., Cell 126, 205–218 (2006).
44. S. C. Khouili et al., Cell Rep. 33, 108468 (2020).
45. J. M. Rybicka, D. R. Balce, S. Chaudhuri, E. R. O. Allan,
R. M. Yates, EMBO J. 31, 932–944 (2012).
46. M. Samie, P. Cresswell, Nat. Immunol. 16, 729–736 (2015).
47. C. Bosteels et al., Immunity 52, 1039–1056.e9 (2020).
48. M. Enders et al., J. Immunol. 204, 87–100 (2020).
49. M. Oberkampf et al., Nat. Commun. 9, 2241 (2018).
50. N. M. Kretzer et al., J. Exp. Med. 213, 2871–2883 (2016).
51. I. Voskoboinik, J. C. Whisstock, J. A. Trapani, Nat. Rev.
Immunol. 15, 388–400 (2015).
52. P. Cresswell, Immunogenetics 71, 141–160 (2019).
53. M. L. Skowyra, P. H. Schlesinger, T. V. Naismith, P. I. Hanson,
Science 360, eaar5078 (2018).

cDC1s (Lin−CD11c+XCR1+), as well as other cell 2–mediated endocytic escape, however, appears 54. I. Maejima et al., EMBO J. 32, 2336–2347 (2013).

types previously shown to cross-present, includ-
ing splenic macrophages (Lin−F4/80+), plasma-
cytoid DC (pDCs, Lin−F4/80−CD11cintSiglecH+),
and Ly6C+ monocytes (Fig. 5A and fig. S14A)
(47–50). In the spleen, only a small, CX3CR1+
to occur without compromising the overall
stability of the endocytic compartments.
The restricted expression of Mpeg1 to sub-
sets of professional antigen-presenting cells
points to an important role of perforin-2 during
AC KNOWLED GME NTS
We thank members of the Kozik laboratory for discussions and
feedback regarding this work; we also thank G. M. Griffiths and
L. C. James for reading of the manuscript and their feedback. We are
indebted to N. Hacohen and T. Eisenhauer for their help setting up
CRISPR-Cas9 screens and G. Slodkowicz for help with statistical

Rodríguez-Silvestre et al., Science 380, 1258–1265 (2023) 23 June 2023 7 of 8

RES EARCH | R E S E A R C H A R T I C L E
analysis. We also thank A. Crisp for help with data processing.
We are grateful to M. Mann for his support and I. Paron and
T. Heymann at the Max Planck Institute of Biochemistry for technical
assistance with mass spectrometry. Finally, we thank the animal
facility/ARES staff, genotyping facility, flow cytometry core, and
microscopy core at the MRC Laboratory of Molecular Biology for
their technical assistance. Funding: This work was supported
by the UK Medical Research Council (MRC grant no. MC_UP_1201/
26). P.R.-S. was supported by an MRC CellTech Research
Fellowship (MRF-104-0007-S-RODRI). P.A.K. was supported by the
Boehringer Ingelheim Fonds PhD Fellowship. G.H.H.B., A.K.D., and
J.P.S. were funded by the Max Planck Society for the Advancement
of Science. B.J.T. was supported by the Cambridge Trust Vice
Chancellor Award and Hughes Hall Edwin Leong PhD scholarship.
W.A.M. is a Lister Institute Fellow and supported by a Sir Henry
Dale Fellowship jointly funded by the Wellcome Trust and the Royal
Society (206248/Z/17/Z). W.A.M. was further supported by the
Rodríguez-Silvestre et al., Science 380, 1258–1265 (2023)
UK Dementia Research Institute, which receives its funding from reserved; exclusive licensee American Association for the
DRI Ltd, funded by the UK Medical Research Council, Alzheimer’s Advancement of Science. No claim to original US government
Society, and Alzheimer’s Research UK. Author contributions: works. https://www.sciencemag.org/about/science-licenses-
P.R.S. designed and performed experiments and wrote the manuscript. journal-article-reuse
M.L. designed and performed experiments and contributed to the
manuscript. P.A.K. performed experiments and provided scientific SUPPLEMENTARY MATERIALS
insight. A.K.D., J.P.S., and G.H.B. carried out mass spectrometry
science.org/doi/10.1126/science.adg8802
experiments and proteomics data analysis. R.P. performed
Materials and Methods
experiments. B.J.T. and W.A.M. carried out Tau-HiBit assay. P.K.
Figs. S1 to S16
supervised the study, designed and performed experiments, and
Tables S1 to S5
wrote the manuscript. Correspondence and requests for materials
References (55–61)
should be addressed to Patrycja Kozik (pkozik@mrc-lmb.cam.ac.uk).
MDAR Reproducibility Checklist
Competing interests: The authors declare no competing interests.
Data and materials availability: All data are available in the main View/request a protocol for this paper from Bio-protocol.
text or the supplementary materials. Data are available via
ProteomeXchange with identifier PXD041861. License Submitted 27 January 2023; accepted 3 May 2023
information: Copyright © 2023 the authors, some rights 10.1126/science.adg8802
p
g
y
y g
,
23 June 2023 8 of 8
RES EARCH

QUANTUM INFORMATION tion applied—much faster than the timescale

over which the data and spectator qubits de-
Mid-circuit correction of correlated phase errors correlate. This second requirement has limited
the experimental implementation of such pro-
using an array of spectator qubits tocols because a substantial number of mea-
surements are required to reliably estimate
K. Singh1†, C. E. Bradley2†, S. Anand2†, V. Ramesh3, R. White3, H. Bernien2* the effects of a dynamic noise environment.
Furthermore, the spectator qubit readouts must
Scaling up invariably error-prone quantum processors is a formidable challenge. Although quantum error be performed mid-circuit without perturbing
correction ultimately promises fault-tolerant operation, the required qubit overhead and error thresholds are the data qubits.
daunting. In a complementary proposal, colocated, auxiliary “spectator” qubits act as in situ probes of In this work, we overcome these challenges
noise and enable real-time, coherent corrections of data qubit errors. We used an array of cesium spectator and demonstrate real-time correction of cor-
qubits to correct correlated phase errors on an array of rubidium data qubits. By combining in-sequence related phase errors using a dual-species array
readout, data processing, and feedforward operations, these correlated errors were suppressed within of individually trapped neutral atoms. The
the execution of the quantum circuit. The protocol is broadly applicable to quantum information platforms protocol is outlined in Fig. 1A. Data qubits
and establishes key tools for scaling neutral-atom quantum processors: mid-circuit readout of atom (rubidium atoms) and spectator qubits (cesium
arrays, real-time processing and feedforward, and coherent mid-circuit reloading of atomic qubits. atoms) are laser-cooled into optical tweezer
arrays (26). During logic operations on the

R
data qubits, mid-circuit readouts (MCRs) on
ealizing large-scale programmable quan- susceptible to the same noise sources. Specta- the array of ~60 spectator qubits enable single-
tum systems that can overcome inevita- tor qubits act as in situ probes of that noise, shot estimation of globally correlated phase

ble noise sources is a central challenge
for modern physics (1, 2). Environmental
noise and experimental parameter drift
necessitate strategies to reduce their impact
and overcome resulting qubit errors. Although
quantum error correction will ultimately be
such that measurement and feedforward can
be used to coherently protect the data qubits
during the execution of a quantum algorithm
(23–25). Notably, under two key conditions,
spectator protocols are agnostic to the spec-
trum and correlation time of the noise source.
p
errors. The readout results are processed in
real time and used to infer the noise-induced
phase accrued by the ~60 data qubits. Crucially,
owing to the cross-talk–free operation of the
two species, these readouts do not disturb the
coherence of the data qubits. We leverage a

required, achieving the necessary qubit opera-
tion fidelities is an outstanding challenge for
present quantum computing platforms (3–9).
Moreover, the effectiveness of error-correction
codes is reduced by correlated errors (10, 11),
First, the noise-induced dynamics must be
correlated between the spectator and data
qubits. Second, an estimate of those dynam-
ics must be made by reading out the spectator
qubits—and a subsequent feedforward opera-
g
classical control architecture to perform in-
sequence feedforward, such that correlated
errors on the data qubits are mitigated within
the execution of the quantum circuit. Finally,
we show that the spectator qubits can be

y
which may naturally occur when the qubits
are in close spatial proximity or are controlled
by shared hardware (12–16).
To address these challenges, a number of
techniques have been developed to mitigate
the effects of noise, such as composite pulses
(17), optimal control (18), dynamical decou-
pling (17, 19), Hamiltonian learning (20), and
machine learning–based control engineering
(21). These techniques have found great suc-

y g
cess, but they are typically tailored to specific
noise models or require careful calibration and
thus face challenges when used in realistic, fluc-
tuating environments. For example, dynamical
decoupling generates a filter function that miti-

,
gates a particular spectrum of noise, with pass-
bands remaining that are not suppressed (22).
Additionally, it is only effective if the correla-
tion time of the noise is long with respect to the
interpulse delay.
Recent theoretical work has proposed a com-
plementary technique based on “spectator”
qubits—additional qubits that are colocated
with the computational “data” qubits and are

1
Intelligence Community Postdoctoral Research Fellowship
Program, Pritzker School of Molecular Engineering,
University of Chicago, Chicago, IL 60637, USA. 2Pritzker
School of Molecular Engineering, University of Chicago,
Chicago, IL 60637, USA. 3Department of Physics, University
of Chicago, Chicago, IL 60637, USA.
*Corresponding author. Email: bernien@uchicago.edu
†These authors contributed equally to this work.
Fig. 1. Spectator qubit protocol with a dual-species atom array. (A) Feedforward loop for real-time correction
of correlated phase errors between data qubits (Rb atoms, blue) and spectator qubits (Cs atoms, yellow). A mid-circuit,
single-shot phase estimation on the spectators is used to infer the noise-induced phase accrued by the data qubits.
This information enables a real-time correction on the data qubits before the final readout, which suppresses
dephasing. Subsequently, spectator qubits lost during readout can be replenished while maintaining data qubit
coherence. (B) Example fluorescence image of the dual-species atom array. Scale bar indicates ~10 mm. (C) Microwave
Rabi oscillations of the data and spectator qubits. Dashed lines are fits to exponentially decaying sinusoids.

Singh et al., Science 380, 1265–1269 (2023) 26 June 2023 1 of 5

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

replenished within the data qubit coherence

time, an essential step toward repeated mea-
surements and the continuous operation of
atom-based quantum processors.

MCR of spectator qubits

Our experiment is performed on arrays of 10-
by-10 and 11-by-11 sites for the spectator and
data qubits, respectively (Fig. 1B), which are
stochastically loaded with an average loading
fraction of ~55%. The experimental apparatus
has been upgraded from our previous work (26)
to incorporate qubit initialization, manipulation,
and readout, along with classical hardware to
implement real-time processing and feedfor-
ward. Here, the qubits are encoded into long-
lived hyperfine states ( jF ¼ 1; mF ¼ 0i :¼
j0iRb and jF ¼ 2; mF ¼ 0i :¼ j1iRb for Rb;
jF ¼ 3; mF ¼ 0i :¼ j0iCs and jF ¼ 4; mF ¼
0i :¼ j1iCs for Cs, where F is the total angular
momentum and mF is the magnetic quantum

p
number). Microwave driving of the data and
spectator qubits after optical pumping into
j1iRb andj1iCs reveals coherent Rabi oscillations
(Fig. 1C).
An essential ingredient for the spectator
protocol is to perform MCR of the spectator

g
qubits without inducing additional data qubit
decoherence. This is challenging in single-
species atom arrays because all atoms are
resonant with the excitation laser and the
measured qubits scatter light, which can de-

cohere the data qubits through reabsorption.
To overcome this, several ideas have been pro-
posed and demonstrated, including coherently
transporting qubits into readout cavities (27)
or using additional shelving states to hide
atoms from excitations from the readout light,
as demonstrated for trapped ions (4). How-
ever, realizing cross-talk–free imaging in large
atom arrays has remained an outstanding
challenge. A key motivation behind the dual-
y
Fig. 2. MCR of atomic qubits. (A) Pulse diagram depicting MCR of atomic qubits. Lowercase and uppercase
letters indicate p/2 and p pulses, respectively, along that axis of rotation. The readout light is left on for the remaining
duration of the sequence after MCR. (B) Measurement of spectator qubit dynamics while preserving data qubit
coherence. During an XY8 decoupling sequence on the data qubits (red diamonds), we performed an XY4
decoupling sequence and subsequent projective measurement on the spectator qubits. The data qubit coherence
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
( hsx i2 þ hsy i2 ) is unchanged in the absence of MCR (blue circles). Dashed lines are fits (29). The inset shows
coherence measurements for early (square) and late (triangle) evolution times. (C) Example fluorescence histogram

y g
of a spectator qubit. Solid lines are fits to a bimodal Poisson distribution. (D) Cumulative histogram of the
species approach is that the different atomic discrimination infidelities of the spectator qubits during MCR (29). eCDF, empirical cumulative distribution function.
species have distinct optical transitions, and
(E) Coherence of spectator qubits. The measured spectator coherence time is T2XY4 = 136(7) ms.
measurements on one species are not expected
to influence the other (26, 28).
In a first experiment, we characterized the
[fitted T2;MCR
XY8
= 0.68(1) s, T2;No
XY8

spectator qubit MCR and measured its impact
on the data qubit coherence. The quantum
circuit is shown in Fig. 2A. During an XY8
decoupling sequence on the data qubits, an
XY4 sequence was performed on the spectators.
The spectator qubits were measured within
the XY8 sequence by selectively removing all
atoms in the j1iCs state by a resonant laser pulse
and then fluorescence imaging for 15 ms. The
coherences of the data and spectator qubits
as a function of their individual decoupling
times are shown in Fig. 2, B and E, respec-
tively. Although the camera exposure time is
fixed, the imaging light is applied for a var-
iable time, 5t (of a total of 16t), to determine
its effect on the data qubits. Crucially, the data
qubit coherence time is unaltered by the MCR
MCR = 0.65(2) s].
The large detuning of the imaging light leads
to negligibly low spontaneous scattering rates
of ~10−7 Hz. Moreover, spontaneous Raman
scattering events that change these mF states
are further suppressed by a factor of 0.009
owing to destructive interference of the off-
resonant transition amplitudes (12). The the-
oretical T1 time from this decay channel is
thus ~108 s, resulting in a data qubit bit-flip
rate from readout cross-talk of ~10−11 during
the 15-ms MCR. This readout duration was
chosen to balance the requirements for achiev-
ing a high discrimination fidelity while mini-
mizing the time for a feedforward operation
(29). The discrimination fidelity of the specta-
tor qubit states (Fig. 2D) is extracted from a
,
bimodal fit to the fluorescence histogram of
each spectator qubit, as exemplified in Fig. 2C.
Across the spectator array, we find a mean
fidelity of 0.989(5), showing that the spectator
qubit states are well resolved by MCR.
Spectator protocol and correction
of phase errors
The preservation of data qubit coherence dur-
ing spectator readout opens the possibility for
feedforward operations within a quantum cir-
cuit. Under simultaneous evolution, noise chan-
nels can induce correlated phase errors between
the data and spectator qubits. Importantly, the
large number of spectator qubits allows single-
shot estimation of the acquired phase from one
simultaneous MCR. The phase accrued by the

Singh et al., Science 380, 1265–1269 (2023) 26 June 2023 2 of 5

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

Fig. 3. Mid-circuit correction of correlated phase errors. (A) Noise channels
induce correlated phase errors (red arrows) between the two sets of qubits.
Measurement of the spectators along the y axis enables single-shot phase
estimation, from which the phase accrued by the data qubits can be inferred
and corrected in real time (green arrow). (B) Gate sequence. The data and
spectator qubits are synchronously decoupled and acquire correlated errors
owing to magnetic field noise dBz. The spectator qubit decoupling sequence is
truncated, with the remaining time assigned for MCR and feedforward. CPU,
central processing unit; QPU, quantum processing unit. (C) Example coherence
p
g
y
measurement of the data qubits at the end of the sequence, with the feed-
forward turned on (green squares) and off (blue triangles). Field noise is
applied at fAC = 36.2 Hz and 10.7 mG RMS. Dashed lines are fits, from which
we extract hsx i = 0.53(1) and 0.02(2) for feedforward on and off, respectively.
(D) Data qubit hsx i as a function of the RMS noise strength at fAC. The shaded
green region indicates the correctable range (see text). (E) Data qubit hsx i
as a function of the noise frequency at 10.7 mG RMS. The shaded gray region
indicates an absolute gain in the measured coherence. For (D) and (E), solid
lines are the results of numerical simulations (see text).

data qubits can then be inferred and corrected
in real time, as illustrated in Fig. 3A.
To demonstrate this capability, we injected
global magnetic field noise with amplitudes
and frequencies comparable to those typically
is the time between p-pulses (22). The specta-
tors sample this noise for three-quarters of the
total evolution time of the data qubits, with the
remainder of the time assigned for MCR and
feedforward. To achieve fast camera process-
y g
in the absence of injected noise (29). F0S is
uniquely defined when the accrued phase lies
within [−p/2, p/2], beyond which the protocol
breaks down. The estimated noise-induced
phase accrued by the data qubits is given by
F0D ¼ gbF0S , where g = 4/3 is the ratio of the

found in laboratory environments. The phase
of the noise was random in each experimen-
tal repetition, without shot-to-shot temporal
correlations. We focused on monochromatic
noise for ease of synthesis and interpretation
of protocol performance but note that our
scheme is generally agnostic to the noise spec-
trum. The pulse sequence for the experiment
is shown in Fig. 3B. The data and spectator
qubits underwent synchronous dynamical de-
coupling and acquired correlated errors from
the common noise. Although the filter func-
tion of the Carr-Purcell-Meiboom-Gill (CPMG)–
type dynamical decoupling sequence partially
mitigates such noise, certain frequencies still
couple into the sequence, occurring at odd-
harmonics of fAC = 1/(4t) = 36.2 Hz, where 2t
ing and feedback, we used a camera-linked
classical control architecture for in-sequence
processing of the fluorescence images, which
in turn triggers an arbitrary-waveform gener-
ator to perform real-time updates of the phase
of the final data qubit p/2 pulse (29). The
phase update of this final p/2 pulse is equiv-
alent to a z-axis qubit rotation, which is used
to correct the noise-induced phase error on
the data qubits.
To estimate the phase acquired by the spec-
tators, FS , MCR was performed along an axis
orthogonal to the state preparation axis. Ac-
cordingly, the collective expectation value of
the array can be inverted to give an estimate,
F0S ¼ arcsinðhsy i=C Þ, where C is a scaling fac-
tor that describes the amplitude of the signal
,
sensing times and b = 1.35 is the ratio of the
second-order Zeeman shifts of the clock states
(29). With this knowledge, a real-time correc-
tion can be applied.
We first probed the case for which the noise
is maximally coupled, at fAC [10.7 mG root
mean square (RMS)]. Without the spectator
protocol, the random phase of the noise leads
to complete dephasing of the data qubits.
Notably, the feedforward corrects the noise-
induced phase in each experimental repeti-
tion, resulting in a recovery of the data qubit
coherence (Fig. 3C). The coherence as a func-
tion of the noise amplitude is shown in Fig. 3D.
In stark contrast to the rapid decay observed
in the absence of feedforward, the spectator

Singh et al., Science 380, 1265–1269 (2023) 26 June 2023 3 of 5

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

Fig. 4. Reloading of spectator qubits while maintaining data qubit loading fraction of 0.49. (B) Reloading spectators using PGC during data
coherence. (A) Reloading spectators using a pulsed MOT while decoupling the qubit decoupling. The data qubit coherence time is T2XY8 = 0.64(5) s with the

p
data qubits. The data qubit coherence time is T2XY4 = 0.42(3) s with the pulsed PGC light and T2XY8 = 0.65(2) s without it. Reloading occurs on a faster timescale
MOT and T2XY4 = 0.45(1) s without it. Spectators are reloaded on a timescale of of 90(30) ms compared with (A), saturating at a fraction of 0.32. Dashed lines
150(50) ms (time required to reach 1 − 1/e of asymptote), saturating at a are fits (29).

protocol robustly preserves coherence for field average number of loaded spectator qubits), out itself and the trapping lifetime. Therefore,
we find f ≈ 0.88, which is in good agreement

strengths less than 11 mG. Beyond this value,
the accrued phases on the spectator qubits can
exceed ±p/2, where the protocol can no longer
unambiguously detect phase errors.
Next, we studied the dependence on the noise
with the numerical simulations.
In the context of quantum information pro-
cessing, it is interesting to consider the re-
quirements to reach f ≫ 0:99 . Without any
g
continuous operation of atom-based quantum
processors will require reload and reset op-
erations that overcome these erasure errors
(32, 33). In this work, we explored two meth-
ods for reloading spectators while maintaining

frequency for an RMS noise strength of 10.7 mG
(Fig. 3E). For a range of frequencies close to
fAC, real-time correction results in an absolute
gain in the measured signal, shielding the data
qubits from otherwise deleterious decoherence.
A pair of small additional features occur near
fAC in the “feedforward-on” spectrum, arising
from the finite spectator readout time, which
leads to decorrelation between the data and
spectator qubits. Reducing the fraction of time
change in g or b, f = 0.99 could be achieved for
N = 165 and C = 1. At present, the value of C is
limited primarily by uncorrelated dephasing of
the spectator qubits, which is caused by thermal
motion in the optical tweezers and tweezer-
induced T1 processes. Thermal motion can be
reduced by additional cooling schemes, and T1
can be improved by increased detuning of the
optical tweezers.
Beyond optimizing for g ≈ 1, f can be fur-
y
coherent data qubits. These build on our stan-
dard procedure, in which a two-dimensional
magneto-optical trap (MOT) generates a beam
of atoms that is laser-cooled into the tweezer
array via a three-dimensional MOT.
The first reloading approach uses a strobo-
scopic MOT that is applied synchronously with
an XY4 sequence on the data qubits to de-
couple them from the magnetic field gradient
(Fig. 4A). Without the gradient, this decou-

used for MCR would suppress these effects.
Outside this region, feedforward causes a slight
reduction in the measured coherence, which
results from imperfect phase estimation. For
both the amplitude and frequency sweep, the
ther improved by reducing b, at the cost of a
reduced range of correctable data qubit er-
rors, FD;max ¼ Tgbp=2. This could be achieved
with alternative spectator qubit states, such
as magnetic field–sensitive states.
y g
pling sequence gives T2XY4 = 0.45(1) s. With it,
we find T2XY4 = 0.42(3) s, but the functional
form is modified (29). The spectator array
is reloaded on a much shorter timescale of
150(50) ms, defined as the time taken to reach
1 − 1/e of the asymptotic loading fraction.

salient features of the data are well described by
simple simulations of the experiment with no
free parameters aside from a global amplitude
rescaling (Fig. 3, D and E). These simulations
are based on the assumption of monochromatic
noise that solely perturbs the frequencies of the
qubits (29). At stronger noise strengths, a slight
discrepancy occurs, which likely arises from a
breakdown of these assumptions.
Alongside our numerical simulations, analytic
expressions can be derived for the error due to
quantum projection noise (QPN) in the phase-
estimation step. In the absence of any correlated
dephasing, QPN-induced feedforward errors
modulate the data qubit expectation values
g2 b2
hsx i by f ≈ 1  2NC 2 (29). For our experimental
parameters (C = 0.46, N = 61, where N is the
Although in this work we focus on mag-
netic field noise, the protocol can also mitigate
common-mode control errors. For instance, by
cotrapping the data and spectator qubits using
the same laser system (such as a far-detuned
1064-nm laser), phase errors induced by in-
tensity fluctuations of the trapping-laser light
could be corrected.
Coherent reloading of spectator qubits
In these experiments, fluorescence-based de-
tection of the spectators involves selectively
removing those in the j1iCs state before imag-
ing. Therefore, performing repetitive MCRs
will continuously deplete the array. Although
low-loss readout techniques exist (30, 31), fi-
nite losses always remain from both the read-
,
The pulsed MOT saturates at a loading frac-
tion of 0.49, which is comparable to that achieved
with the standard procedure. Residual de-
phasing from the field gradient can be over-
come by using low-inductance coils with faster
switching times and by performing decou-
pling pulses using a Raman laser system, which
would enable Rabi frequencies in the mega-
hertz range.
In the second approach, we used polarization-
gradient cooling (PGC) to load spectators di-
rectly from the atomic beam without a field
gradient (Fig. 4B). This both increases the load-
ing speed and allows an arbitrary choice of
decoupling parameters: Here, we used a single
cycle of XY8. In this reloading paradigm, the

Singh et al., Science 380, 1265–1269 (2023) 26 June 2023 4 of 5

RES EARCH | R E S E A R C H A R T I C L E
data qubit coherence time [T2XY8 = 0.64(5) s]
is unchanged from the values presented in
Fig. 2, and the spectator qubit array is re-
loaded on a timescale of 90(30) ms. The frac-
tion of total reloaded spectators is lower than
that in the previous method, saturating at
0.32. We hypothesize that this is limited by the
2-mm-diameter cooling beams. Incorporating
larger cooling beams will likely increase the
loading fraction for both approaches and would
enable reloading times of a few tens of milli-
seconds (34). Coherence times on the order of
seconds can be achieved by using further de-
tuned trapping light and a larger number of
decoupling pulses (9).
Discussion and outlook
A central challenge for all quantum architec-
tures is to increase system sizes while maintain-
ing low physical error rates. Our demonstration
of the use of spectator qubits to measure and
correct correlated phase noise is a broadly ap-
plicable strategy that can be used to reduce
error rates in quantum computing platforms.
Furthermore, spectator protocols could be
used in conjunction with standard quantum
error correction strategies to protect against
correlated errors as well as increase the fi-
delity of operations beyond the fault-tolerance
threshold. An attractive feature of this protocol
is that it does not necessitate interactions (two-
qubit gates) or individual spectator qubit con-
trol, which reduces hardware complexity. The
use of spectator qubits for noise measurements
may provide opportunities in quantum sens-
ing and metrology (22, 35, 36) and for im-
proving clock coherence within a single device
through differential spectroscopy between the
data and spectator qubits (37). Whereas in
this work we focused on global noise, arrays
of spectator qubits may also enable the de-
tection of spatially varying noise fields that
can be suppressed by local qubit addressing
(24). Careful engineering of the spectator qubits
and their control sequences may improve pro-
tocol performance. For example, spectator qubits
could be encoded in states with enhanced or
reduced noise sensitivity to increase the phase
resolution or the range of tolerable noise (25).
This can be achieved by using nonzero mF states
or by entangling the spectator qubits (22).
Singh et al., Science 380, 1265–1269 (2023) 26 June 2023
The methods demonstrated in this work
constitute a set of quantum-control techniques
that are essential for atom-array quantum pro-
cessors, including MCR, feedforward opera-
tions, and the reloading of auxiliary qubits
while maintaining quantum data. Combining
these capabilities with programmable intra-
species (9, 38) and interspecies Rydberg gates
will enable auxiliary qubit–assisted measure-
ments as required for quantum error correction
(32, 33, 39) and efficient preparation of long-
range entangled states (40). These same ca-
pabilities also enable the exploration of complex
dynamical quantum behavior under continuous
observation, including measurement-induced
phase transitions (41).
Note added in proof: After the consideration
of this manuscript, a preprint paper was pub-
lished that describes MCR of a neutral atom
by shelving in hyperfine states (42).
RE FERENCES AND NOTES
1. E. T. Campbell, B. M. Terhal, C. Vuillot, Nature 549, 172–179 (2017).
2. B. M. Terhal, Rev. Mod. Phys. 87, 307–346 (2015).
3. C. Ryan-Anderson et al., Phys. Rev. X 11, 041058 (2021).
4. L. Postler et al., Nature 605, 675–680 (2022).
5. Y. Zhao et al., Phys. Rev. Lett. 129, 030501 (2022).
6. S. Krinner et al., Nature 605, 669–674 (2022).
7. R. Acharya et al., Nature 614, 676–681 (2023).
8. M. H. Abobeih et al., Nature 606, 884–889 (2022).
9. D. Bluvstein et al., Nature 604, 451–456 (2022).
10. J. Preskill, Quantum Inf. Comput. 13, 181–194 (2013).
11. A. G. Fowler, J. M. Martinis, Phys. Rev. A 89, 032316 (2014).
12. S. Kuhr et al., Phys. Rev. A 72, 023406 (2005).
13. T. Monz et al., Phys. Rev. Lett. 106, 130506 (2011).
14. C. E. Bradley et al., Phys. Rev. X 9, 031045 (2019).
15. J. M. Boter et al., Phys. Rev. B 101, 235133 (2020).
16. C. D. Wilen et al., Nature 594, 369–373 (2021).
17. L. M. Vandersypen, I. L. Chuang, Rev. Mod. Phys. 76,
1037–1069 (2005).
18. H. A. Rabitz, M. M. Hsieh, C. M. Rosenthal, Science 303,
1998–2001 (2004).
19. T. Gullion, D. B. Baker, M. S. Conradi, J. Magn. Reson. 89,
479–484 (1990).
20. M. D. Shulman et al., Nat. Commun. 5, 5156 (2014).
21. S. Mavadia, V. Frey, J. Sastrawan, S. Dona, M. J. Biercuk,
Nat. Commun. 8, 14106 (2017).
22. C. L. Degen, F. Reinhard, P. Cappellaro, Rev. Mod. Phys. 89,
035002 (2017).
23. S. Majumder, L. Andreta de Castro, K. R. Brown, NPJ Quantum
Inf. 6, 19 (2020).
24. R. S. Gupta, L. C. Govia, M. J. Biercuk, Phys. Rev. A 102,
042611 (2020).
25. H. Song, A. Chantasri, B. Tonekaboni, H. M. Wiseman,
Phys. Rev. A 107, L030601 (2023).
26. K. Singh, S. Anand, A. Pocklington, J. T. Kemp, H. Bernien,
Phys. Rev. X 12, 011040 (2022).
27. E. Deist et al., Phys. Rev. Lett. 129, 203602 (2022).
28. J. T. Zhang et al., Quantum Sci. Technol. 7, 035006 (2022).
29. See supplementary materials.
30. A. Fuhrmanek, R. Bourgain, Y. R. P. Sortais, A. Browaeys,
Phys. Rev. Lett. 106, 133003 (2011).
31. M. J. Gibbons, C. D. Hamley, C.-Y. Shih, M. S. Chapman,
Phys. Rev. Lett. 106, 133002 (2011).
32. I. Cong et al., Phys. Rev. X 12, 021049 (2022).
33. Y. Wu, S. Kolkowitz, S. Puri, J. D. Thompson, Nat. Commun. 13,
4657 (2022).
34. A. Steane, M. Chowdhury, C. Foot, J. Opt. Soc. Am. B 9, 2142
(1992).
35. I. S. Madjarov et al., Phys. Rev. X 9, 041052 (2019).
36. M. A. Norcia et al., Science 366, 93–97 (2019).
37. M. E. Kim et al., Nat. Phys. 19, 25–29 (2023).
38. T. M. Graham et al., Nature 604, 457–462 (2022).
39. M. Morgado, S. Whitlock, AVS Quantum Sci. 3, 023501 (2021).
40. R. Verresen, N. Tantivasadakarn, A. Vishwanath, Efficiently
preparing Schrödinger’s cat, fractons and non-Abelian
topological order in quantum devices. arXiv:2112.03061
[quant-ph] (2021).
41. Y. Li, X. Chen, M. P. A. Fisher, Phys. Rev. B 98, 205136
(2018).
42. T. M. Graham et al., Mid-circuit measurements on a neutral
atom quantum processor. arXiv:2303.10051 [quant-ph]
(2023).
43. K. Singh et al., Data and plotting code for “Mid-circuit correction
of correlated phase errors using an array of spectator qubits.”
Zenodo (2023); https://doi.org/10.5281/zenodo.7612718.
AC KNOWLED GME NTS
p
We thank H. Levine, J. Covey, and A. Clerk for fruitful discussions
and critical reading of the manuscript. Funding: We acknowledge
funding from the Office for Naval Research (N00014-20-1-2510),
the Air Force Office of Scientific Research (FA9550-21-1-0209),
the NSF Quantum Leap Challenge Institutes (QLCI) for Hybrid
Quantum Architectures and Networks (NSF award 2016136), and
the Sloan Foundation. This material is based upon work supported
by the US Department of Energy, Office of Science, National
g
Quantum Information Science Research Centers, and was
supported by an appointment to the Intelligence Community
Postdoctoral Research Fellowship Program at the Pritzker School
of Molecular Engineering administered by the Oak Ridge Institute
for Science and Education through an interagency agreement
y
between the US Department of Energy and the Office of the
Director of National Intelligence. Author contributions: K.S., C.E.B.,
S.A., V.R., R.W., and H.B. contributed to the experiments. K.S., C.E.B.,
S.A., and H.B. contributed to the analysis of the results and
preparation of the manuscript. Competing interests: The work
described in this article is included in a patent application filed with
the US Patent and Trade Office by the University of Chicago.
Data and materials availability: The data and code supporting
this study are available at Zenodo (43). License information:
Copyright © 2023 the authors, some rights reserved; exclusive
licensee American Association for the Advancement of Science.
No claim to original US government works. https://www.science.org/
about/science-licenses-journal-article-reuse
y g
SUPPLEMENTARY MATERIALS
science.org/doi/10.1126/science.ade5337
Materials and Methods
Supplementary Text
Figs. S1 to S5
Tables S1 and S2
,
References (44–48)
Submitted 24 August 2022; accepted 15 May 2023
Published online 25 May 2023
10.1126/science.ade5337
5 of 5
RES EARCH

NEUROSCIENCE long-term stability without sacrificing acqui-

sition speed.
Wide-field fluorescence lifetime imaging of neuron We implemented our approach through in-
corporation of a Pockels cell into the fluorescence
spiking and subthreshold activity in vivo detection path of a standard epifluorescence
microscope (Fig. 1A, fig. S1, and materials and
Adam J. Bowman1*, Cheng Huang2,† Mark J. Schnitzer2,3,4, Mark A. Kasevich1* methods). The Pockels cell design was optimized
for resonant drive and wide-field imaging, in-
The development of voltage-sensitive fluorescent probes suggests fluorescence lifetime as a promising corporating thermal control and transverse
readout for electrical activity in biological systems. Existing approaches fail to achieve the speed crystal geometry to cancel off-axis birefrin-
and sensitivity required for voltage imaging in neuroscience applications. We demonstrated that gence. A high-voltage modulation was applied
wide-field electro-optic fluorescence lifetime imaging microscopy (EO-FLIM) allows lifetime imaging at to the Pockels cell with a resonant transformer
kilohertz frame-acquisition rates, spatially resolving action potential propagation and subthreshold (figs. S2 and S3), resulting in a fast polariza-
neural activity in live adult Drosophila. Lifetime resolutions of <5 picoseconds at 1 kilohertz were tion rotation that was synchronous with the
achieved for single-cell voltage recordings. Lifetime readout is limited by photon shot noise, and the excitation pulses from a 100-ps laser source.
method provides strong rejection of motion artifacts and technical noise sources. Recordings revealed Fluorescence from the sample was first po-
local transmembrane depolarizations, two types of spikes with distinct fluorescence lifetimes, and larized, then polarization was modulated with
phase locking of spikes to an external mechanical stimulus. the Pockels cell and finally split with a polariz-
ing beamsplitter into two wide-field images on

R
a scientific complementary metal-oxide semi-
ecording the electrical activity of neu- lifetime for static measurements of membrane conductor (sCMOS) camera corresponding

rons at high spatial and temporal reso-
lution is central to understanding brain
function. Fluorescent probes of cal-
cium activity enable optical studies of
large neuron populations in vivo (1, 2). How-
ever, the response time of calcium indicators
potential in vitro (11). However, existing life-
time detectors—for example, single-photon
counters or cameras with modulated pixels
(12)—fall short of the requirements for detect-
ing fast dynamics and imaging neurons in vivo,
either because of their limited photon through-
p
to gated (G) and ungated (U) intensity. These
two images encoded nanosecond time infor-
mation in their intensity ratio. Because gating
was performed with a beamsplitter, it was pos-
sible to capture the entire fluorescence decay
with photon efficiency limited by optical coat-

is much slower than the underlying electri-
cal signals. Fluorescent voltage sensors are a
complementary approach, providing direct
readout of neuron membrane potential with
the capability to resolve action potentials.
put or because of prohibitively high noise.
EO-FLIM technique
We demonstrated an approach that uses electro-
optic fluorescence lifetime imaging micros-
g
ings. In this study, we modulated one input
polarization and discarded half of the avail-
able signal on a first polarizer. This can be
avoided in the future through addition of a
second beam path (13). The fluorescence de-

Although voltage probes have rapidly devel-
oped with a variety of genetically encoded
(3–7) and chemical dyes (8) in use, there re-
main considerable challenges to their appli-
cation in vivo because of low sensitivity, rapid
photobleaching, and motion artifacts. To
achieve high-speed recording and sufficient
signal-to-noise ratios (SNRs), most voltage
probes use fluorescence intensity to read out
an underlying sensing mechanism on the basis
copy (EO-FLIM), an all-optical technique for
lifetime imaging that is based on nanosecond
gating of a wide-field image (13, 14). EO-FLIM
allows efficient photon collection and is com-
patible with detection on high-speed, low-
noise scientific cameras. With this method, we
achieved lifetime imaging of action potentials
in vivo.
EO-FLIM enables significant suppression of
intensity artifacts, allowing robust imaging
y
cay at each pixel was convolved with the tem-
poral gating function of the Pockels cell and
then sampled at a single modulation phase
relative to the excitation laser (Fig. 1B). Each
image thus provided a lifetime estimate for
every pixel in parallel at the frame rate of the
scientific camera. When imaging genetically
targeted neurons in vivo, this technique al-
lowed for 1-kHz–frame rate recordings with
a lifetime sensitivity of 2.53 ± 0.48 ps with

of absorption, Förster resonance energy trans-
fer (FRET), or quenching.
We applied an alternative strategy for de-
tecting fast probe dynamics that is based on
lifetime imaging (9). Voltage sensors that use
in the presence of tissue motion and fluctua-
tions in illumination intensity. Such artifacts
are ubiquitous in recordings of neural activity
from awake, behaving animals (15, 16). This
has two consequences: (i) It enables faithful
y g
0.7 × 107 to 1.4 × 107 detected photons per
frame (Fig. 1C), indicating a substantial im-
provement in throughput over previous FLIM
recordings of cellular dynamics. EO-FLIM
approaches fundamental sensitivity limits

FRET and quenching mechanisms modulate
the probe’s nonradiative decay rate, intrinsi-
cally connecting fluorescence intensity with
nanosecond excited-state lifetime. Lifetime is
a promising readout for voltage imaging, es-
pecially because of its capability to provide an
absolute indication of membrane potential
(10). Recent results have validated fluorescence
1
Physics Department, Stanford University, Stanford, CA 94305,
USA. 2James H. Clark Center, Stanford University, Stanford,
CA 94305, USA. 3CNC Program, Stanford University, Stanford,
CA 94305, USA. 4Howard Hughes Medical Institute, Stanford
University, Stanford, CA 94305, USA.
*Corresponding author. Email: abowman2@stanford.edu (A.J.B.);
kasevich@stanford.edu (M.A.K.)
†Present address: Department of Neuroscience, Washington Univer-
sity School of Medicine, St. Louis, MO 63110, USA.
recording of subthreshold voltage waveforms,
and (ii) it improves the SNR by suppressing
high-frequency intensity noise. These follow
from the fact that EO-FLIM estimates lifetime
from the ratio of a pair of simultaneously
recorded intensity channels derived from a
common optical source. In conventional ap-
proaches, corrections for intensity noise use
ratios of measurements that are nonsimulta-
neous and easily corrupted by high-frequency
noise or slower motion artifacts. Optical sen-
sors of neuron activity are typically reported
by DF =F , referencing fast intensity changes
(DF) to a nonsimultaneous, average fluores-
cence baseline (F ). Fluorescence lifetime can
read out an intensity-optimized sensor with
improved temporal noise performance and
,
for estimating lifetimes between 1 and 4 ns
(fig. S2).
Results
We used EO-FLIM to image a genetically en-
coded voltage indicator (GEVI) in Drosophila
melanogaster. We expressed pAce positive
polarity GEVI in a subtype of fly mushroom
body output neuron (MBON), MBON-g1pedc>ab.
pAce works through FRET and is a fusion of
the bright fluorescent protein mNeonGreen
with a voltage-sensitive opsin from Acetabularia
(3). We surgically prepared Drosophila before
imaging to provide optical access to the brain
(17, 18). Action potentials and subthreshold
dynamics were readily resolved with action
potentials corresponding to a 20- to 50-ps

Bowman et al., Science 380, 1270–1275 (2023) 23 June 2023 1 of 6

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

Fig. 1. Lifetime imaging of action potentials.

(A) Schematic of EO-FLIM microscope. Wide-field
fluorescence images were modulated with a Pockels
cell (PC) placed between crossed polarizers (P
and PBS) and driven at 20 MHz by a high-voltage
resonant transformer. Two spatially offset output
images were simultaneously captured after a second
polarizing beamsplitter (PBS) on a sCMOS camera,
corresponding to gated (G) and ungated (U)
intensities. (B) Instrument response function (IRF)
and fluorescence traces for the U channel were
measured by varying the Pockels-cell drive phase
relative to the excitation laser. The pAce GEVI was fit
to 2.2-ns lifetime. For kilohertz imaging, a single
optimal phase point was captured (vertical line
at 0-ns delay), and the G/U image intensity ratio
was converted to a lifetime estimate (fig. S1).
(C) Histogram of measurements (highpass filtered)
obtained at 1 kHz for a single neuron in vivo
demonstrate a lifetime sensitivity of 3 ps (full trace in
Fig. 2A). (D) Wide-field image of a neuron with

p
structures indicated. Scalebar, 25 mm. (E) Whole-cell
lifetime trace resolves action potentials and sub-
threshold transitions. (F and G) Average spike shape
is plotted in intensity and lifetime from color-coded
regions. (H) Frames from an interpolated lifetime
movie demonstrate spike propagation, averaging the

g
signal from ~300 individual spikes. The point of
initiation is indicated by the arrow, and bidirectional
propagation was observed both along the axon
and backward toward soma and dendrites (movies S1
to S3). Spike propagation was also imaged directly

y
without averaging (movies S4 and S5). (I) Applying a
10-frame moving average allowed subthreshold
signals to be localized to neuron structures (movies
S6 and S7). Example frames demonstrate localization
in the dendrite for both positive and negative
subthreshold signals.

shift in the fluorescent lifetime (Fig. 1, D and
E). Average spike readouts in lifetime and in-
tensity for different neuron subregions were
in strong agreement both in their relative tim-
ing and amplitudes (Fig. 1, F and G).
The donor fluorescence lifetime of a FRET
GEVI depends on radiative decay rate, kf, and
the voltage-sensitive nonradiative decay rate,
knr(V), associated with FRET as t ¼ kf þ k1nr ðV Þ
(10, 19). In pAce, the Ace opsin acts as acceptor
and provides voltage sensitivity. The donor’s
fluorescence intensity is directly quenched by
FRET, giving a signal, DF º sDknr , where s
is the donor excitation cross section. For an
ideal FRET process, one expects to find Dt=t ¼
DF =F. pAce gave a linear but attenuated life-
time response of ð0:70 T 0:07ÞDF =F (fig. S4).
This may indicate components of the GEVI
response in intensity—for example, modulation
of cross section s—that did not affect lifetime
readout.
Wide-field lifetime imaging correlates neu-
ron activity with spatial structure. The point
of action potential initiation in the axon is
resolved along with bidirectional propagation
along the axon and backward toward the den-
drite and soma (20). Action potentials were
attenuated and broadened in the dendrites
y g
,
and soma (Fig. 1, F and G, and figs. S5 and S6).
Spike propagation is shown in movies S1 to
S3 with still frames from movie S1 displayed
in Fig. 1H, generated with spike-triggered av-
eraging over ~300 spikes and interpolating
between frames. We also observed individ-
ual spikes and spike propagation in real time
without temporal averaging (movies S4 and
S5). Comparison of recordings from multiple
neuronal subregions revealed local depolari-
zations in the axons, which fail to initiate
action potentials across the entire cell. These
depolarizations were not resolved in inten-
sity readout but are clearly visible in lifetime

Bowman et al., Science 380, 1270–1275 (2023) 23 June 2023 2 of 6

RES EARCH | R E S E A R C H A R T I C L E
Fig. 2. Lifetime suppresses intensity noise and improves fidelity of subthreshold
recording. Six example MBON neurons are shown comparing lifetime (blue)
with DF=F intensity recordings (black). Recordings were obtained by averaging
over high-resolution images shown at far left. Scalebar, 25 mm. Shaded boxes
highlight some notable regions of the traces for improved lifetime readout. For
each example, the distributions of spike SNR are compared for intensity and
lifetime, with calculated spike detection fidelity d′ indicated. (A and B) Two
examples of flies without motion demonstrate improvement of technical noise
floor at high frequencies by up to 7 dB. The noise power spectra for the traces
are compared at far right, with dotted lines indicating the photon shot-noise
limits. (C to E) Three examples of flies having low-frequency noise associated with
motion artifacts. Lifetime improves noise power spectrum across temporal
frequencies, rejecting intensity noise by up to 9 dB at low frequencies (further
analysis is provided in Fig. 3). (F to J) Lifetime provided an improved readout of
two spike amplitudes in response to mechanical stimulus at 60 Hz. Large-amplitude
(L) spikes showed an enhanced lifetime responsivity and tripled detection SNR
Bowman et al., Science 380, 1270–1275 (2023) 23 June 2023
p
g
y
y g
,
and d′ over the small-amplitude (S) spikes. L spikes occurred independent of
subthreshold waveform level but synchronized with spiking on plateaus shown in
the inset in (F). (G) and (H) show average spike waveforms for color-coded
regions. The point of initiation for S spikes was a central region of the axon
(consistent with movies S1 and S2), whereas L spikes were diffuse and associated
with background fluorescence. L spikes also correspond to local spikes in the
dendrite and soma in (H). The L-spike background component possibly resulted
from out-of-focus neurons (movies S8 and S9). In (I) and (J), histograms of action
potential amplitudes are compared. In intensity, the L and S populations were not
resolved and strongly overlap, but they were clearly separated in lifetime.
Lifetime is used to identify the spikes, and the intensity histogram in (I) is shaded
with two colors to show overlapping populations. (K) A polar histogram demonstrates
strong phase locking of the L spikes to mechanical stimulus by referencing
phase to a bandpass-filtered lifetime trace. S spikes do not show phase locking.
P
(L) The average phase vector length, i cosðqi Þ=NAP, is plotted versus bandpass
center frequency to show narrow-band locking response.
3 of 6
RES EARCH | R E S E A R C H A R T I C L E

recordings (figs. S5 and S6). When we then

applied a 10-frame moving average, the spatial
distribution of slower subthreshold voltage
signals could also be studied. Subthreshold
signals were often strongest and localized in
the dendrites (movies S6 and S7). Still frames
from these movies are shown in Fig. 1I and
figs. S7 and S8.
By measuring a ratio of simultaneous in-
tensities, EO-FLIM removes noise sources that
are common mode to both the modulated
channels. In Fig. 2, we show example record-
ings of MBON-g1pedc>a=b neurons from six
flies, comparing intensity DF =F with lifetime
readouts. In all recordings, lifetime detec-
tion enhanced the SNR for action potential
detection by approximately twofold. This SNR
was quantified by comparing spike ampli-
tude with the high-frequency noise floor. We
also analyzed traces by using the spike de-
tection fidelity, d′, a discriminability index

p
that quantifies spike detection by comparing
the statistical distributions of spike ampli-
tudes and background-noise fluctuations (21).
Lifetime detection improved d′ by 1.5 to 2.4
times (Fig. 2, A to E).
Noise power was also compared across tem-

g
poral frequencies, demonstrating broad sup-
pression of intensity noise (Fig. 2, A to E) and
showing that EO-FLIM approaches the pho-
ton shot-noise limit. To allow a direct compar-
ison of noise power spectrum, the responsivity

y
of spikes in intensity and lifetime channels
was normalized. For flies not displaying much
motion (Fig. 2, A and B), lifetime primarily
Fig. 3. Lifetime improves uniformity in action potential amplitude and threshold level. (A to D). Histograms
reduced technical noise at high frequencies
of action potential amplitudes (lifetime in blue) and action potential levels on the subthreshold waveform
that resulted from the excitation laser (4 to
(lifetime in green) are plotted for each activity trace, overlayed on the same histograms for intensity (gray).
7 dB). Even in these well-behaved examples,
Action potential amplitudes are normalized to the mean. Subthreshold level is also mean normalized as
lifetime readout resulted in improved SNR  
L  L =L, where L is the spike’s corresponding level on a low-pass filtered trace, and its distance is
and d′.
Lifetime recordings also improved long-term measured relative to the mean level of all other spikes, L. A perfectly uniform threshold would thus result in
stability in the voltage readout. Intensity-based L = 0 for all spikes. In each histogram, the ratio of standard deviation between intensity and lifetime readouts,

voltage imaging often displays strong motion
artifacts, which degrade stability. In Drosophila,
these artifacts result from movements such as
extension of the proboscis. For flies displaying
motion (Fig. 2, C to E, and fig. S9), lifetime
y g
sF =st , is given as a figure of merit for uniformity. (A) and (B) to (D) correspond to Fig. 2, A and C to E,
respectively.
of subthreshold voltage plateaus, whereas initiated in a local area of the axon, as shown
large-amplitude (L) spikes were observed in in movies S1 and S2. The L spikes may be as-

readout suppressed artifacts at low frequen-
cies by up to 9 dB as compared with intensity
readout (in these examples, subthreshold wave-
forms may only be resolved in lifetime). Figure 3
shows trace stability quantified according to
the spike-amplitude distribution and the uni-
formity of threshold voltage level at action
potential locations. Histograms of mean nor-
malized spike heights and mean normalized
subthreshold level are plotted for each spike,
showing that lifetime improves spike unifor-
mity by up to 2.5 times and threshold unifor-
mity by up to 5.8 times.
With the improved readout stability af-
forded by lifetime recording, we observed
two spike amplitudes (Fig. 2, F to L). The
small-amplitude (S) spikes occurred on top
bursts that were independent of subthresh-
old voltage level. In this sample, the fly was
mechanically stimulated with sound waves
near a resonance of the microscope stage.
Figure 2, I and J, show histograms of spike
heights in both intensity and lifetime. The
two spike populations were clearly distin-
guished by using lifetime readout but were
not separable with intensity readout. The L
spikes showed enhanced lifetime responsiv-
ity (1.34 DF =F, compared with 0.68 DF =F for
S spikes). The difference in responsivity may
indicate kinetic differences in the GEVI re-
sponse to different action potential waveforms
(3). Spike-triggered averages (movies S8 and
S9) showed that the L spikes originated dif-
fusely across the image, whereas the S spikes
,
sociated with out-of-focus neurons displaying
off-target GEVI expression and were also syn-
chronous with spiking of the targeted neuron
(Fig. 2, F to H). The L spikes displayed strong
phase locking in response to mechanical stim-
ulus, whereas the S spikes did not. A histogram
of phases is displayed in Fig. 2K, where phase
is determined by each spike’s location on a
trace that has been filtered at the stimulus
frequency. Narrow-band locking response is
demonstrated in Fig. 2L. Similar L spikes were
also observed during the mechanical stimulus
sweep (Fig. 4C).
To further demonstrate the noise-rejection
capabilities of lifetime detection, we placed
the fly on a piezoelectric stage to provide
mechanical shaking in the xy plane. This

Bowman et al., Science 380, 1270–1275 (2023) 23 June 2023 4 of 6

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

p
Fig. 4. Phase locking of spikes to direct mechanical stimulus. (A) Lifetime line sweep, and phase locking appears as increased frequency content at
provided strong rejection of intensity noise associated with shaking the the stimulus frequency during spike bursts. (D) To show phase locking
sample (40-Hz square wave, ~0.8 mm peak-to-peak). (B) Intensity noise at the visually, a sliding window autocorrelation of the lifetime trace is plotted with

stimulus frequency and second harmonic were attenuated by 21 and 20 dB.
(C) A spectrogram of the lifetime trace is plotted as stimulus is swept from
50 to 150 Hz. Vertical lines of activity in the spectrogram correspond to
spike bursts in the lifetime trace. Mechanical cross-talk is seen as the diagonal
g
a 150-ms window. Phase locking may be seen by observing alignment of
autocorrelation peaks during activity bursts to the peaks resulting from
mechanical cross-talk signal. Examples of bursts showing phase locking are
highlighted (blue vertical bars).

direct mechanical stimulus resulted in high
levels of intensity noise that obscured neu-
ron activity, but lifetime readout suppressed
this noise by up to 21 dB (Fig. 4, A and B).
Using this direct stimulus, we could observe
phase-locked spiking behavior from 30 to
>100 Hz (Fig. 4 and fig. S10). As shown in
these figures, we observed phase locking
through increased spectral power at the stim-
ulus frequency in a spectral waterfall plot as
quantitative FRET measurements with only the
donor channel. In voltage imaging, lifetime will
enable improved measurement of FRET-opsin
(3, 7), hybrid FRET (6), and dye indicators (8).
We also anticipate application to imaging cal-
cium (27), neurotransmitters (28, 29), and cyclic
adenosine monophosphate (30).
Use of GEVIs in vivo is often accompanied
by a large fluorescence background that re-
sults from protein expression outside the cel-
y
in vivo, overcoming the throughput and sensi-
tivity limitations of existing FLIM techniques.
We expect straightforward application to other
systems, including mammalian brains, which
feature both larger neurons and action poten-
tials as compared with those of Drosophila
(3, 7). Voltage imaging in neuroscience is one
example application, but membrane potential
is also broadly interesting throughout biology,
from bacteria (33, 34) and plants (35) to car-

the excitation frequency was swept. This
phase locking may also be visualized with
the autocorrelation of the lifetime trace (Fig.
4, C and D). These observations are consis-
tent with previous studies on mechanical
and auditory effects in Drosophila that iden-
lular membrane (31) or leaky gene expression
from nontargeted cells (32). This background
signal had a different fluorescence lifetime
and photobleaches at a different rate, resulting
in slow drifts of the measured lifetime traces
y g
diac (36–38) and muscle tissue (39). The ability
of EO-FLIM to strongly reject motion noise in
vivo is relevant for brain- and cardiac-imaging
applications in which it is challenging to faith-
fully distinguish voltage dynamics from mo-
tion and hemodynamic artifacts (15, 16, 36, 37).

tified a broad auditory response across the
central brain (22) and responses to substrate
vibrations (23, 24).
Discussion
EO-FLIM may be applied to both existing
lifetime-sensitive probes and to donor read-
out of FRET-based biosensors. Standard FRET
sensors are read out by the ratio of optical
intensities in spectrally separated donor and
acceptor channels (25), requiring an acceptor
molecule with high quantum yield. Two-color
readout frequently limits detection sensitivity
because of spectral cross-talk and also pre-
vents probe multiplexing (26). Lifetime mea-
surement removes these limitations and allows
(fig. S11). (We expect that in vitro studies will
not be affected by such backgrounds.) To mea-
sure absolute voltage, signal and background
populations would need to be unmixed by
discriminating between two closely spaced ex-
ponential decays. For this reason, we focus on
the improved stability and noise performance
afforded by lifetime measurement rather than
on absolute quantification. In our current im-
plementation, we measured the population-
weighted average lifetime by acquiring images
at a single modulation phase. In the future,
multiple modulation phases can be combined
to unmix lifetime components and improve
absolute measurement.
We have shown that EO-FLIM enables flu-
orescence lifetime imaging of neuron activity
,
It may become possible to perform voltage
imaging during natural movements such as
insect flight, or while tracking a freely moving
organism (40). Further, recent advances in
GEVI probes have enabled voltage imaging of
populations of neurons (3, 5). Lifetime imag-
ing will establish accurate long-term readout
of subthreshold activity across a neural cir-
cuit, allowing functional connectivity mapping
where spike activity may be correlated with
subthreshold modulation of downstream neu-
rons. Last, by using lifetime detection in com-
bination with optogenetic tools (4, 41), it will
be possible to improve techniques for targeted
optical activation and control (42) of neuron
membrane potential.

Bowman et al., Science 380, 1270–1275 (2023) 23 June 2023 5 of 6

RES EARCH | R E S E A R C H A R T I C L E
RE FE RENCES AND N OT ES
1. T. H. Kim, M. J. Schnitzer, Cell 185, 9–41 (2022).
2. M. B. Ahrens, M. B. Orger, D. N. Robson, J. M. Li, P. J. Keller,
Nat. Methods 10, 413–420 (2013).
3. M. Kannan et al., Science 378, eabm8797 (2022).
4. H. Tian et al., All-optical electrophysiology with improved
genetically encoded voltage indicators reveals interneuron
network dynamics in vivo, bioRxiv 2021.11.22.469481 [Preprint]
(2021). https://doi.org/10.1101/2021.11.22.469481.
5. K. D. Piatkevich et al., Nature 574, 413–417 (2019).
6. A. S. Abdelfattah et al., Science 365, 699–704 (2019).
7. Y. Gong et al., Science 350, 1361–1366 (2015).
8. P. Liu, E. W. Miller, Acc. Chem. Res. 53, 11–19 (2020).
9. R. Datta, T. M. Heaster, J. T. Sharick, A. A. Gillette, M. C. Skala,
J. Biomed. Opt. 25, 071203 (2020).
10. D. Brinks, A. J. Klein, A. E. Cohen, Biophys. J. 109, 914–921
(2015).
11. J. R. Lazzari-Dean, A. M. Gest, E. W. Miller, eLife 8, e44522
(2019).
12. M. Raspe et al., Nat. Methods 13, 501–504 (2016).
13. A. J. Bowman, M. A. Kasevich, ACS Nano 15, 16043–16054
(2021).
14. A. J. Bowman, B. B. Klopfer, T. Juffmann, M. A. Kasevich,
Nat. Commun. 10, 4561 (2019).
15. J. D. Marshall et al., Cell 167, 1650–1662.e15
(2016).
16. M. S. Creamer, K. S. Chen, A. M. Leifer, J. W. Pillow,
PLOS Comput. Biol. 18, e1010421 (2022).
17. S. Sinha et al., Proc. Natl. Acad. Sci. U.S.A. 110, 18374–18379
(2013).
18. C. Huang et al., Nat. Commun. 9, 872 (2018).
19. R. M. Clegg, in FRET and FLIM Techniques, T. W. J. Gadella, Ed.,
vol. 33 of Laboratory Techniques in Biochemistry and Molecular
Biology (2009), pp. 1–57.
20. C. Huang et al., Dopamine signals integrate innate and learnt
valences to regulate memory dynamics, Research Square
[Preprint], Version 1 (2022). https://assets.researchsquare.com/
files/rs-1915648/v1_covered.pdf?c=1660253275.
21. B. A. Wilt, J. E. Fitzgerald, M. J. Schnitzer, Biophys. J. 104,
51–62 (2013).
Bowman et al., Science 380, 1270–1275 (2023) 23 June 2023
22. D. A. Pacheco, S. Y. Thiberge, E. Pnevmatikakis, M. Murthy,
Nat. Neurosci. 24, 93–104 (2021).
23. P. Hehlert, W. Zhang, M. C. Göpfert, Trends Neurosci. 44,
323–335 (2021).
24. W. Zhang, Z. Yan, L. Y. Jan, Y. N. Jan, Proc. Natl. Acad. Sci. U.S.A.
110, 13612–13617 (2013).
25. W. Akemann et al., J. Neurophysiol. 108, 2323–2337
(2012).
26. D. M. Grant et al., Biophys. J. 95, L69–L71
(2008).
27. F. H. van der Linden et al., Nat. Commun. 12, 7159
(2021).
28. P. Ma et al., Fluorescence lifetime enables high-resolution
analysis of neuromodulator dynamics across time and animals,
bioRxiv 2022.09.28.510014 [Preprint] (2022). https://doi.org/
10.1101/2022.09.28.510014.
29. L. Sistemich, P. Galonska, J. Stegemann, S. Kruss,
Angew. Chem. Int. Ed. Engl. 62, e202300682
(2023).
30. J. Klarenbeek, J. Goedhart, A. van Batenburg, D. Groenewald,
K. Jalink, PLOS ONE 10, e0122513 (2015).
31. J. Platisa, V. A. Pieribone, Curr. Opin. Neurobiol. 50, 146–153
(2018).
32. Y. Aso et al., eLife 3, e04577 (2014).
33. A. Prindle et al., Nature 527, 59–63 (2015).
34. J. M. Kralj, D. R. Hochbaum, A. D. Douglass, A. E. Cohen,
Science 333, 345–348 (2011).
35. N. B. C. Serre et al., Nat. Plants 7, 1229–1238 (2021).
36. C. D. Acker, P. Yan, L. M. Loew, Prog. Biophys. Mol. Biol. 154,
3–10 (2020).
37. P. Lee et al., Cardiovasc. Res. 115, 1659–1671 (2019).
38. J. P. Wikswo Jr., S. F. Lin, R. A. Abbas, Biophys. J. 69,
2195–2210 (1995).
39. N. Azimi Hashemi et al., Proc. Natl. Acad. Sci. U.S.A. 116,
17051–17060 (2019).
40. J. P. Nguyen et al., Proc. Natl. Acad. Sci. U.S.A. 113,
E1074–E1081 (2016).
41. Y. Adam et al., Nature 569, 413–417 (2019).
42. A. C. F. Bergs et al., Nat. Comm. 14, 1939 (2023).
43. A. J. Bowman, C. Huang, M. J. Schnitzer, M. A. Kasevich,
Wide-field fluorescence lifetime imaging of neuron spiking and
sub-threshold activity in vivo, Dryad (2023); https://doi.org/
10.5061/dryad.nzs7h44wd.
44. A. J. Bowman, C. Huang, M. J. Schnitzer, M. A. Kasevich,
Wide-field fluorescence lifetime imaging of neuron spiking and
sub-threshold activity in vivo, Version 1, Zenodo, (2023);
https://doi.org/10.5281/zenodo.7706488.
AC KNOWLED GME NTS
Funding: We acknowledge funding from the Gordon and Betty
Moore Foundation; the US Department of Energy, Office of Science,
Office of Biological and Environmental Research, under award
DE-SC0021976; NIH grant U01NS120822 (M.J.S. and G. Vasan);
and NSF NeuroNex grant DBI-1707261 (M.J.S. and K. Deisseroth).
A.J.B. acknowledges support from the NSF Graduate Research
Fellowship under grant 1656518 and the Stanford Graduate
Fellowship. Author contributions: A.J.B. developed the microscope.
C.H. prepared Drosophila for imaging. A.J.B. and C.H. performed the
experiments. A.J.B. and M.A.K. analyzed data and wrote the
manuscript. All authors contributed to experiment conception and
manuscript revision. Competing interests: A.J.B. and M.A.K. are
inventors on PCT/US2019/062640, US17/153438, and US17/
898093. Data and materials availability: All data are available in
the manuscript, in the supplementary material, or deposited at
Dryad (43). Code is available at Zenodo (44). License
information: Copyright © 2023 the authors, some rights reserved;
exclusive licensee American Association for the Advancement of
Science. No claim to original US government works. https://www.
p
science.org/about/science-licenses-journal-article-reuse
SUPPLEMENTARY MATERIALS
science.org/doi/10.1126/science.adf9725
Materials and Methods
Figs. S1 to S11
Movies S1 to S9
g
MDAR Reproducibility Checklist
View/request a protocol for this paper from Bio-protocol.
Submitted 23 November 2022; accepted 16 May 2023
10.1126/science.adf9725
y
y g
,
6 of 6
RES EARCH

PLANT SCIENCE sue stresses can be tensile or compressive; the

epidermis is under tissue tension (Fig. 1B, di-
Brassinosteroid coordinates cell layer interactions in vergent red arrows), whereas internal regions
are under tissue compression (convergent blue
plants via cell wall and tissue mechanics arrows).
Tissue stresses can be revealed by immedi-
Robert Kelly-Bellow1*†, Karen Lee1†, Richard Kennaway1, J. Elaine Barclay1, Annabel Whibley1, ate outward recurvature of median slices through
Claire Bushell1, Jamie Spooner1, Man Yu1, Paul Brett2, Baldeep Kular2, Shujing Cheng3, internodes or by the formation of epidermal
Jinfang Chu3,4, Ting Xu5, Brendan Lane6, James Fitzsimons7, Yongbiao Xue5, Richard S. Smith6*, cracks when adhesion between cells is weak-
Christopher D. Whitewoods1,7*, Enrico Coen1* ened (6, 8, 9). They can be quantified by stretch-
ing detached epidermal tissue to the point that
Growth coordination between cell layers is essential for development of most multicellular organisms. its original length is restored (10, 11). However,
Coordination may be mediated by molecular signaling and/or mechanical connectivity between cells, little is known about how tissue stresses are
but how genes modify mechanical interactions between layers is unknown. Here we show that genes controlled genetically and thus the role they
driving brassinosteroid synthesis promote growth of internal tissue, at least in part, by reducing may play in non–cell-autonomous gene action.
mechanical epidermal constraint. We identified a brassinosteroid-deficient dwarf mutant in the aquatic In this study, we addressed this problem through
plant Utricularia gibba with twisted internal tissue, likely caused by mechanical constraint from a slow- the analysis of dwarf mutants in the aquatic
growing epidermis. We tested this hypothesis by showing that a brassinosteroid mutant in Arabidopsis plant Utricularia gibba and the terrestrial
enhances epidermal crack formation, indicative of increased tissue stress. We propose that by plant Arabidopsis thaliana.
remodeling cell walls, brassinosteroids reduce epidermal constraint, showing how genes can control
U. gibba dwarf has twisted internal tissue
growth coordination between layers by means of mechanics.

p
U. gibba is a carnivorous plant with a spiral

M
any multicellular organisms are formed
from multiple cell layers, raising the
question of how growth is coordinated
between layers to produce an integrated
produce axial growth. There are no growth
conflicts between cells. However, if the epi-
dermal walls (Fig. 1B, purple) yield less to
turgor (e.g., because they are thicker or less
vegetative growing tip, comprising an apex
that produces stolons bearing filiform leaves
and traps (12) (Fig. 2A). The stolons and leaves
have internal air spaces that allow the plant to
float just below the water surface. To obtain
developmental mutants in U. gibba, we carried

final form. In plants, evidence from
genetic chimeras and from layer-specific modi-
fication of gene function shows that genes
active in one layer can act nonautonomously
to influence growth in other layers (1–4).
extensible than inner walls), an epidermal
growth constraint is generated, and load is
transferred from inner to outer walls. Each
cell experiences mechanical stresses caused by
the cell’s own turgor (cell-autonomous stresses)
g
out ethyl methanesulfonate mutagenesis.
Obtaining large numbers of progeny proved
difficult because of poor seed set and germi-
nation rates. Rather than mutagenizing seed,

y
Nonautonomy could be explained through and by mechanical effects from surrounding we therefore mutagenized small stolon explants
chemical signaling between layers and/or tissue (non–cell-autonomous stresses) (7) called and grew each on to flowering (see methods in
mechanical interactions. tissue stresses (5). Whereas the cell-autonomous the supplementary materials for details). M1
Mechanics may act nonautonomously through stress is always tensile, by our definition, tis- seed was collected from 441 explants and gave
the generation of tissue stresses (5), as demon-
strated experimentally by Hofmeister more
than 150 years ago (6). To understand the
origin of tissue stresses, consider a cylindrical
tissue in which cells are tightly stuck together,
with all cells having the same size, turgor, wall

y g
material properties, and wall thickness (Fig.
1A). If cell walls are anisotropic, such that they
yield more readily in the vertical orientation,
the vertical component of turgor forces in each
cell can cause stresses (highlighted for three

,
cells with black double-headed arrows) that

1
Department of Cell and Developmental Biology, John Innes
Centre, Norwich NR4 7UH, UK. 2Department of Biochemistry
and Metabolism, John Innes Centre, Norwich NR4 7UH, UK.
3
National Centre for Plant Gene Research (Beijing), Institute
of Genetics and Developmental Biology, Chinese Academy of
Sciences, Beijing 100101, China. 4College of Advanced
Agricultural Sciences, University of Chinese Academy of
Sciences, Beijing 100039, China. 5State Key Laboratory of
Plant Cell and Chromosome Engineering, Institute of Fig. 1. Origin of tissue stresses. (A) Cross section of a stem, with epidermal cells in gray and cell walls
Genetics and Developmental Biology, Chinese Academy of in black. All cell walls have the same material properties. Vertical component of turgor forces autonomous to
Sciences, Beijing 100101, China. 6Department of
each cell causes stresses (highlighted for three cells with black double-headed arrows) that produce
Computational and Systems Biology, John Innes Centre,
Norwich NR4 7UH, UK. 7Sainsbury Laboratory, University of axial growth. There are no growth conflicts between cells. (B) If the epidermal walls (purple) yield less to
Cambridge, Cambridge CB2 1LR, UK. turgor, an epidermal growth constraint is generated. Each cell now experiences two types of stress: cell-
*Corresponding author. Email: robert.bellow@jic.ac.uk (R.K.-B.); autonomous stress caused by the cell’s own turgor (black double-headed arrows) and tissue stress caused
richard.smith@jic.ac.uk (R.S.S.); chris.whitewoods@slcu.cam.ac.uk
(C.D.W.); enrico.coen@jic.ac.uk (E.C.) by mechanical effects from surrounding tissue. Tissue stresses can be tensile (divergent red arrows) or
†These authors contributed equally to this work. compressive (convergent blue arrows).

Kelly-Bellow et al., Science 380, 1275–1281 (2023) 23 June 2023 1 of 7

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

M2 phenotypes including altered traps, absent

traps, reduced leaf and stolon growth, long
flower spurs, spiky leaves, multiple traps on
leaves, and fasciation. One M2 family con-
tained two dwarf plants, and self-seed from a
wild-type sib gave 37 wild-type, 9 dwarf, and
3 extreme-dwarf plants (Fig. 2, A to C), con-
sistent with segregation of two recessive muta-
tions: dwarf and enhancer of dwarf.
Both the dwarf and extreme-dwarf plants
had short internodes, short leaves, and small
traps (Fig. 2, A to D). To follow their devel-
opment, we numbered internodes sequentially
relative to the spiral apex, with internode 1 cor-
responding to the first clearly visible in-
ternode to emerge from the apex (fig. S1).
Wild-type internode length increased until
about internode 4, after which it plateaued
to give a mature internode length of ~2 mm
(Fig. 2E). By contrast, dwarf and extreme-dwarf
plants exhibited very little growth after inter-

p
node 1, generating mature internode lengths
of ~0.7 and ~0.3 mm, respectively (Fig. 2E).
Epidermal cells of mutant stolons were shorter
and smaller than those of wild type (Fig. 2, F
to J). Measurements of cell lengths parallel Fig. 2. External phenotype of U. gibba wild type and dwarf mutants. (A to C) U. gibba vegetative
to the stolon axis indicated that 70% of the plants comprise a spiral apex (ap), filiform leaves (l), stolons (st), and traps (t). (A) Wild type. (B) Dwarf.

reduction in dwarf internode length was
caused by reduced longitudinal growth after
cell division arrest (fig. S2A). Further reduc-
tion in internode length in extreme dwarf
was caused by reduced growth before di-
g
(C) Extreme dwarf. Scale bar, 1 mm. (D) Violin plots of wild type [ x = 3.07 mm ± 0.11 (SEM), n = 10], dwarf
[x = 0.72 mm ± 0.02 (SEM), n = 10], and extreme dwarf [x = 0.29 mm ± 0.01 (SEM), n = 13] mature internode
lengths from plants grown in continuous culture. Block indicates interquartile range, and horizontal line
the mean. Both mutants have reduced lengths compared with wild type (***P < 0.001). (E) Internode

vision arrest. In addition to reduced internode
length, both dwarf and extreme dwarf ex-
hibited a significant increase in stolon cir-
cumference and number of epidermal cells
in transverse sections compared with wild
type, indicating increased radial and circum-
ferential growth before division arrest (fig.
S2, B and C).
We next determined the phenotype of inter-
nal tissues. Wild-type stolons had a cylindrical
y
lengths from growing explants of wild type (red), dwarf (orange) and extreme dwarf (blue) plotted
against internode number. Dashed line shows mean from internode 10 onwards (n > 4 plants). (F to H) Cell
areas in mature stolons of wild type (F), dwarf (G), and extreme dwarf (H), color coded according to the
color scale shown at the left of (F). Scale bar, 100 mm. (I and J) Violin plots of cell area (I) and cell anisotropy
[cell maximum length/(cell maximum length + cell minimum length)] (J) of mature stolons of wild type
(n = 1817 cells from eight explants), dwarf (n = 2289 cells from five explants), and extreme dwarf (n = 1494
cells from four explants). Both mutants have significantly lower values than wild type.
as in wild type (fig. S3A). Twisted vascular tissue growth using continuum mechanics.
tissue in dwarf plants was only observed from For these purposes, we distinguished between

epidermis (Fig. 3C, purple) connected by five
or six straight “blades” (cyan) to an axial cylinder
of large cells (yellow) surrounding a vascular
bundle (orange), with air spaces (magenta) be-
tween the blades (Fig. 3, A to C). Dwarf stolons
internode 4 onward (Fig. 3K, orange line). Con-
tortions of the blade were evident in dwarf
mutants earlier, at internode 1, as tissue strips
running perpendicular to the vascular axis in
longitudinal sections (Fig. 3J, blue arrow).
y g
two types of regional growth: specified and
resultant (13). Specified growth corresponds
to the growth driven by a cell’s own turgor,
in mechanical isolation from other cells. Re-
sultant growth corresponds to the growth

had smaller air spaces, twisted blades, and a
sinuous contorted vascular bundle (Fig. 3, D
to F). Extreme-dwarf stolons had smaller air
spaces and less-twisted vasculature than dwarf
plants (Fig. 3, G to I). Both dwarf and extreme-
dwarf plants sank in water, presumably be-
cause of their reduced air spaces.
The twisted internal tissue of the dwarf
plants might be caused by a contorted pattern
of early vascular and blade cell-type specifica-
tion or by altered tissue growth after specifi-
cation has occurred. To distinguish between
these possibilities, we determined the devel-
opmental timing of the twisted phenotype in
dwarf plants. Straight vasculature cell types
surrounded by blade and air spaces were
evident at internodes 0 in dwarf (Fig. 3J),
These blade contortions were only seen after
air spaces had formed. Thus, contortion and
twisting of internal tissue in dwarf plants
arose through altered growth after cell type
specification and air space formation, lead-
ing to excess vascular length compared with
epidermal length (fig. S2D). In extreme-dwarf
plants, which showed little contortion, orga-
nized vasculature and surrounding tissue were
evident in early internodes, but air spaces
were not (fig. S3B).
Twisted dwarf phenotype explained by
epidermal constraint
To evaluate hypotheses that might account
for both the internal twisting and shortened
internode length of dwarf mutants, we modeled
,
generated when tissue stresses, which act
non–cell-autonomously, are also factored in.
Computational models allow tissue stresses
and resultant growth to be calculated from
an input pattern of specified growth rates and
orientations.
We modeled a small length of U. gibba
stolon as a stiff cylindrical epidermal sheet
connected by blades to an axial core (Fig. 4,
A to C, and fig. S9, A and B). Specified growth
was oriented parallel to an axial (initially vertical)
polarity field (Fig. 4A, arrows). To reduce
boundary effects, each stolon end was con-
strained to remain flat and horizontal.
If all regions had the same specified growth
rate, the cylinder elongated without genera-
tion of tissue stresses or twisting of internal

Kelly-Bellow et al., Science 380, 1275–1281 (2023) 23 June 2023 2 of 7

RES EARCH | R E S E A R C H A R T I C L E
Fig. 3. Internal phenotype of U. gibba wild type and dwarf mutants. Wild type (A to C), dwarf (D to F),
and extreme dwarf (G to I) longitudinal confocal sections [(A), (D), and (G)], freeze-fracture scanning
electron microscopy [(B), (E), and (H)], and toluidine blue–stained transverse sections [(C), (F), and
(I)]. Arrows and cells are color coded purple for epidermis (e), cyan for blades (b), magenta for air
spaces (s), yellow for axial core (a), and orange for vasculature (v). Scale bars, 50 mm. (J and K) Confocal
Z-slice of early dwarf internodes 0 to 4. Scale bars, 100 mm.
tissue (Fig. 4, D to F). If specified growth rate
was set to zero in the epidermis, the epidermal
constraint caused a dwarf phenotype (Fig. 4, G
and H). Tissue tension was generated in the
epidermis (Fig. 4I, red), and tissue compres-
sion was generated in the blades and core (Fig.
4, I and J, blue). The tissue tension caused the
epidermis to grow to some extent, despite
its specified growth rate being zero [com-
pare epidermal resultant growth rate (Fig.
4K) with specified growth rate (Fig. 4G)].
Conversely, tissue compression in blades and
core caused a lower resultant growth rate than
Kelly-Bellow et al., Science 380, 1275–1281 (2023)
specified (compare Fig. 4L with Fig. 4H), but
higher than in the epidermis (zero). The tissue
stresses also caused twisting of blades and
core (Fig. 4, M to P, and movie S1). Thus, re-
duced specified growth rate in the epidermis
captured both the dwarf phenotype and inter-
nal contortion.
Twisting of the axial core still occurred
when blades were removed from a middle
segment of the cylinder (fig. S4, A to F),
showing that tissue compression could be
transmitted to the core from above and below.
No twisting occurred if the cylinder was solid
23 June 2023
(fig. S4, G to K), showing that air spaces were
needed to accommodate buckling, and ac-
counting for the reduced twisting observed
in extreme dwarf plants. Restricting specified
growth to the axial core led to a dwarf pheno-
type and a sinuous core but little twisting of
the blades (fig. S4, L and M). Restricting spec-
ified growth to the blades gave a dwarf pheno-
type with twisted blades but little twisting
of the core (fig. S4, N to P). Radial specified
growth of the blades led to blade twisting,
but cylinder elongation and axial core straight-
ness were not affected (fig. S4, Q to V). Thus,
both the dwarfism and internal axial and blade
twisting could be most readily accounted for
by reduced specified growth rate of the epi-
dermis alone.
DWARF encodes a brassinosteroid
biosynthetic enzyme
To understand the molecular basis of the
p
dwarf mutant, we sequenced the wild-type
progenitor and 33 wild-type, 10 dwarf, and 3
extreme-dwarf segregants. Only one single-
nucleotide polymorphism (SNP) was absent
from the progenitor, heterozygous or absent
in wild-type segregants, and homozygous in
g
all mutants, indicating that it was located in
the DWARF gene. Extreme-dwarf plants carried
four additional SNPs absent from the progeni-
tor (table S1) that were candidate mutations in
ENHANCER OF DWARF. Plants homozygous
y
for enhancer of dwarf and heterozygous or
homozygous for DWARF were scored as wild
type, suggesting that the enhancer of dwarf
mutation alone did not have a strong pheno-
typic effect. However, the mutation may have
caused a subtle phenotype that we missed
when initially scoring the families.
The DWARF SNP introduced an early stop
codon in a gene encoding a cytochrome P450
90B1 enzyme, which catalyzes the C22-alpha-
y g
hydroxylation step in the brassinosteroid bio-
synthesis pathway (14). This gene is homologous
to DWARF4 (DWF4) in Arabidopsis, which
affects cell area and cell anisotropy in a similar
way to U. gibba DWARF (15, 16). Brassinosteroid
,
precursors after the C22-alpha-hydroxylation
step were undetectable or at a low level in
dwarf mutants, whereas a precursor before the
step was present (fig. S5). Inhibiting brassi-
nosteroid biosynthesis in wild type using brassi-
nazole led to short stolons, smaller cells, and
contorted vasculature, similar to dwarf mutants
(fig. S6). Adding brassinosteroid, by growing
mutants in epibrassinolide, rescued dwarf
plants and partially rescued extreme dwarf
plants (fig. S6). Thus, DWARF likely encodes a
brassinosteroid biosynthesis gene.
To determine the timing of brassinosteroid
action, we tracked dwarf stolons after treat-
ment with epibrassinolide (Fig. 4S). Inter-
nodes that were not readily visible when the
treatment began, because they were concealed
3 of 7
RES EARCH | R E S E A R C H A R T I C L E

in wild type. However, we cannot rule out the

possibility that internodes above 5 are imper-
meable to exogenous brassinosteroid.

Arabidopsis brassinosteroid mutant has

elevated tissue stresses
Our experimental and modeling results indi-
cate that brassinosteroid promotes U. gibba
stolon growth starting just before cell division
arrest by counteracting an epidermal con-
straint, thus reducing tissue stresses. If gener-
ally applicable, this hypothesis predicts that
Arabidopsis dwf4 mutants should also exhibit
elevated tissue stresses. However, the effect
of these stresses might be masked because
Arabidopsis stems are solid and therefore lack
air spaces to accommodate buckling (fig. S4, G
to K). To determine whether tissue stresses are
enhanced in dwf4 mutants, we therefore ex-
ploited the quasimodo2-1 (qua2-1) mutation,
which weakens cell-cell adhesion (17). As illus-

p
trated in Fig. 1B (red arrows), tissue stresses
generated a force that acts to pull epidermal
cells apart. In wild-type Arabidopsis, cell-cell
adhesion is strong enough to resist this force,
but in qua2-1 mutants, epidermal cracks are
observed between cells in dark-grown hypo-

g
cotyls, confirming that epidermal tissue tension
is present (9, 18). If brassinosteroid normally
acts to reduce tissue tension, cracks are pre-
dicted to be exacerbated in qua2-1 dwf4 double
mutants, or in qua2-1 mutants treated with a

y
brassinosteroid inhibitor.
To test these predictions, we intercrossed
dwf4 and qua2-1 mutant lines. About 1/16
(58/885) of the F2 dark-grown seedlings ex-
hibited a distinctive novel phenotype: Hypocotyls
were dwarf and seemed devoid of epidermis
(Fig. 5, A and B), unlike qua2-1 single mutants,
which showed small epidermal cracks at a
similar stage (Fig. 5C). To clarify the devel-
opmental origin of the double-mutant pheno-

y g
Fig. 4. Simulations of U. gibba wild type and dwarf mutant and timing of brassinosteroid action.
type, we imaged seedlings at different days
(A) Initial state for wild-type and dwarf mutant models [epidermis (purple), blades (cyan), and axial core (yellow)].
after germination. Seedlings of dwf4 qua2-1 were
Arrows indicate polarity. (B) Initial state without epidermis. (C) Transverse slice of (A). (D) Final state of wild-
indistinguishable from dwf4 seedlings until
type simulation, color coded for specified growth rate, which is uniformly high. (E) Same as (D), but color coded for
~3 days after stratification, when wide cracks
tissue type. (F) Same as (E), with epidermis removed. (G) Final state for simulation of dwarf mutant, color coded
appeared in the double mutant (Fig. 5D). These
for specified growth rate, which is excluded from the epidermis and gives reduced elongation. (I) Same as (G), but

,
cracks were much larger than those observed in
color coded for tissue stresses. (K) Same as (G), but color coded for resultant growth rate. (M) Same as (G),
qua2-1 single mutants at the same stage (Fig.
but color coded for tissue type. (H, J, L, and N) Same as (G), (I), (K), and (M), respectively, but with epidermis
5E). By 5 days after stratification, the cracks in
clipped away. (O) Transverse slice of (M). (P) Same as (M), but showing the axial core only. (Q) Color scale for
dwf4 qua2-1 had enlarged to the extent that
specified and resultant growth rates, in strain per time step of simulation. (R) Color scale for tissue stresses, with red
much of the epidermis was no longer evident
indicating tension (t) and blue indicating compression (c). (S) A dwarf explant imaged on days 0 and 14 after
(Fig. 5B). Crack formation was also enhanced
treatment with 0.01 mM epibrassinolide. Internode numbers labeled on day 14. Scale bar, 5 mm. (T) Average
when qua2-1 single mutants were grown in
internode lengths of dwarf explants before treatment (day 0, solid orange line) and at 14 days (day 14, solid brown
the presence of a brassinosteroid inhibitor,
line) (n ≥ 10 explants). Day 14 internode −10 to 0 lengths were not significantly different from the mean of Fig. 2E
brassinazole (Fig. 5, F to H). These results thus
(red dashed line). Error bars show SEM.
support the hypothesis that brassinosteroid
promotes stem growth by counteracting an
within the spiral vegetative shoot tip or had response to treatment (P < 0.05), with the epidermal constraint.
not yet initiated, were assigned consecutive magnitude of the increase declining with in- To further validate this interpretation, we
negative numbers, starting from 0. These in- ternode number. Thus, brassinosteroid likely modeled the growth of a solid cylinder with a
ternodes grew to a length similar to those of acts from around internode 0, when cell divi- stiff epidermis in which cracks can form when
mature wild type (Fig. 4T). Internodes 1 to 5 sion is nearing arrest, until around internode 5, tension exceeds a threshold value. Uniform
also showed a significant length increase in by which stage cell elongation has arrested specified growth rate gave elongation without

Kelly-Bellow et al., Science 380, 1275–1281 (2023) 23 June 2023 4 of 7

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

p
g
y
y g
,

Kelly-Bellow et al., Science 380, 1275–1281 (2023) 23 June 2023 5 of 7

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

Fig. 5. Effect of reduced brassinosteroid on epidermal cracks in
Arabidopsis qua2-1 and explanatory computational models. (A to C)
Confocal images 5 days after stratification. (A) qua2-1 dwf4 double mutant.
(B) qua2-1 dwf4 shown in (A), with cells artificially colored for clarity (purple
indicates epidermal cells, and cyan indicates interior cells). It is possible that the
cyan-colored cells include some disorganized epidermal cells. Close-up shown on
the right. (C) qua2-1 single mutant with a close up of a region with cracks.
(D and E) Confocal images 3 days after stratification. (D) qua2-1 dwf4 double
mutant. (E) qua2-1 single mutant. Scale bars, 100 mm. Purple arrows highlight
curled free ends of epidermal cells. (F to H) Phenotypes of qua2-1 Arabidopsis
hypocotyls without or with treatment with brassinosteroid inhibitor [1 mM
brassinazole (BRZ)]. [(F) and (G)] Confocal images of seedlings after 9 days
growth in the dark. (F) qua2-1 with close up of a region with cracks selected for
magnification. (G) qua2-1 grown on 1 mM BRZ with close up. Curved cells at
crack boundaries are indicated with an arrow. Epidermal cells are in purple,
internal cells are in cyan. Scale bars, 100 mm except in (F) (gray scale bar),
which is 1000 mm. (H) Violin plots of crack widths measured by number of cells.
Mean crack width covers more cell files in qua2-1 + BRZ [ x = 4.089 ± 0.38
(SEM)] than qua2-1 untreated [ x = 1.611 ± 0.14 (SEM)], but not significantly
more (P = 0.0531, untreated hypocotyls n = 74 cracks from two plants,
BRZ-treated hypocotyls n = 83 cracks from six plants). (I to L) Tissue-level
computer simulations. (I) Initial state, with epidermis in purple and inner regions
in cyan. Arrows indicate polarity. (J) Final state with all regions having the same
specified growth parallel to polarity, leading to elongation without epidermal
cracking. (K) Final state with reduced specified growth in the epidermis leads to
a shorter cylinder and epidermal cracks. (L) Longitudinal section through (K),
showing longitudinal tissue tension (t) in red and compression (c) in blue.
(M to Z) Cellular-level computer simulations. (M) Initial state, with outer epidermal
wall in dark purple, epidermis in purple, and inner tissue in cyan. [(N) to (Y)] Final
state. [(N) to (Q)] All cell walls have the same thickness and material properties. [(R)
to (U)] Outer epidermal wall is 10 times thicker. [(V) to (Y)] Outer wall is 10 times
thicker, with higher extensibility and reduced yield threshold. [(Q), (U), and (Y)]
Epidermal fracture introduced at an early stage. White arrow in (Q) indicates position
of fracture. Fracture is two cells wide for (Q) and (U) and eight cells wide for (Y).
[(N), (R), and (V)] Specified growth rate. [(O), (S), (W), (Q), (U), and (Y)] Tissue
stresses. [(P), (T), and (X)] Resultant longitudinal wall stresses. (Z) Color scales. (Top)
Specified growth rate (7% per time step). (Middle) Tissue stresses (23). (Bottom)
Longitudinal wall stresses (23). Scale bar, 50 mm.

tissue stresses or cracks (Fig. 5, I and J), whereas
low epidermal specified growth led to reduced
elongation, elevated tissue stresses, and crack
formation (Fig. 5, K and L).
Release from epidermal constraint by
wall remodeling
verting a proportion of reversible elastic wall
strain, above a yield threshold, into irreversible
strain at each time step. The proportion corre-
sponded to the extensibility of the wall. Walls
were seven times stiffer (a sevenfold larger
Young’s modulus) in the transverse compared
p
because the thick outer wall grows more slowly)
(Fig. 5Y), capturing the qua2-1 phenotype (Fig.
5C). Thus, brassinosteroid likely acts, at least
in part, by loosening of the thick outer wall,
counteracting the epidermal constraint.
In addition to wall thickness, epidermal

The above results raise the question of how
brassinosteroid reduces epidermal constraint.
The most obvious source of epidermal con-
straint is the thick outer wall of the epidermal
with longitudinal orientation, leading to ver-
tical specified growth.
If all cell walls had the same material prop-
erties, the cylinder elongated with uniform
specified growth rates (Fig. 5N), low tissue
g
constraint may be further enhanced by the
orientation of microfibrils, which are less
transverse in outer than in inner walls for
wild-type Arabidopsis hypocotyls (23). Brassi-
nosteroid treatment can cause microtubules

cells (6). A constraining outer wall is also con-
sistent with the concave shape of the outer
wall in epidermal cells released by crack for-
mation (Fig. 5, C to E, purple arrows). Outer
epidermal walls of U. gibba dwarf stolons
were about two to three times thicker than
inner walls at internode 1 (fig. S7, D and F),
by which time growth had ceased (Fig. 2E).
Outer epidermal walls of A. thaliana dwf4
dark-grown hypocotyls were ~20 times thicker
stresses (Fig. 5O), and uniform longitudinal
wall stresses (Fig. 5P). Introducing an epi-
dermal fracture caused rounding of the cell
ends but did not cause further cell separa-
tion (Fig. 5Q, arrow indicates the position of
fracture). Setting the outer epidermal wall to be
10 times thicker than the inner walls lowered
the epidermal specified growth rate (Fig. 5R).
The growth constraint generated longitudinal
epidermal tissue tension and internal tissue
y
of the outer epidermal plasma membrane to
orient more transversely (24, 25). Thus, brassi-
nosteroid may reduce epidermal constraint by
remodeling the thick outer wall in two ways:
wall loosening and reducing the proportion of
longitudinally oriented microfibrils. Such an
effect on microfibril orientation might explain
why Utricularia dwarf mutants have wider
stolons (fig. S2, B and C). The differential
properties of the outer epidermal wall (e.g.,

than inner walls at 4 days after stratification
(fig. S7, C and E), by which time growth had
largely ceased (fig. S8). Similar wall thicknesses
were observed for wild types at comparable
stages (fig. S7, A, B, E, and F), even though
compression (Fig. 5S). Resultant longitudinal
wall stresses were uniform but lower than if
all cell walls had the same material properties
(compare Fig. 5T with Fig. 5P). The cylinder
therefore grew less, capturing the dwf4 pheno-
y g
thickness, extensibility, microfibril orientation)
may depend on cell polarity factors that confer
differences between outer and inner cell faces
(26–28). Brassinosteroids may also reduce
epidermal constraint by increasing turgor in

growth continued afterwards, suggesting that
brassinosteroid does not reduce epidermal
constraint primarily by altering wall thickness.
A possible mechanism for reduction of
epidermal constraint is wall loosening. Brassi-
nosteroids promote hypocotyl elongation with-
in 6 hours of application through increased
wall relaxation properties (i.e., wall loosening)
(19, 20), possibly through phosphorylation of
plasma membrane proton adenosine triphos-
phatase (21). To explore the possible contribu-
tion of wall loosening, we modeled hypocotyl
tissue growth at the cellular level. A segment
of hypocotyl was modeled as a vertical cylinder
of tightly attached cells of similar size and
under the same turgor (Fig. 5M). Wall growth
by means of creep (22) was simulated by con-
type. Introducing a wide epidermal fracture
(eight cells wide) released epidermal cell ends
to peel back (Fig. 5U and movie S2), capturing
the dwf4 qua2-1 phenotype (Fig. 5D).
To simulate wild type, the thick outer wall
was loosened by increasing its extensibility
and reducing its yield threshold. This modi-
fication increased the epidermal specified
growth rate (compare Fig. 5V with Fig. 5R),
lowered tissue stresses (compare Fig. 5W with
Fig. 5S), and raised longitudinal wall stresses
of internal tissue (compare Fig. 5X with Fig. 5T).
The cylinder therefore elongated more than in
the dwf4 simulation, capturing the wild-type
phenotype. Introducing a narrow epidermal
fracture (two cells wide) led to released epi-
dermal cell ends peeling back (they curve
,
epidermal cells, although there is currently no
experimental evidence to support this possibility.
Conclusions
When brassinosteroid synthesis or perception
genes are expressed only in the epidermal cell
layer of Arabidopsis brassinosteroid mutants,
a near–wild-type phenotype is generated, even
though these genes are normally expressed in
both epidermis and ground tissue (4, 29). Our
results indicate that this nonautonomous effect
of epidermal brassinosteroid gene expression
on resultant growth of internal tissue involves
release of internal tissue from epidermal me-
chanical constraint, although this does not
preclude additional contributions from mo-
lecular signaling. Mechanical interactions

Kelly-Bellow et al., Science 380, 1275–1281 (2023) 23 June 2023 6 of 7

RES EARCH | R E S E A R C H A R T I C L E
between cell layers also play a role in animal
development, such as formation of crocodile
skin cracks (30) and intestinal villi (31). Here
we show how genes may modify tissue layer
interactions by changing cellular growth prop-
erties and thus tissue stresses. Gene activity
may therefore have coordinated effects on
tissue development not only via molecular sig-
naling but also via mechanics.
RE FE RENCES AND N OT ES
1. S. Hake, B. R. Char, Genes Dev. 11, 1087–1097 (1997).
2. M. H. Frank, D. H. Chitwood, Dev. Biol. 419, 41–53 (2016).
3. R. A. Tilney-Bassett, Plant Chimeras (Edward Arnold Publishers
Ltd., 1986).
4. S. Savaldi-Goldstein, C. Peto, J. Chory, Nature 446, 199–202
(2007).
5. Z. Hejnowicz, A. Sievers, J. Exp. Bot. 46, 1035–1043
(1995).
6. U. Kutschera, K. J. Niklas, J. Plant Physiol. 164, 1395–1409
(2007).
7. F. Boudon et al., PLOS Comput. Biol. 11, e1003950 (2015).
8. W. S. Peters, A. D. Tomos, Ann. Bot. 77, 657–665 (1996).
9. S. Verger, Y. Long, A. Boudaoud, O. Hamant, eLife 7, e34460
(2018).
10. Z. Hejnowicz, A. Sievers, J. Exp. Bot. 46, 1045–1053 (1995).
11. U. Kutschera, J. Plant Physiol. 138, 460–465 (1991).
12. P. Taylor, The Genus Utricularia: A Taxonomic Monograph
(Royal Botanic Gardens, 1989).
13. E. Coen, R. Kennaway, C. Whitewoods, Development 144,
4203–4213 (2017).
14. S. Fujioka et al., Biosci. Biotechnol. Biochem. 59, 1543–1547 (1995).
Kelly-Bellow et al., Science 380, 1275–1281 (2023)
15. R. Azpiroz, Y. Wu, J. C. LoCascio, K. A. Feldmann, Plant Cell 10,
219–230 (1998).
16. S. Choe et al., Plant Cell 10, 231–243 (1998).
17. S. Verger, S. Chabout, E. Gineau, G. Mouille, Development 143,
2536–2540 (2016).
18. G. Mouille et al., Plant J. 50, 605–614 (2007).
19. D. M. Zurek, D. L. Rayle, T. C. McMorris, S. D. Clouse,
Plant Physiol. 104, 505–513 (1994).
20. T. W. Wang, D. J. Cosgrove, R. N. Arteca, Plant Physiol. 101,
965–968 (1993).
21. A. Minami, K. Takahashi, S. I. Inoue, Y. Tada, T. Kinoshita,
Plant Cell Physiol. 60, 935–944 (2019).
22. E. Coen, D. J. Cosgrove, Science 379, eade8055 (2023).
23. E. F. Crowell et al., Plant Cell 23, 2592–2605 (2011).
24. X. Wang et al., Plant Cell 24, 4012–4025 (2012).
25. M. Catterou et al., Planta 212, 673–683 (2001).
26. J. Takano et al., Proc. Natl. Acad. Sci. U.S.A. 107, 5220–5225
(2010).
27. L. Łangowski, K. Růzicka, S. Naramoto, J. Kleine-Vehn, J. Friml,
Curr. Biol. 20, 904–908 (2010).
28. V. Gorelova, J. Sprakel, D. Weijers, Nat. Plants 7, 1548–1559
(2021).
29. D. M. Friedrichsen, C. A. Joazeiro, J. Li, T. Hunter, J. Chory,
Plant Physiol. 123, 1247–1256 (2000).
30. M. C. Milinkovitch et al., Science 339, 78–81 (2013).
31. A. E. Shyer et al., Science 342, 212–218 (2013).
32. R. Kennaway, JIC-Enrico-Coen/Cell_Layer_Interactions_2023:
GFtbox_Cell_Layer_Interactions_2023 (v1.0.1), Zenodo (2023);
https://doi.org/10.5281/zenodo.7913146.
ACKN OWLED GMEN TS
We thank B. and P. Steward at The Fly Trap Plants and T. Bailey
from the Carnivorous Plant Society for plants, seeds, and advice.
We thank M. Majda for qua2-1 seeds, E. Wegel and S. Lopez of
John Innes Centre (JIC) Bioimaging for help with light microscopy,
R. Wightman for help with freeze-fracture scanning electron
23 June 2023
microscopy, L. Perkins and the JIC horticulture team for large-
scale U. gibba cultivation, G. Mosca for MorphoMechanX, and
D. Bradley for critical reading of the manuscript. Funding: This
work was supported by a European Research Council grant
(323028-CarnoMorph) and Biotechnology and Biological Sciences
Research Council grants (BBS/E/J/000PR9787, BB/M023117/1,
BB/L008920/1, and BB/X01102X/1) awarded to E.C. Author
contributions: Biological experiments, data analysis, and
conceptualization: R.K.-B., K.L., J.E.B., M.Y., P.B., B.K., S.C., J.C.,
T.X., B.L., J.F., Y.X., R.S., and C.D.W. Computational modeling: R.K.,
R.S.S., and E.C. Software development: R.K. and R.S.S., Development
of U. gibba resources: K.L., C.B., J.S., M.Y., and C.D.W. Bioinformatic
analysis: R.K.-B. and A.W. Supervision, funding acquisition, and
conceptualization: E.C. Competing interests: The authors declare
that they have no competing interests. Data and materials
availability: All data are available in the manuscript or the
supplementary materials or are deposited at Zenodo (32). License
information: Copyright © 2023 the authors, some rights reserved;
exclusive licensee American Association for the Advancement of
Science. No claim to original US government works. https://www.
science.org/about/science-licenses-journal-article-reuse
SUPPLEMENTARY MATERIALS
science.org/doi/10.1126/science.adf0752
Materials and Methods
Supplementary Text
Figs. S1 to S9
p
Tables S1 and S2
References (33–53)
MDAR Reproducibility Checklist
Movies S1 and S2
View/request a protocol for this paper from Bio-protocol.
Submitted 5 October 2022; accepted 18 May 2023
g
10.1126/science.adf0752
y
y g
,
7 of 7
RES EARCH

ARCTIC ECOLOGY tundra, both of which are associated with

declines in abundance of reindeer and caribou
Large herbivore diversity slows sea ice–associated (both Rangifer tarandus) (12, 16). Such conse-
quences may be reciprocally important to tundra
decline in arctic tundra diversity vegetation dynamics because large herbivores
influence patterns of vegetation abundance
Eric Post1*, Elina Kaarlejärvi2, Marc Macias-Fauria3, David A. Watts4, Pernille Sporon Bøving1, and community composition across multiple
Sean M. P. Cahoon5, R. Conor Higgins1†, Christian John1‡, Jeffrey T. Kerby6,7, Christian Pedersen8, spatial and temporal scales in the Arctic (17, 18).
Mason Post9, Patrick F. Sullivan10 Moreover, vertebrate herbivores mediate nu-
merous vegetation responses to climate change
Biodiversity is declining globally in response to multiple human stressors, including climate forcing. (19), including plant diversity loss under warm-
Nonetheless, local diversity trends are inconsistent in some taxa, obscuring contributions of local processes ing (20) (Fig. 1). Accordingly, conservation of
to global patterns. Arctic tundra diversity, including plants, fungi, and lichens, declined during a 15-year herbivores and rewilding of lost herbivore
experiment that combined warming with exclusion of large herbivores known to influence tundra vegetation assemblages have emerged as novel mitiga-
composition. Tundra diversity declined regardless of experimental treatment, as background growing tion strategies for effects of climate change in
season temperatures rose with sea ice loss. However, diversity declined slower with large herbivores many biomes, including the rapidly warming
than without them. This difference was associated with an increase in effective diversity of large herbivores arctic tundra (19, 21–23).
as formerly abundant caribou declined and muskoxen increased. Efforts that promote herbivore diversity, To investigate interactive effects of warming
such as rewilding, may help mitigate impacts of warming on tundra diversity. and herbivory on community composition of
arctic tundra, including plants, fungi, and

W
lichens, we initiated a large-scale herbivore

ithout efforts to mitigate carbon emis-
sions, 2°C warming above the preindus-
trial baseline is expected to precipitate
rapid extinctions across multiple taxa
and biomes (1, 2). Notably, the rela-
tively species-poor Arctic is already 2°C warmer
may alter biodiversity in adjacent terrestrial
systems through effects on temperature and
precipitation that influence the abundance
and distribution of local species (13–15). In-
directly, sea ice decline has also been linked
to rain-on-snow events and shrubification of
p
exclusion and warming experiment in 2002
near Kangerlussuaq, Greenland (24) (figs. S1
and S2). The experiment ran continuously for
15 years (concluding in 2017) and included
annual nondestructive sampling of tundra
community diversity (24). The site is one of

g
than this baseline seasonally and will exceed
this threshold annually decades sooner than
anywhere else on Earth (3). Despite such rapid
warming, plant diversity responses to climate
change in the Arctic are ostensibly incon-

y
sistent and difficult to predict (4, 5), perhaps A B
in part because of heterogeneity in abiotic
conditions including soil moisture and nu-
trient availability across the Arctic and the
limitations this imposes on tundra plant re-
sponses to warming (6).
Although warming itself is a major driver of
ecological change in the Arctic (7), associated
declines in sea ice extent and seasonal melt
onset (8–10) also pose direct and indirect con-

y g
sequences, including threats to Indigenous
livelihoods based on wildlife harvest (11) and
husbandry (12). For instance, sea ice decline

1
Department of Wildlife, Fish, and Conservation Biology,

,
University of California Davis, Davis, CA 95616, USA.
2
Research Center for Ecological Change, Organismal and
Evolutionary Biology Research Programme, University of
Helsinki, Helsinki, Finland. 3School of Geography and the Fig. 1. Conceptualization of potential interactions between warming and large herbivore diversity and
Environment, University of Oxford, Oxford, UK. 4Alaska State abundance in arctic tundra vegetation dynamics motivating this study. Both panels depict arctic tundra
Virology Laboratory, Division of Public Health, Alaska under climatic warming. (A) represents a scenario in which warming occurs with an abundant and diverse
Department of Health, Fairbanks, AK 99775, USA. 5Pacific
Northwest Research Station, USDA Forest Service, Anchorage,
assemblage of large herbivores (in this case caribou and muskoxen) and (B) represents a scenario in which
AK 99501, USA. 6Aarhus Institute of Advanced Studies, Aarhus warming occurs with a sparse and species-poor assemblage of large herbivores. The arrows indicate differences in
University, Aarhus, Denmark. 7Section for Ecoinformatics and arctic tundra composition under warming that might arise with a decline in large herbivore diversity and
Biodiversity, Department of Biology, Aarhus University, Aarhus
C, Denmark. 8Department of Landscape Monitoring, Norwegian
abundance (upper arrow) or an increase in large herbivore diversity and abundance (lower arrow). Hence, from
Institute of Bioeconomy Research, 1431 Ås, Norway. (A) to (B) a lower or declining diversity or abundance of large herbivores, or their complete absence, may facilitate
9
Department of Genome Sciences and Brotman Baty Institute, a warming-induced increase in deciduous shrub height and abundance, decline in graminoids and forbs, and
University of Washington, Seattle, WA 98195, USA.
10 a consequent reduction in tundra diversity. Conversely, from (B) to (A) greater or increasing diversity or abundance
Environment and Natural Resources Institute, University of
Alaska Anchorage, Anchorage, AK 99508, USA. of large herbivores may constrain a warming-induced increase in deciduous shrub height and abundance, thereby
*Corresponding author. Email: post@ucdavis.edu maintaining tundra diversity by promoting the persistence of graminoids and forbs that might otherwise be
†Present address: Yolo County Resource Conservation District, outcompeted or interfered with by shrubs for space, light, or soil nutrients. This study combined long-term
Woodland, CA 95695, USA.
‡Present address: Marine Science Institute, University of California, experimental warming and herbivore exclusion with monitoring of background variation in abiotic conditions and
Santa Barbara, CA 93106, USA. herbivore diversity and abundance to investigate such interactions. [Artwork by SciStories]

Post et al., Science 380, 1282–1287 (2023) 23 June 2023 1 of 6

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

the few locations in the Arctic tundra biome
with multiple species of large herbivores (19),
caribou and muskoxen (Ovibos moschatus).
After the first five years of the experiment,
we reported that both caribou and muskoxen
influenced tundra community responses to
the experimental treatments (25). Ongoing
work at the site revealed associations between
sea ice extent and several components of the
system, including local spring warming (26)
and increasing shrub abundance (27). The
long-term nature of the experiment was there-
fore conducive to monitoring natural changes
in abiotic conditions and herbivore abundance
(24) presumed important to tundra community
composition and diversity dynamics (Fig. 1)
and ultimately presented an opportunity to
investigate their interactions with the experi-
mental treatments.
Patterns of tundra diversity decline
(Fig. 2 and tables S3 and S4). Peak growing
season (July) temperature effects on tundra di-
versity were negative across all experimental
treatments (tables S3 and S4), indicating a
decline in tundra diversity with local warming.
These associations varied only marginally
among treatments at both the plot and site
scales (tables S3 and S4, respectively), but
were significantly weaker under grazing than
under herbivore exclusion at the plot scale
(table S3).
Abiotic drivers of tundra diversity decline:
Relation of local weather to sea ice decline
Linking sea ice decline and local tundra com-
munity dynamics requires, at a minimum, ex-
amining associations among sea ice and local
abiotic conditions. During the experimental
period (2002 to 2017), Arctic mean annual tem-
perature increased by nearly 1°C (3), ASIE de-
clined by 1.34 million km2 (28), and peak annual
the study site increased by >1°C (29). Of the
candidate local abiotic predictors of tundra
diversity decline considered here (24), the only
ones significantly related to May-July ASIE
during the experimental period were mean
daily temperature and mean daily maximum
temperature in July, the latter of which was
also the single best local abiotic predictor of
variation in tundra diversity (tables S2 and
S5). Both of these variables were also correlated
with July ASIE (tables S5), the second-best in-
dividual predictor of variation in tundra di-
versity (table S2). Additionally, July mean and
maximum temperatures were more highly
correlated with ASIE than with BBSIE in both
July and May-July (table S6), and multivariate
models testing competing abiotic predictors (24)
revealed the dominant effect of May-July ASIE
(table S7). In all cases, July temperature variables
were negatively correlated with ASIE, suggesting
that declining spring and summer ASIE pro-

p
Effective species number of plants, fungi, and growing season (July) mean temperature at motes local July warming. Additional analyses
lichens (i.e., True Simpson’s diversity, here-
after tundra diversity) (24) declined across
the study site under all experimental treat-
ments as the 15-year experiment progressed
(Fig. 2, top row; see also fig. S3). However,

g
rates of tundra diversity decline differed across
experimental treatments at both the plot
and site scale (table S1). Tundra diversity
declined nearly twice as fast under herbi-
vore exclusion as under grazing, most slowly

y
under grazing with experimental warming
and, at the site scale, most rapidly under her-
bivore exclusion with warming (Fig. 2 and
table S1).
The decline in tundra diversity across all
experimental treatments suggests a common
abiotic driver whereas different rates of de-
cline under grazing versus herbivore exclusion
suggest mediation of that abiotic driver by
herbivory. We examined relationships among

y g
declining tundra diversity and local precipi-
tation and temperature during the growing
season (May through July; hereafter May-
July), as well as both Arctic-wide sea ice extent
(ASIE) and regional Baffin Bay sea ice extent

,
(BBSIE) (24). Growing season mean ASIE
was the best overall predictor of interan-
nual variation in tundra diversity at the site
whereas July (peak growing season) temper-
ature was the best local predictor (table S2).
Moreover, ASIE outperformed BBSIE (tables
S2 and S3). Associations between May-July
ASIE and tundra diversity were positive
across all experimental treatments, both at Fig. 2. Arctic tundra diversity trends and associations with sea ice decline. Decline in arctic tundra
the plot scale and across the study site (Fig. community diversity (True Simpson’s diversity, or effective species number) across treatments over the
2 and tables S3 and S4), indicating that course of a 15-year field experiment near Kangerlussuaq, Greenland, 2003 to 2017 (upper row) and
tundra diversity declined with diminishing association with declining Arctic-wide sea ice extent during the annual growing season (May through July)
ASIE during the growing season. Moreover, (bottom row). Panels in the upper row show plot-scale (small, pale symbols; n = 174 for exclosed plots and
associations between May-July ASIE and tun- n = 189 for grazed plots) and site-scale (large, dark symbols; n = 15 in all cases) community diversity during each
dra diversity were stronger under herbivore year of the experiment. Panels in the bottom row show plot- and site-scale community diversity in relation to
exclusion than under grazing and weakest mean Arctic-wide sea ice extent during the annual growing season (May-July). Trend lines are fit to site-scale mean
under grazing with experimental warming values. Plot- and site-scale trends and sea ice coefficients (b ± 1SE) were significant for all treatments (24).

Post et al., Science 380, 1282–1287 (2023) 23 June 2023 2 of 6

RES EARCH | R E S E A R C H A R T I C L E
indicate that this relationship operates through
effects of synoptic-scale air mass configurations
that increase the frequency of easterly winds
over the study area in July during low sea ice
years (figs. S4 and S5).
Mediation of tundra diversity responses
to the experiment by large herbivore diversity
and abundance
Greater rates of decline in tundra diversity
under herbivore exclusion than under grazing,
and stronger associations between abiotic
conditions and declining tundra diversity on
exclosed than on grazed plots, suggest a role
of vertebrate herbivores in mediating sea ice
effects at this site. Hence, we examined the
response of tundra diversity to our experiment
in relation to background variation in large
herbivore diversity (or abundance) and May-
July ASIE over the course of the experiment
(24). At the inception of this experiment, lo-
cal caribou abundance was 20 times that of
muskoxen but declined throughout the exper-
iment whereas muskox abundance more than
tripled (30) (Fig. 3, upper panels). Opposing
abundance trends of these two species resulted
in an increase in effective species number (True
Simpson’s diversity) of large herbivores from a
low of approximately 1 to a high of 1.7 between
2003 and 2017 (Fig. 3). Across the arctic tundra
biome, local richness of large herbivores typically
ranges from 0 to 2 species (19), so the magnitude
of the increase in herbivore diversity over the
course of this experiment is comparable to
extant biome-wide variation in large herbivore
Fig. 3. Dynamics of
large herbivore abun-
dance and diversity at
the study site near
Kangerlussuaq, Green-
land. Increase in large
herbivore diversity (True
Simpson’s diversity, or
effective species number)
(24) at the study site
near Kangerlussuaq,
Greenland (2003 to
2017; linear regression
b = 0.06 ± 0.01, t = 5.02,
P < 0.001). Upper panels
show annual maximum
abundances of caribou
and muskoxen observed
at the site used to
calculate True Simpson’s
diversity of the large
herbivore assemblage
(24, 30). Caribou
and muskox silhouettes
in the lower panel are
intended to illustrate a shift from dominance of effective species number by caribou at the beginning of the
15-year tundra warming and herbivore exclusion experiment to contributions of both species to effective
species number by the end of the experiment.
Post et al., Science 380, 1282–1287 (2023) 23 June 2023
diversity (but not density, which is compara-
tively low at this site) (24, 31).
Tundra diversity responses to the experi-
mental treatments were unrelated to May-July
ASIE but were strongly related to variation in
large herbivore diversity over the course of the
experiment (table S7) (Fig. 4). The exclosure
effect on tundra diversity on warmed plots was
negatively associated with large herbivore di-
versity, whereas the warming effect on tundra
diversity on grazed plots was positively asso-
ciated with large herbivore diversity (Fig. 4).
Both relationships indicate that a large herbi-
vore diversity index above an effective species
number of ~1.5 promoted greater tundra di-
versity on grazed versus exclosed plots, and on
warmed versus ambient plots under grazing.
Hence, although tundra diversity declined across
the study site in association with trends in
background abiotic conditions, the concurrent
increase in large herbivore diversity likely fa-
cilitated a slower decline and greater overall
tundra diversity on grazed and experimentally
warmed plots than on plots under any other
experimental treatment combination. This was
especially evident in the latter years of the ex-
periment when herbivore diversity was greatest
(Fig. 2, top row, small symbols).
Tundra diversity responses to the experi-
ment were also more strongly related to var-
iation in large herbivore diversity than to
variation in caribou or muskox abundance
individually [table S8; see table S9 for analy-
ses that also considered arctic hares (Lepus
arcticus; fig. S6), which were not significant].
Moreover, tundra diversity responses to the
experiment were differentially associated with
variation in caribou versus muskox abun-
dance (table S8), likely reflecting distinct in-
fluences of the two herbivores on tundra taxa
(fig. S7). For instance, the exclosure effect
on warmed plots was positively associated
with caribou abundance but negatively as-
sociated with muskox abundance (Fig. 4).
Consequently, the slower sea ice–associated
decline in tundra diversity under grazing with
warming eventually resulted in greater tundra
diversity under this treatment than under her-
bivore exclusion with warming (Fig. 2). This
difference developed even as caribou abun-
dance declined and muskoxen increased at the
study site. Moreover, the strength of the effect
of the warming treatment on tundra diversity on
grazed plots increased with increasing muskox
abundance at the site but not with caribou
abundance (Fig. 4). This suggests distinct inter-
p
actions of the two herbivores with warming,
further emphasizing the potential for varia-
tion in diversity of the herbivore assemblage
to influence tundra community responses to
background abiotic change (Fig. 1).
Contributions of individual tundra taxa
g
to tundra diversity
Finally, we sought to determine which taxa
contributed most to variation in tundra di-
versity under each experimental treatment
y
combination (24) (fig. S8). Tundra diversity
declined with increasing abundances of de-
ciduous shrubs (Fig. 5A) and, consistent with
other findings (32), shrub abundance at the
Kangerlussuaq site increased with both her-
bivore exclusion and warming (33). The de-
cline in tundra diversity with increasing shrub
abundance occurred under all treatments but
was strongest under exclosed ambient con-
ditions and weakest under grazed warmed
y g
conditions (Fig. 5A and table S10). Contribu-
tions of shrub increases to declining diver-
sity were, moreover, generally weaker under
herbivory than under herbivore exclusion
(Fig. 5A and table S10). Deciduous shrubs were
,
also the only functional group contributing
to tundra diversity decline as they became more
common (33) during the experiment (Fig. 5B).
Tundra diversity decline appears, therefore,
attributable mainly to increases in the two
most common species at the site: dwarf birch
and gray willow, the abundances of which are
constrained by herbivory and promoted by
warming, respectively (18). The abundances
of both species also co-vary negatively at the
site with sea ice extent (27). The role of large
herbivore diversity in this suite of interactions
likely derives from the distinct yet comple-
mentary seasonal presence and behavioral-
foraging impacts of caribou and muskoxen.
The impact of migratory caribou on vegetation
at the site is likely limited mainly to growing
3 of 6
RES EARCH | R E S E A R C H A R T I C L E

Fig. 4. Arctic tundra

diversity responses to
experimental manipula-
tions in relation to
background variation in
herbivore diversity,
caribou abundance, and
muskox abundance.
Variation in the magnitude
of annual experimental
treatment effects on tun-
dra community diversity,
quantified as delta-
corrected log response
ratios (24), in relation to
interannual variation in
large herbivore diversity
(upper row), abundance of
caribou (middle row), and
abundance of muskoxen
(bottom row) at the study

p
site near Kangerlussuaq,
Greenland, 2003 to 2017.
Shown are the exclosure
effect on ambient plots (E on
A), for which positive (neg-
ative) values indicate that

g
tundra diversity was greater
(lower) on exclosed plots
than on grazed plots under
ambient conditions; the
exclosure effect on warmed

y
plots (E on W), for which
positive (negative) values
indicate that tundra diver-
sity was greater (lower) on
exclosed plots under
warming than on grazed
plots under warming; the
warming effect on grazed
plots (W on G), for which
positive (negative) values

y g
indicate that tundra diversity
was greater (lower) on
warmed plots under grazing
than on ambient plots under
grazing; and the double

,
treatment double control
effect (DTDC), for which
positive (negative) values indicate that tundra diversity was greater (lower) on exclosed warmed plots than on grazed ambient plots. Solid and dashed lines indicate significant
and nonsignificant relations, respectively; multivariate generalized linear model Wald Chi-Square values and significance tests for each relationship are reported in table S8.

season herbivory whereas that of resident
muskoxen is annual (fig. S8). Moreover, be-
havioral and rumen content analyses of the
two species in the area indicate that they ex-
hibit distinct intra- and interseasonal prefer-
ences for dwarf birch and gray willow (34–36)
(fig. S8). Such differences, although in some
ways subtle, may foster a “portfolio effect”
(37) of heterogeneous herbivore impacts on
tundra community composition, buffering re-
sponses to climate change that might other-
wise develop more quickly in concert with
changes in herbivore abundance in a single
large herbivore system.
Implications for ongoing climate change
impacts on tundra diversity and rewilding
Biodiversity loss is projected to be one of the
most likely and ecologically consequential out-
comes of climate change (1, 2, 38). In addition
to cultural and economic impacts to humans
(39, 40), biodiversity loss will affect key eco-
logical properties, including community stability
and ecosystem function (41). Recent assessment
reports have indicated that polar ecosystems
have experienced the greatest impacts of climate
change on biodiversity over the past century
(39), that climate change impacts experienced
by Indigenous Peoples and local communi-
ties are greater in tundra habitats than in any
other biome (40), and that there is a “high to
very high” risk of imminent biodiversity loss
in the Arctic with ongoing climate change (42).
Our results likewise signal potential adverse
consequences for tundra biodiversity of sea ice
loss and warming.

Post et al., Science 380, 1282–1287 (2023) 23 June 2023 4 of 6

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

A or reverse ecosystem state shifts resulting from

the loss of herbivore diversity (43). Hence, our
results may have relevance to discussions of
nature-based solutions (44) to climate change
impacts such as rewilding of the Arctic to mit-
igate tundra diversity losses associated with
woody plant encroachment (22, 45) (Fig. 1).
Our study demonstrates that climate-driven
vegetation diversity decline may be medi-
ated by an increase in large herbivore diver-
sity, even at the modest level of increase seen
here. Efforts focused on maintenance or en-
hancement of large herbivore diversity may
therefore under certain conditions help miti-
gate climate change impacts on at least one
important element of ecosystem health and
function: tundra diversity.

REFERENCES AND NOTES

1. C. H. Trisos, C. Merow, A. L. Pigot, Nature 580, 496–501
(2020).
2. M. C. Urban, Science 348, 571–573 (2015).

p
3. E. Post et al., Sci. Adv. 5, eaaw9883 (2019).
4. F. S. Chapin 3rd, S. Díaz, Proc. Natl. Acad. Sci. U.S.A. 117,
B 6295–6296 (2020).
5. M. Vellend et al., Proc. Natl. Acad. Sci. U.S.A. 110, 19456–19459
(2013).
6. S. C. Elmendorf et al., Ecol. Lett. 15, 164–175 (2012).
7. AMAP, Arctic Climate Change Update 2021: Key Trends and
Impacts (Arctic Monitoring and Assessment Programme, 2021).

g
8. D. Notz, J. Stroeve, Science 354, 747–750 (2016).
9. I. H. Onarheim, T. Eldevik, L. H. Smedsrud, J. C. Stroeve,
J. Clim. 31, 4917–4932 (2018).
10. J. C. Stroeve et al., Geophys. Res. Lett. 39, 2012GL052676
(2012).
11. D. D. W. Hauser et al., Environ. Res. Lett. 16, 095003 (2021).

y
12. B. C. Forbes et al., Biol. Lett. 12, 20160466 (2016).
13. U. S. Bhatt et al., Annu. Rev. Environ. Resour. 39, 12.11–12.33
(2014).
14. M. Macias-Fauria, E. Post, Biol. Lett. 14, 20170702 (2018).
15. A. Buchwal et al., Proc. Natl. Acad. Sci. U.S.A. 117, 33334–33344
(2020).
16. P. Fauchald, T. Park, H. Tømmervik, R. Myneni, V. H. Hausner,
Sci. Adv. 3, e1601365 (2017).
17. K. A. Bråthen, V. T. Ravolainen, A. Stien, T. Tveraa, R. A. Ims,
Ecol. Appl. 27, 2416–2427 (2017).
18. E. Post, S. M. P. Cahoon, J. T. Kerby, C. Pedersen, P. F. Sullivan,
Proc. Natl. Acad. Sci. U.S.A. 118, e2015158118 (2021).
19. J. Olofsson, E. Post, Philos. Trans. R. Soc. 373, 20170437

y g
(2018).
Fig. 5. Taxa contributing most to variation in arctic tundra diversity under long-term experimental 20. E. Kaarlejärvi, A. Eskelinen, J. Olofsson, Nat. Commun. 8, 419
warming and large herbivore exclusion. (A) Species or taxa contributing most to variation in tundra community (2017).
diversity (24) at the plot scale under each experimental treatment combination at the study site near Kangerlussuaq, 21. O. J. Schmitz et al., Science 362, eaar3213 (2018).
22. J. C. Svenning et al., Proc. Natl. Acad. Sci. U.S.A. 113, 898–906
Greenland, 2003 to 2017. Black arrows indicate the direction and magnitude of contributions of each species or (2016).
taxon to tundra diversity and are scaled according to the values of the standardized linear coefficients quantifying 23. Y. Malhi et al., Curr. Biol. 32, R181–R196 (2022).
their individual contributions to variation in tundra diversity (24). Taxa are arranged within each treatment from 24. Methods are available as supplementary materials.

,
25. E. Post, C. Pedersen, Proc. Natl. Acad. Sci. U.S.A. 105,
top to bottom in order of decreasing contributions to tundra diversity. (B) Magnitude of the contribution of individual 12353–12358 (2008).
taxa to tundra diversity [standardized linear model coefficient; (24)] in relation to changes in their commonness 26. J. T. Kerby, E. Post, Nat. Commun. 4, 2514 (2013).
[abundance-weighted occurrence; (33)] over the course of the experiment. Positive y-axis values indicate that 27. E. Post et al., Science 341, 519–524 (2013).
28. NASA, Arctic sea ice minimum extent - annual September
increases in the abundance of a taxon enrich tundra diversity, and negative values indicate that increases in
Minimum Extent (National Aeronautics and Space
abundance of a taxon reduce tundra diversity. Positive x-axis values indicate that a taxon has become more common Administration, 2020); https://climate.nasa.gov/vital-signs/
under that experimental treatment, whereas negative values indicate that a taxon has become rarer under that arctic-sea-ice/.
29. J. E. Saros et al., Environ. Res. Lett. 14, 074027 (2019).
treatment, over the duration of the experiment (33). Experimental treatment abbreviations in the lower panel are as
30. M. Eikelenboom, R. C. Higgins, C. John, J. T. Kerby, E. Post,
follows: EA, exclosed ambient; EW, exclosed warmed; GA, grazed ambient; and GW, grazed warmed. Common names Food Webs 27, e00196 (2021).
refer to the following species, genera, or families: dwarf birch (Betula nana), moss (Aulacomnium sp.), graminoids 31. K. E. M. Vuorinen et al., Environ. Res. Lett. 17, 034013
(Gramineae), horsetail (Equisetum sp.), dog lichen (Peltigera sp.), gray willow (Salix glauca), and draba (Draba sp.). (2022).
32. J. Olofsson et al., Glob. Change Biol. 15, 2681–2693 (2009).
33. E. Post, C. Pedersen, D. A. Watts, Sci. Rep. 12, 1292
The extent to which management or con- indicated both warming-driven losses of arctic (2022).
servation of large herbivores might slow or plant diversity and mediation or reversal of 34. H. Thing, Dan. Rev. Game Biol. 12, 1–53 (1984).
35. M. C. Forchhammer, Can. J. Zool. 73, 1344–1361 (1995).
possibly reverse tundra diversity losses in the such losses by large herbivores (20), and ad-
36. M. C. Forchhammer, J. J. Boomsma, Oecologia 104, 169–180
Arctic is difficult to generalize from our re- ditional evidence suggests that herbivore re- (1995).
sults. However, other recent work has also introductions or replacements may prevent 37. D. E. Schindler et al., Nature 465, 609–612 (2010).

Post et al., Science 380, 1282–1287 (2023) 23 June 2023 5 of 6

RES EARCH | R E S E A R C H A R T I C L E
38. S. Diaz et al., . Science 366, eaax3100 (2019).
39. R. Hassan, R. Scholes, N. Ash, Millennium Ecosystem
Assessment: Ecosystems and Human Well-being: Current state
and trends (World Resources Institute, Island Press, 2005).
40. IPBES, Global assessment report on biodiversity and
ecosystem services of the Intergovernmental Science-Policy
Platform on Biodiversity and Ecosystem Services (IPBES
Secretariat, 2019).
41. F. S. Chapin III et al., Nature 405, 234–242 (2000).
42. Working Group II, Climate Change 2022: Impacts, Adaptation,
and Vulnerability (IPCC Sixth Asssment Report, Cambridge
Univ. Press, 2002).
43. J. C. Svenning, One Earth 3, 657–660 (2020).
44. E. Cohen-Shacham, G. Walters, C. Janzen, S. Maginnis, Nature-
based solutions to address societal challenges (International
Union for Conservation of Nature, 2016).
45. M. Macias-Fauria, P. Jepson, N. Zimov, Y. Malhi, Philos. Trans.
R. Soc. 375, 20190122 (2020).
46. E. Post, Tundra taxa annual abundances under experimental
herbivore exclosure and warming near Kangerlussuaq, Greenland,
2003-2017, NSF Arctic Data Center (2022); https://arcticdata.
io/catalog/view/doi%3A10.18739%2FA2CR5ND85.
47. E. Post, Annual summer maximum daily counts of adult caribou
and muskoxen at the Kangerlussuaq, Greenland, study site
2003-2017, NSF Arctic Data Center (2022); https://arcticdata.io/
catalog/view/doi%3A10.18739%2FA2HH6C72X.
48. NSIDC, Sea Ice Index, Version 3 (G02135) (National Snow and Ice
Data Center, 2018); https://nsidc.org/data/g02135/versions/3.
Post et al., Science 380, 1282–1287 (2023) 23 June 2023
49. NOAA, Baffin Bay Sea Ice Extent (National Centers for
Environmental Information, 2023); https://www.ncei.noaa.gov/
access/monitoring/regional-sea-ice/extent/Baffin/0.
50. J. Cappelen, Weather observations from Greenland 1958-2020
(DMI Report 21-08, 2021).
ACKN OWLED GMEN TS
We thank M. Forchhammer and H. Thing for guidance on
experimental design and assistance with the permitting process;
local residents of Kangerlussuaq, including B. Brodersen, R. Møller,
C. Sørensen, and B. Vængtoft at Kangerlussuaq International
Science Support, for long-term assistance with fieldwork planning
and operations; the 109th Airlift Wing of the New York Air
National Guard for assistance with logistics; numerous volunteers
for assistance with fieldwork; J. C. Yde for assistance with the
Kangerlussuaq weather data; and J. C. Svenning and four
anonymous reviewers for constructive comments on the
manuscript. Funding: This research was supported by grants to
E.P. from the US National Science Foundation (NSF) (0124031,
0217259, 0732168, 0713994, 1107381, 1525636) and the National
Geographic Society; to E.K. from the Finnish Cultural Foundation
and Academy of Finland (347188); to J.T.K. from the European
Union’s Horizon 2020 research and innovation program under the
Marie Skłodowska-Curie grant (agreement 754513), and the
Aarhus University Research Foundation; and to M.M-F. from the
UK’s Natural Environment Research Council (NERC; grant NE/
L011859/1). Author contributions: Experimental design and
implementation: P.S.B., C.P., and E.P. Data collection: P.S.B.,
S.M.P.C., R.C.H., C.J., J.T.K., C.P., E.P., M.P., and D.A.W. Conceptual
development of the study: All authors. Analytical design: R.C.H., C.J.,
E.K., J.T.K., M.M.-F., E.P., P.F.S., and D.A.W. Data analysis: M.M.-F.,
E.P., and P.F.S. Writing and editing: E.P., with contributions from all
co-authors. Competing interests: The authors declare no competing
interests. Data and materials availability: Data on tundra taxa
abundances and large herbivore abundances are available at the
Arctic Data Center (46, 47). Data on sea ice extent are available at
the US National Snow and Ice Data Center (48) and the US National
Oceanic and Atmospheric Administration (49). Kangerlussuaq
weather data are available from the Danish Meteorological
Institute (50). License information: Copyright © 2023 the authors,
some rights reserved; exclusive licensee American Association
for the Advancement of Science. No claim to original US
government works. https://www.sciencemag.org/about/science-
licenses-journal-article-reuse
SUPPLEMENTARY MATERIALS
science.org/doi/10.1126/science.add2679
Materials and Methods
Supplementary Text
Figs. S1 to S8
Tables S1 to S10
References (51–64)
MDAR Reproducibility Checklist
Submitted 1 June 2022; accepted 16 May 2023
p
10.1126/science.add2679
g
y
y g
,
6 of 6
RES EARCH

STRUCTURAL BIOLOGY processing, this dataset diverged into three

subclasses that produced three-dimensional
Structures of the free and capped ends of the reconstructions of the free pointed end, Tmod-
capped pointed end, and CapZ-capped barbed
actin filament end at resolutions of 2.84, 3.26, and 2.79 Å,
respectively (fig. S1D). Free barbed ends were
Peter J. Carman1,2†, Kyle R. Barrie1,2†, Grzegorz Rebowski1, Roberto Dominguez1,2* not observed, consistent with CapZ’s subna-
nomolar affinity for the barbed end and long
The barbed and pointed ends of the actin filament (F-actin) are the sites of growth and shrinkage half-time for dissociation (12). Datasets were
and the targets of capping proteins that block subunit exchange, including CapZ at the barbed end collected without CapZ to obtain the structure
and tropomodulin at the pointed end. We describe cryo-electron microscopy structures of the free and of the free barbed end. Attempts to shorten
capped ends of F-actin. Terminal subunits at the free barbed end adopt a “flat” F-actin conformation. the filaments by shearing mostly failed because
CapZ binds with minor changes to the barbed end but with major changes to itself. By contrast, subunits filaments reannealed fast before vitrification
at the free pointed end adopt a “twisted” monomeric actin (G-actin) conformation. Tropomodulin binding with or without Tmod. A total of 413,076 par-
forces the second subunit into an F-actin conformation. The structures reveal how the ends differ ticles were collected from two datasets, which,
from the middle in F-actin and how these differences control subunit addition, dissociation, capping, combined with a different data processing
and interactions with end-binding proteins. strategy, yielded a 3.30-Å resolution structure
of the free barbed end (fig. S2). A 2.26-Å

A
resolution structure of the middle of F-actin
ctin is the most abundant cytosolic pro- cells, the main capping proteins are CapZ (also was obtained by helical reconstruction and
tein in eukaryotes. The cellular pool of known as capping protein or CP) (12) at the used as reference for comparisons. This struc-

actin is approximately evenly divided
between monomeric (G-actin) and fil-
amentous (F-actin) forms (1). G-actin
consists of four subdomains which, based on
their disposition in F-actin, form the outer
(subdomains 1 and 2) and inner (subdomains
barbed end and tropomodulin (Tmod) at the
pointed end (13).
Recent cryo-electron microscopy (cryo-EM)
helical reconstructions of F-actin in different
nucleotide states [ADP, ADP-Pi (inorganic
phosphate), and ATP analogs] provide high-
p
ture is similar to other structures of the middle
of F-actin, with a Ca root mean square devia-
tion of 0.35 Å with the highest-resolution
structure (2.24 Å) reported thus far (5). Poly-
merization started from Mg2+-ATP-actin, but
subunits in all the structures described here

3 and 4) domains (2) (Fig. 1A). Two clefts
separate the outer and inner domains: the
nucleotide-binding cleft and, opposite of it, a
hydrophobic cleft that mediates interactions
with numerous actin-binding proteins and
resolution insights into the structure of subunits
in the middle of F-actin and the conformational
changes that lead to ATP hydrolysis upon polym-
erization (5, 14, 15). However, the structure of
F-actin changes little as a function of the bound
g
contain bound ADP, including subunits at
the barbed and pointed ends, indicating that
hydrolysis and g-phosphate release occurred
in the ~3 hours that elapsed before vitrifica-
tion. The helical rise and twist of subunits is

with actin itself in the filament (3). Adenosine
triphosphate (ATP) hydrolysis by actin is one
of several factors that control the G- to F-actin
transition. Although G-actin is a slow adeno-
sine triphosphatase (ATPase), ATP hydrolysis
is accelerated manyfold in F-actin as a result
of a rotation of the outer domain relative to the
inner domain, which produces a flatter subunit
conformation and reorients side chains and
water molecules in the catalytic site for hydro-
nucleotide or divalent cation (Ca2+ or Mg2+).
Thus, although the rate constants for the re-
actions of ATP- and ADP-actin addition and
dissociation at the ends of F-actin were deter-
mined ~40 years ago (1, 6), the structural bases
of kinetic asymmetry at the ends of F-actin re-
main poorly understood. On the basis of single-
particle cryo-EM structures, we reveal how the
barbed and pointed ends are conformationally
different and primed for subunit addition and
y
also similar in all the structures. Data collec-
tion, refinement, and structure validations are
shown in figs. S3 to S7 and table S1.
Tropomyosin density was only observed in
a subset (~13%) of the Tmod-capped pointed
end subclass, resulting in a 4.8-Å resolution
reconstruction that shows tropomyosin inter-
acting with Tmod on both sides of F-actin (fig.
S8A). The median length of the filaments in
micrographs was 48 nm, which corresponds to

lysis (4, 5). In this way, F-actin acquires
polarity, i.e., structural and kinetic asymmetry:
ATP-actin binds to the barbed (or +) end of
F-actin, followed by fast ATP hydrolysis, slow
g-phosphate release, and dissociation of adeno-
dissociation, respectively, and how CapZ and
Tmod block subunit exchange.
Results
Cryo-EM structures of the filament ends
y g
~8.7 actin protomers and ~1.25 tropomyosin
molecules along the long-pitch helix (fig. S1B).
The low occupancy of tropomyosin is therefore
consistent with the extremely low (millimolar)
affinity of single tropomyosin molecules for

sine diphosphate (ADP)–actin from the pointed
(or −) end (1, 6). Unbound monomers undergo
rapid ADP-to-ATP exchange and become ready
to rejoin the barbed end. This process, known
as filament treadmilling, is regulated by pro-
teins that either accelerate subunit addition at
the barbed end (formins, Ena/VASP) (7–9),
accelerate subunit dissociation from the
pointed end (cyclase-associated protein and
cofilin) (10, 11), or cap the ends to stop subunit
exchange. In muscle sarcomeres and nonmuscle
1
Department of Physiology, Perelman School of Medicine,
University of Pennsylvania, Philadelphia, PA, USA.
2
Biochemistry and Molecular Biophysics Graduate Group,
Perelman School of Medicine, University of Pennsylvania,
Philadelphia, PA, USA.
*Corresponding author. Email: droberto@pennmedicine.upenn.edu
†These authors contributed equally to this work.
We used single-particle cryo-EM to determine
structures of the ends of F-actin. Actin alone
tends to form long filaments in cryo-EM mi-
crographs, which is ideal for helical recon-
struction but produces few filament ends
per micrograph. To increase the number of
ends, we used an optimized mixture of CapZ,
Tmod, tropomyosin, and actin that produced
shorter filaments and thus, more ends per
micrograph (materials and methods; fig. S1,
A to C). Filament end particles were harvested
with the program Topaz, using a particle-
picking pipeline based on custom-trained
neural networks (16). Starting from a subset
of manually picked ends, iterations of Topaz
training and particle picking resulted in a
dataset of 1,102,793 particles. During data
,
F-actin (17). That tropomyosin was only ob-
served in conjunction with Tmod probably re-
sults from the interaction of these two proteins
at the pointed end, which mutually enhances
their affinities for F-actin (13, 18). The low res-
olution of the Tmod-tropomyosin map pre-
cludes a detailed analysis of this interaction.
The barbed end
With one exception, the conformation of the
two terminal subunits at the barbed end cor-
responds to the classical flat conformation
of subunits in the middle of the filament
(4, 5, 14, 15) (Fig. 1, A to C, movie S1, and
table S2). Three main features characterize
this conformation: (i) a ~20° scissorlike rota-
tion of the outer domain relative to the inner

Carman et al., Science 380, 1287–1292 (2023) 23 June 2023 1 of 6

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

A B

C D E

Fig. 1. Free barbed end. (A) Schematic representation of the G- to F-actin
transition. A scissorlike ~20° rotation of the outer domain (subdomains 1 and 2)
relative to the inner domain (subdomains 3 and 4) produces a flatter
conformation of subunits in F-actin. (B) Cryo-EM map of the free barbed end at
3.30-Å resolution. The terminal and penultimate subunits are shown in two
different shades of blue. (C) The terminal and penultimate subunits (blue) adopt
the classical F-actin conformation of subunits in the middle of F-actin (gray).
Subunits were superimposed based on the inner domain (surface representation)
to highlight differences in orientation of the outer domain (ribbon representation).
p
g
y
(D) Fit to the cryo-EM map of the W loop of the terminal subunit (blue) versus a
subunit from the middle of the filament (gray). (E) In the middle of F-actin,
the W loop and the C terminus of actin form a pincer-like structure that grasps
the D loop of the subunit below (gray and cyan, respectively). This interaction
is absent for subunits at the barbed end (blue), resulting in the W loop adopting
a G-actin conformation and the C terminus moving by 2.0 Å (red arrows).
Single-letter abbreviations for amino acids referenced throughout are as
follows: R, Arg; K, Lys; I, Ile; P, Pro; F, Phe; Q, Gln; G, Gly; E, Glu; H, His;
L, Leu; N, Asn; A, Ala; T, Thr

y g
domain that flattens the actin protomer and barbed end of the actin-related protein 1 (Arp1) CapZ binding but is likely favored by weakened
changes the overall conformation of both the minifilament of dynactin (20). Yet, Arp1 shares contacts between the two long-pitch helices at
nucleotide and hydrophobic clefts (Fig. 1A); only 51% sequence identity with actin and its the barbed end. Indeed, the interstrand plug

(ii) the D loop (residues R39 to K50) adopts
an extended conformation to insert into the
hydrophobic cleft of the subunit immediately
above; and (iii) residues in the hydrophobic
cleft change conformation to host the D loop
of the subunit immediately below. Specifi-
cally, the W loop (I165 to P172) and the C
terminus of actin (P367 to F375) form a pincer-
like structure that grasps the D loop of the
subunit below. The latter change does not
occur in barbed-end subunits because they do
not interact with the D loop (Fig. 1, D and E).
A cryo-EM structure of the CapZ hetero-
dimer (Fig. 2, A and B) at the barbed end was
recently reported (19), but CapZ was only
partially visualized and at low resolution
(6 to 7 Å). CapZ was also visualized at the
interaction with CapZ differs somewhat. Com-
pared with these structures, the 2.79-Å resolu-
tion structure described here (Fig. 2C) displays
large differences in the overall disposition and
specific interactions of CapZ at the barbed end,
and the relative orientation of actin (or Arp1)
subunits in the filament (fig. S9A). As in the
structure of the free barbed end, the terminal
and penultimate subunits in the CapZ-capped
structure have an F-actin conformation with a
G-actin–like W loop (table S2). However, the
short-pitch pair formed by these two subunits
is splayed apart by ~1.5 Å compared with a short-
pitch pair in the middle of F-actin (Fig. 2D).
Subunit splaying results from a ~2° rotation of
the terminal subunit away from the longitudi-
nal axis of F-actin, which may be induced by
,
(Q263 to G273) provides the main contact be-
tween the long-pitch helices (or strands) of
F-actin (2). In the middle of the filament, the
plug from one subunit binds at the interface
between two subunits of the opposite strand, a
pattern that is broken at the barbed end where
the plug of the terminal subunit contacts only
one subunit of the opposite strand (Fig. 2D).
CapZ is a heterodimer of related a and b
subunits (Fig. 2A) (12). The heterodimer is
described as having a mushroom-like shape,
where the mushroom head binds the barbed
end (Fig. 2B). CapZ can adopt distinct filament-
bound and unbound conformations. Alloste-
ric regulators of CapZ, including CARMIL,
CD2AP, and CKIP, have a capping-protein
interaction (CPI) motif that binds around the

Carman et al., Science 380, 1287–1292 (2023) 23 June 2023 2 of 6

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

A E

C D F

p
G

Fig. 2. CapZ-capped barbed end. (A) Domain diagram of CapZ subunits a and
b. (B) Mushroom-like structure of CapZ and transition between filament-bound
and unbound states. In filament-bound CapZ, the mushroom head flattens and
the tentacles engage the hydrophobic clefts of barbed-end subunits. (C) Cryo-EM
map of the CapZ-capped barbed end at 2.79-Å resolution, showing the terminal
and penultimate actin subunits in two shades of blue and CapZa and CapZb in pink
and magenta, respectively. (D) Splaying of the short-pitch pair formed by the
terminal and penultimate subunits (blue) as compared to a short-pitch pair from
g
y
(E) Comparison of the filament-bound (pink, magenta) and unbound (22)
(gray, PDB code 3AA7) structures of CapZ. The superimposition, based on
CapZa, highlights a ~15° rotation of CapZb (red arrow) that flattens the
mushroom head. (F) The actin-binding surface of CapZ consists mostly
of two antiparallel helices and the a and b tentacles. Comparison of the
filament-bound (pink, magenta) and unbound (gray) structures shows
that the helices contain p bulges that change conformation between these two
states (red arrows). The tentacles, which are disordered in the unbound

the middle of F-actin (gray). The short-pitch pairs were superimposed based
on the penultimate subunit to highlight the splaying of the terminal subunit,
showing a maximum displacement of 1.5 Å resulting from a rotation of ~2° (red
arrow). Splaying is likely favored by missing contacts of the interstrand plug
at the barbed end compared to the middle of F-actin (dashed red curves).
y g
structure, project out in the filament-bound structure to engage the two barbed-
end subunits. (G) Close-up view of CapZ’s interaction with the barbed end,
with the binding interface colored pink (CapZa) and magenta
(CapZb). CapZ residues participating in the interaction are shown and
labeled in fig. S9C.

mushroom stalk and stabilizes the unbound
conformation to promote uncapping (21, 22).
The filament-bound state is characterized by
a flatter conformation of the mushroom head,
resulting from the distal ends of the head moving
upwards toward the barbed end (19, 20). This
conformation has also been observed in CapZ
complexes with twinfilin-actin (23) and V-1 (22),
a protein that sterically competes with CapZ
binding to the barbed end (fig. S9B). In the
current structure, CapZ also adopts a flat con-
formation, bending by ~15° from its filament-
unbound conformation (Fig. 2E and movie S1).
Flattening of the mushroom head leads to
changes in two long, pseudo-symmetric helices
of CapZa (E221 to R259) and CapZb (H209 to
L243) that run along the barbed-end interface
(Fig. 2F). Both helices contain a p bulge, a
deformation of the a-helical fold that often
plays a functional role (24). In the barbed end–
bound structure, the bulge in the CapZb helix
travels along the helical axis toward the N
terminus by about half a helical turn, which
reorients the side chains of N229, E230, and
I231 at the barbed-end interface. In CapZa,
the p bulge coincides with a kink in the helix,
which becomes more pronounced in the
filament-bound structure to reorient the side
chains of D252 and K256 at the barbed-end
interface. Other important elements of the
interaction are the a tentacle (CapZa residues
R260 to A286) and the b tentacle (CapZb
,
residues R244 to N277) (25), immediately
C-terminal to the p-bulge helices (Fig. 2G).
The tentacles are well resolved in our struc-
ture and their conformation differs from other
filament-bound structures (19, 20), as well as
unbound structures in which the tentacles are
disordered (21–23). Both tentacles comprise
amphipathic helices that project upwards and
present their hydrophobic sides to the hydro-
phobic clefts of the terminal (b tentacle) and
penultimate (a tentacle) subunits of the filament.
The pointed end
Unexpectedly, the first two subunits at the free
pointed end adopt a G-actin conformation and
their D loops are disordered, as observed in

Carman et al., Science 380, 1287–1292 (2023) 23 June 2023 3 of 6

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

most G-actin structures (Fig. 3, A and B, movie
S1, and table S2). However, their hydrophobic
clefts host the D loop of subunits immediately
below, which imposes changes in the W loop
and C terminus characteristic of the F-actin
conformation. The opposite was observed at
the barbed end, where the last two subunits
A B
have an F-actin conformation with G-actin-like
features in their hydrophobic cleft (Fig. 1). We
draw three conclusions from these observa-
tions. First, the G- to F-actin transition is caused
by interactions of subdomains 2 and 4 with
the subunit immediately above along the long-
pitch helix. These interactions impose an F-
actin conformation to subunits at the barbed
end, whereas their absence at the pointed end
allows subunits to revert to the G-actin confor-
mation. Second, subunit addition at the pointed
end is conformationally unfavorable, analo-
gous to the way G-actin to G-actin interactions
are rate limiting during nucleation (1, 26).
Third, subunits at the pointed end appear con-
formationally primed for dissociation.
Contrary to CapZ, which capped all barbed
ends, Tmod capped only ~20% of pointed
ends, resulting in a 3.26-Å resolution structure
(Fig. 4, A and B, and movie S1). Tmod’s low
pointed-end occupancy may be in part attri-
buted to inefficient pointed-end capping in
the absence of tropomyosin (13, 18), which
binds weakly to short filaments (17). Tmod
comprises alternating tropomyosin and actin-
binding sites (Fig. 4A). Likely owing to the lack
of interactions with tropomyosin, the two tropo-
myosin binding sites are disordered in the

p
Tmod-capped structure (Fig. 4B), but appear
ordered in the low-resolution Tmod-tropomyosin
map (fig. S8A). In the Tmod-capped structure,
actin binding site 1 (ABS1, P58 to K99) has an
extended conformation, interrupted by a single
a helix, and interacts from N to C terminus

g
Fig. 3. Free pointed end. (A) Cryo-EM map of the free pointed end at 2.84-Å resolution, showing the with subdomains 4, 2, and 1 of the first actin
first and second subunits in two different shades of green. The D loop of subunits at the pointed end is subunit (Fig. 4B and fig. S9D). Actin binding
disordered. (B) The first (left) and second (right) subunits at the pointed end adopt a G-actin conformation, site 2 (ABS2, L161 to T348) consists mostly of
where the outer domain is rotated ~20° (red arrow) compared with subunits in the middle of F-actin a leucine-rich repeat (LRR) domain, which
(gray). Subunits were superimposed based on their inner domains (surface representation) to highlight forms a wedge-like structure that inserts into

y
differences in the orientation of their outer domains (ribbon representation). a cleft at the interface between the first three

B
C

y g
,

Fig. 4. Tmod-capped pointed end. (A) Domain diagram of Tmod, comprising
alternating tropomyosin (TMBS1 and TMBS2, disordered in the structure) and
actin (ABS1 and ABS2) binding sites. (B) Cryo-EM map of the Tmod-capped
pointed end at 3.26-Å resolution, showing the first and second subunits in two
shades of green and Tmod in orange. The view on the right is approximately
down the longitudinal axis of F-actin and shows a ribbon representation of Tmod.
ABS1 caps the first actin subunit, whereas ABS2 wedges into a cleft formed
by the first three actin subunits. Most of the interactions with actin are mediated
by the b-strand to a-helix loops of the LRR domain of ABS2. Tmod residues
involved in the interaction are shown and labeled in fig. S9D. (C) The first
actin subunit (left) adopts a G-actin conformation with the outer domain
rotated by ~20° (red arrow) compared with subunits in the middle of F-actin
(gray). The second actin subunit (right) adopts an F-actin conformation.
Subunits were superimposed based on their inner domains (surface represen-
tation) to highlight differences in orientation of their outer domains
(ribbon representation).

Carman et al., Science 380, 1287–1292 (2023) 23 June 2023 4 of 6

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

pointed-end subunits. The loops that connect

the b strands to the a helices of the LRR do-
main account for most of the interactions
with actin subunits (Fig. 4B). We had previ-
ously proposed a model of Tmod at the pointed
end that was based on crystal structures of
ABS1 and ABS2 bound to G-actin (18). This
model accurately predicted Tmod’s binding
mode, but not the conformation of actin sub-
units at the pointed end, which is crucial to
understand the details of this interaction.
Thus, another reason for Tmod’s inefficient
pointed-end binding appears to be that it re-
quires a conformational change that forces
the second subunit into an F-actin conforma-
tion, whereas the first subunit remains in the
G-actin conformation (Fig. 4C and table S2).
The interaction of ABS2 with the three pointed-
end subunits is likely responsible for the
conformational change in the second subunit,
because binding of ABS2 would be sterically

hindered for the G-actin conformation. In a
recently reported structure of the spectrin-
actin complex (27), the pointed end is capped
by Tmod together with a previously unknown
protein (SH3BGRL2). This complex also con-
tains tropomyosin, whose position is con-
p
Fig. 5. Model of subunit association and dissociation at the free and capped ends of F-actin. Structures
described here show that subunits in F-actin have different conformations depending on whether they are in
the middle or at the ends of F-actin, and are different from those of G-actin. Notably, these are not
nucleotide-dependent conformational differences, which are relatively minor in both G- and F-actin (see text).
The conformational differences at the ends of F-actin correlate with the association and dissociation

strained by interactions with several other
proteins. Likely because of these added inter-
actions, the first actin subunit in the spectrin-
actin complex has an F-actin conformation
and tropomyosin is shifted azimuthally by
g
constants of subunits at the ends of F-actin. Only the main pathway at equilibrium is depicted, with ATP-actin
preferentially adding to the barbed end and ADP-actin dissociating from the pointed end [see (1) for other
possible reactions]. Structural differences explain the asymmetric association of ATP-actin monomers to
the barbed and pointed ends of F-actin, with ATP-bound monomers more likely to undergo the G- to F-actin

~2.0 Å compared with its apparent position
in our Tmod-tropomyosin map (fig. S8).
Conclusions
We found that the ends of F-actin display
conformational differences that correlate with
kinetic differences in subunit addition and dis-
sociation (Fig. 5 and movie S1). Notably, these
structural differences depend on the position of
subunits in the filament and not on the
y
transition required for preferential binding to the barbed end than ADP-bound monomers. CapZ and
Tmod inhibit subunit exchange at the barbed and pointed ends, respectively, by structrual mechanisms
revealed in this study.
can be viewed as conceptually analogous to evidence and supports fundamental conclusions
the rate-limiting step during nucleation, i.e., about the structure, kinetics, and capping in-
the formation of a polymerization seed from teractions at the ends of the actin filament.
G-actin subunits (26). Tmod forms a knot-like
structure around the first three subunits, forc- REFERENCES AND NOTES
ing the second subunit into an F-actin confor- 1. T. D. Pollard, Handb. Exp. Pharmacol. 235, 331–347 (2017).

nucleotide state, which is responsible only for
minor differences both in the monomer (28, 29)
and the filament (5, 14, 15). Consistent with
early studies of filament elongation and de-
polymerization kinetics (1, 6), subunits at the
mation and inhibiting subunit addition and
dissociation by sterically blocking the pointed end.
The nucleotide state controls the association
and dissociation kinetics at the ends of F-actin
(1, 6). The ADP-bound structures described here
y g
2. R. Dominguez, K. C. Holmes, Annu. Rev. Biophys. 40, 169–186
(2011).
3. R. Dominguez, Trends Biochem. Sci. 29, 572–578 (2004).
4. T. Oda, M. Iwasa, T. Aihara, Y. Maéda, A. Narita, Nature 457,
441–445 (2009).
5. W. Oosterheert, B. U. Klink, A. Belyy, S. Pospich, S. Raunser,
Nature 611, 374–379 (2022).

barbed end have an F-actin conformation
and are thus structurally primed for subunit
addition. Yet, their exposed hydrophobic cleft
have a G-actin–like conformation, which may
explain why many proteins that bind G-actin
can also stably or transiently bind the barbed
end, including profilin (1, 30), gelsolin (31),
formins (7), twinfilin (23, 32), and WH2 domain–
containing proteins (33). CapZ undergoes major
structural changes upon binding to the barbed
end, which remains mostly unaffected by this
interaction.
An unexpected finding was that subunits at
the pointed end adopt a G-actin conformation
and thus appear primed for dissociation and less
likely to accept incoming subunits. According
to this model, slow pointed-end association
are directly relevant to the situation at the
pointed end, which accumulates ADP-actin.
At the barbed end, ATP hydrolysis likely occurs
immediately upon subunit addition, followed
by slow g-phosphate release. Yet, as in the fila-
ment (5, 14, 15), the structure of the barbed end
is unlikely to change much as a function of the
bound nucleotide. Therefore, we suggest that
the structural differences observed in this work
explain the asymmetric association of ATP-
actin monomers to the barbed and pointed
ends of F-actin, with ATP-bound monomers
being more likely to undergo the G- to F-actin
transition required for preferential binding to
the barbed end than ADP-bound monomers
(Fig. 5). In this way, this work provides a mech-
anistic understanding of years of biochemical
,
6. T. D. Pollard, J. Cell Biol. 103, 2747–2754 (1986).
7. D. R. Kovar, E. S. Harris, R. Mahaffy, H. N. Higgs, T. D. Pollard,
Cell 124, 423–435 (2006).
8. S. D. Hansen, R. D. Mullins, J. Cell Biol. 191, 571–584
(2010).
9. D. Breitsprecher et al., EMBO J. 30, 456–467 (2011).
10. T. Kotila et al., Nat. Commun. 10, 5320 (2019).
11. S. Shekhar, J. Chung, J. Kondev, J. Gelles, B. L. Goode,
Nat. Commun. 10, 5319 (2019).
12. M. Edwards et al., Nat. Rev. Mol. Cell Biol. 15, 677–689
(2014).
13. V. M. Fowler, R. Dominguez, Biophys. J. 112, 1742–1760
(2017).
14. F. Merino et al., Nat. Struct. Mol. Biol. 25, 528–537 (2018).
15. S. Z. Chou, T. D. Pollard, Proc. Natl. Acad. Sci. U.S.A. 116,
4265–4274 (2019).
16. T. Bepler et al., Nat. Methods 16, 1153–1160 (2019).
17. W. M. Schmidt, W. Lehman, J. R. Moore, Cytoskeleton 72,
292–303 (2015).
18. J. N. Rao, Y. Madasu, R. Dominguez, Science 345, 463–467
(2014).
19. J. Funk et al., Nat. Commun. 12, 5329 (2021).

Carman et al., Science 380, 1287–1292 (2023) 23 June 2023 5 of 6

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

20. L. Urnavicius et al., Nature 554, 202–206 (2018).
21. M. Hernandez-Valladares et al., Nat. Struct. Mol. Biol. 17,
497–503 (2010).
22. S. Takeda et al., PLOS Biol. 8, e1000416 (2010).
23. D. M. Mwangangi, E. Manser, R. C. Robinson, Sci. Adv. 7,
eabd5271 (2021).
24. J. P. Cartailler, H. Luecke, Structure 12, 133–144 (2004).
25. A. Narita, S. Takeda, A. Yamashita, Y. Maéda, EMBO J. 25,
5626–5633 (2006).
26. D. Sept, J. A. McCammon, Biophys. J. 81, 667–674 (2001).
27. N. Li et al., Cell 186, 1912–1929.e18 (2023).
28. L. R. Otterbein, P. Graceffa, R. Dominguez, Science 293,
708–711 (2001).
29. P. Graceffa, R. Dominguez, J. Biol. Chem. 278, 34172–34180
(2003).
30. N. Courtemanche, T. D. Pollard, Biochemistry 52, 6456–6466
(2013).
31. S. Nag, M. Larsson, R. C. Robinson, L. D. Burtnick, Cytoskeleton
70, 360–384 (2013).
32. M. Hakala et al., Nat. Cell Biol. 23, 147–159 (2021).
33. M. F. Carlier, C. Husson, L. Renault, D. Didry, Int. Rev. Cell
Mol. Biol. 290, 55–85 (2011).
ACKN OWLED GMEN TS
Data collection was performed at the Electron Microscopy
Resource Lab (EMRL) and The Beckman Center for Cryo-Electron
Microscopy, University of Pennsylvania (Research Resource
Identifier SCR_022375). We thank M. Boczkowska for assistance
with sample preparation and S. Steimle for assistance with data
collection. Funding: This study was supported by National
Institutes of Health (NIH) grant R01 GM073791 (R.D.) and NIH
grant F31 HL156431 (P.J.C.). Computational resources were
supported by NIH grant S10 OD023592. Author contributions:
Conceptualization: P.J.C., K.R.B., and R.D.; Methodology: P.J.C.,
K.R.B., and R.D.; Investigation: P.J.C., K.R.B., and R.D.; Resources:
P.J.C., K.R.B., and G.R.; Visualization: P.J.C., K.R.B., and R.D.;
Funding acquisition: P.J.C. and R.D.; Project administration: R.D.;
Supervision: R.D.; Writing – original draft: R.D.; Writing – review
and editing: P.J.C., K.R.B., and R.D. Competing interests: The
authors declare no competing interests. Data and materials
availability: Molecular models and cryo-EM density maps have
been deposited with the following accession codes: CapZ-capped
barbed end (PDB ID 8F8Q, EMD-28933), free barbed end (PDB
ID 8F8R, EMD-28934), free pointed end (PDB ID 8F8S, EMD-
28935), Tmod-capped pointed end (PDB ID 8F8T, EMD-28936),
and F-actin in the ADP state (PDB ID 8F8P, EMD-28932). All other
data necessary to evaluate the claims in this paper are present
in the text or supplementary materials. License information:
Copyright © 2023 the authors, some rights reserved; exclusive
licensee American Association for the Advancement of Science. No
claim to original US government works. https://www.science.org/
about/science-licenses-journal-article-reuse
SUPPLEMENTARY MATERIALS
science.org/doi/10.1126/science.adg6812
Materials and Methods
Figs. S1 to S9
Tables S1 and S2
References (34–41)
MDAR Reproducibility Checklist
Movie S1
View/request a protocol for this paper from Bio-protocol.
Submitted 13 January 2023; accepted 17 May 2023
Published online 25 May 2023
10.1126/science.adg6812

p
g
y
y g
,

Carman et al., Science 380, 1287–1292 (2023) 23 June 2023 6 of 6

 WORKING LIFE
By Lei Shen

Three labs in three years

I
used to believe a perfect postdoc lab was out there waiting for me. The research would be
exciting, the members happy, the funding ample. It would be similar to my doctoral lab, where
I had a great experience with a passionate principal investigator (PI) and supportive coworkers.
And when I started my first postdoc, I thought I had found it. I had interviewed with several PIs
and received a job offer from one that seemed to check all the boxes for me. The projects were
promising, the lab had a solid track record for productivity and funding, the PI and senior post-
docs were very knowledgeable, and the lab environment seemed friendly. So, eager and enthusiastic,
I had taken the job.

p
A few months after I started, the those risks the first time around;
pandemic hit. Despite the chal- I was focused on finding a place
lenging circumstances, the PI’s where I could be happy, intellec-
expectations and our workloads tually stimulated, and supported.
didn’t change. I was also assigned Should I make matters worse by

g
an additional COVID-19–related changing jobs again?
project, and I had a hard time Ultimately, I decided that being
moving my research forward. My satisfied with my working envi-
mental health suffered, and my PI ronment was worth the downside

y
was not able to offer the support of having an unusual CV. Trying
I needed. The lab no longer felt to persevere in a lab where I was
perfect at all. unhappy would inevitably lead to
But I still held on to that elu- an unsatisfactory CV anyway. I
sive idea of perfection. And I was would keep looking—maybe not
going to keep looking for it. for a mythical perfect lab, but for
As an international scholar I a lab that would be right for me.
was on a visa; to remain in the I emailed professors across the
country, I needed to stay at my in-
stitution and maintain the same
“I have learned there is university who were doing work
I was interested in, which led me
no perfect. There is

y g
job title. That limited my options, to my current lab. I was impressed
but I was determined to find my by the PI’s knowledge and pas-
dream lab. I found a job posting only a lab that suits me.” sion, and touched by the way she
in a lab that looked appealing and listened to and supported all the
applied. I explained to the PI that I had left my previ- members of the lab. I explained to her why I had left my pre-
ous lab because of the lack of mental health support; I’m vious two labs, and again I was grateful to be understood.

grateful they understood. After talking with the PI and
various lab members, I got the job offer. And again, I felt
optimistic I had found my perfect lab.
To my surprise and disappointment, I quickly discov-
ered that was not the case. The lab was very organized and
effective—but for me, the PI’s management style didn’t of-
fer the lab members enough intellectual space to explore
and grow. After just a few months, I once again found my-
self considering looking for a new lab.
I was warier this time because of a conversation with
a former colleague after I had left my first postdoc. He
had taken me aside to tell me that leaving had been a
silly mistake that could jeopardize my career. I had lost
a good reference and given potential future employers
the impression I couldn’t keep a job. I hadn’t considered
,
My current lab may not be as productive as my first lab
or as organized as my second one. But after three rounds
of postdoc “rotations,” I have learned there is no perfect.
There is only a lab that suits me.
Maybe I should been quicker to realize what really mat-
ters to me in a lab and find the right fit. But no matter
how unconventional my postdoc experience might look on
paper, I’m grateful for the experience I had in all three labs.
I learned valuable things along the way—about research,
and about myself. It all built me into a more confident, pro-
fessional, and skillful scientist and person. And I hope any
future employers can see that as a strength, not a liability. j
Lei Shen is a postdoc at Yale University. Send your career story to
SciCareerEditor@aaas.org.

1294 23 JUNE 2023 • VOL 380 ISSUE 6651 science.org SCIENCE