Science - July 8, 2022 USA

E D I TOR IAL
Inclusion doesn’t lower standards
T
he cultural wars raging across the United States The most corrosive aspect of faculty members’ resis-
have sadly found their way into the world of tance to change is their response to students’ requests
science. Some university science faculty and for the use of inclusive language and terminology:
administrators are resistant to making changes “Can you believe this student asked me not to say
that would allow more students from under- this?” Apparently, faculty cannot be inconvenienced
represented groups to participate and thrive in to modify language in a way that creates a supportive
the sciences. The rationale for this opposition is learning environment. Rather than mocking the stu-
often that “accommodating” legitimate social and peda- dents for asking, educators should be inspired to learn H. Holden Thorp
gogical needs of marginalized groups will lower the new ways of communicating that make it easier for all
Editor-in-Chief,
standards of mastery and excellence in these fields. But students to learn. Institutions need to take responsi-
Science journals.
this concern is just a crutch that protects faculty and bility for the needed changes and not put the burden
hthorp@aaas.org;
institutions from having to do the work of correcting on students. Some proposed solutions have even in-
social injustices in higher education. volved sending all the students to cognitive behavioral @hholdenthorp
It’s common to hear that improving student diversity therapy to make them less sensitive. Why should they
in higher education requires lowering the bar to admis- be the ones to do the work needed to fix a dysfunc-
sion and watering down the curric- tional world they did not create?
ulum so that all students can pass The University of Chicago has
the course of study. I’m not aware engaged in a very effective public
of anyone who is advocating such a
trade-off. There are known methods
“...everyone relations campaign over the past
several years to brand itself as the
of teaching that allow more people
from different backgrounds to mas-
should have more “free-speech university.” It has is-
sued many bold statements from
ter scientific material without com-
promising the quality of education.
optimism about faculty and administrators that
welcome all ideas about contro-
These include a greater use of active
learning methods that engage stu-
who can become versial issues and create a space
for conservative ideas to be shared
dents with course material through
discussions and problem solving (as
a scientist.” freely. Many faculty opposed to di-
versity initiatives have praised the
opposed to passively taking in infor- stance and assume that the univer-
mation). Making such reforms may sity does not favor diversity initia-
require faculty to learn new ways of teaching. But isn’t tives or staff who are devoted to equity and inclusion.
that the job—to foster education for everyone? But the institution’s website for the division of student
Another common refrain is that understanding sci- affairs reveals many identity-based resources, such
ence requires a high degree of skill in mathematics. I’ve as a team that helps students report and respond to
heard firsthand from faculty that students can’t pass bias incidents, a center for identity and inclusion, and
their classes unless they have previously achieved a much more. The University of Chicago understands
high score on standardized tests in math such as the that all students need support and resources to learn.
SAT or ACT. That is a breathtakingly pessimistic view. I don’t think anyone would accuse this institution of
These high scorers are often students who’ve had the lowering its standards.
opportunities and resources to prepare for pre-college Opening the doors to science for everyone requires
exams, which vast numbers of students have difficulty that faculty learn the most effective methods for teaching
accessing. Isn’t the whole point of teaching to provide a a diverse student body. Yes, it’s more work on top of the
pathway to achievement? many other faculty duties, so universities must provide
Social psychologist Claude Steele and others have dem- resources to make the adjustments, such as revamping
onstrated repeatedly that stereotype threat—in which classrooms for active learning, providing time for faculty
symbols and signals of exclusion create negative psycho- to redo their curricula, and doing the hard work involved
logical effects—measurably impairs learning. Names of in having the faculty and institution make the cultural
buildings and sports teams that honor racism and racists changes that students need. And everyone should have
send a discouraging message to marginalized groups. If more optimism about who can become a scientist.
PHOTO: CAMERON DAVIDSON
removing symbols of exclusion is helpful, why wouldn’t a

university want to do that? – H. Holden Thorp
10.1126/science.add7259
SCIENCE science.org 8 JULY 2022 • VOL 377 ISSUE 6602 129

NEWS
We just hope we didn’t open a Pandora’s box.
“ Artificial intelligence scientist Almira Osmanovic Thunström, in Scientific American,
”
discussing research articles written by AI algorithms, including one she submitted to a journal.
IN BRIEF Colombia hits conservation goal

Edited by Shraddha Chakradhar
| Colombia is the first
C O N S E R VAT I O N
Western Hemisphere country to achieve
RESEARCH SECURITY the widely adopted 30x30 ocean conserva-
tion goal, Colombian President Iván Duque
New debate over secrecy claimed last week at the United Nations
Oceans Conference in Lisbon, Portugal.
S
cience has learned that the U.S. National Science Foundation More than 100 other countries have signed
(NSF) has asked the National Academies to take a fresh look the pledge to protect 30% of both Earth’s
this fall at a Cold War-era presidential directive that regards ocean area and land area by 2030. To reach
that proportion of protected ocean area in
openness in basic research as a boon to both innovation and Colombia 8 years before the deadline, Duque
national security. Advocates of classifying as little informa- designated three new marine protected
tion as possible say additional restrictions would harm U.S. areas and expanded the existing Malpelo
research without deterring countries seen as adversaries. China’s Fauna and Flora Sanctuary in the Eastern
Pacific Ocean to include a 1400 kilometer
aggressive pursuit of several emerging technologies with both com- underwater ridge that acts as a highway for
mercial and military uses, however, has prompted calls from many sharks and other marine life. Duque has
lawmakers to cordon off the results of basic research on a range of also announced a $245 million initiative
sensitive technologies. A 2019 report to NSF backed the current pol- that aims to protect 32 million hectares of
Colombian land and ocean.
icy, issued in 1985 and known as NSDD-189. But NSF officials say the
world has changed enough for experts across academia, government,
and industry to reexamine the issue. Germany military research urged
RESEARCH FUNDING | German universities
should undertake more military research,
the German national academy of engineer-
wide as a pingpong table is long—are the ing, Acatech, said in a white paper released
Court limits EPA’s climate power world’s largest, they report in Frontiers in last month. The Russian attack on Ukraine
| Siding with a group
C L I M AT E P O L I C Y Plant Science. Victoria boliviana, so named has prompted a shift in Germany’s restrained
of Republican-led states and the electric because it grows the biggest in Bolivia, was approach to defense, predominant since
power industry, the U.S. Supreme Court shown to be a distinct species with genetic World War II, with the legislature recently
last week limited the Environmental analysis and detailed morphological approving €100 billion in new military fund-
Protection Agency’s (EPA’s) ability to observations of specimens at the National ing. Many German universities have long
regulate carbon emissions from existing Herbarium of Bolivia and at the Royal had a voluntary ban on military and dual-use
power plants. Analysts say the 6-3 deci- Botanic Gardens Kew in England, where research, but that position is out of date
sion in West Virginia v. EPA will hamper the new species is on display. since the war in Ukraine, the 24 June report
President Joe Biden’s efforts to achieve his
goal of substantially reducing U.S. carbon
emissions and give more power to states.
Notably, however, the court’s conservative
majority did not use the case to challenge
EPA’s underlying authority to regulate
greenhouse gases, especially carbon
dioxide, the primary planet-warming gas.
But federal courts are still considering
PHOTO: ROYAL BOTANIC GARDENS, KEW
other lawsuits that would further reduce

the agency’s options for regulating carbon
dioxide emissions.
New water lily species named

| Researchers have found a
E C O L O GY
new species of water lily, and its leaves—as Scientists have discovered and named a new species of giant water lily.
134 8 JULY 2022 • VOL 377 ISSUE 6602 science.org SCIENCE

Swallow-tailed hummingbirds are
among the species found to have a
plethora of new feather colors.
ECOLOGY
Scientists find new hummingbird colors
T
he plumage of hummingbirds has more color diversity than the
feathers of all other birds combined, a recent study finds. Researchers
from Yale University collected feathers from specimens of 114 humming-
bird species and, using a spectrometer, documented the wavelengths
of light they reflected. These wavelengths were then compared with those
found in a previous study of 111 other bird species, including penguins and parrots.
The researchers were surprised to find new colors in the hummers, which widened the
known avian color gamut by 56% and included rarely seen saturated greens and blues,
they report in Communications Biology. The newfound variation largely includes colors
in the ultraviolet scale that are invisible to humans and probably only seen by humming-
birds themselves. Researchers note the variation is likely due to the reflective qualities of
nanostructures present in the small barbs that protrude from the end of each humming-
bird feather. The new colors were mostly found on the crowns and throats of the birds,
suggesting a role in mating displays and communication.
says. It calls for German science to focus at-risk species. Among the most vulnerable
equally on security, resilience, and sustain- trees worldwide are the Magnolia ekmanii Airline ends monkey transports
ability and for a broad debate about what from Haiti and the Diospyros egrettarum, ANIMAL MODELS | Air France will stop
research is needed for the country’s security. which has fewer than 10 specimens left in transporting nonhuman primates once its
its native Mauritius. current contracts end. The airline was the
last major holdout on carrying nonhuman
Protections overlook at-risk trees primates as cargo. People for the Ethical
BIODIVERSITY | Initiatives to conserve Next-gen COVID-19 jabs coming Treatment of Animals lauded the deci-
trees largely focus on nonthreatened spe- | Redesigned COVID-19 booster
C OV I D -1 9 sion, but the European Animal Research
cies, according to an analysis presented shots are expected this fall, after advisers Association cautions it will restrict research,
last week at the World Biodiversity Forum. to the U.S. Food and Drug Administration especially as the European Union prepares
The work is an update to the State of the (FDA) voted 19-2 last week in favor of to introduce further limits on primate stud-
World’s Trees report, originally published incorporating an Omicron strain into exist- ies later this year. The U.S. already faces
in September 2021, that showed one-third ing vaccines. FDA concurred, announcing shortages of monkeys for research, in part
of the 60,000 tree species in the world are on 30 June that it’s asking manufacturers because of a Chinese ban on monkey exports.
at risk of extinction, fueled by clearing land to retool their shots for a spike protein
for agriculture, logging, and climate change. component shared by Omicron BA.4 and
BY THE NUMBERS
The new analysis reveals that areas marked BA.5, which are currently gaining ground
PHOTO: LUIZ CLAUDIO MARIGO/MINDEN PICTURES
for conservation, such as national parks, worldwide. The new vaccines, which Pfizer
protect 85% of nonthreatened tree species,
compared with only 56% of threatened
ones. The same is true for trees in scien-
tific collections, such as botanical gardens
and Moderna say they can make available
by about October, will have the same mes-
senger RNA dose as previous boosters but
will target both Omicron and the original
350,000
Google searches in the United States
and seed banks; only 21% of threatened coronavirus strain first detected in 2019. for “abortion pill” and names of specific
species are safeguarded there versus 45% Pfizer and Moderna have reported data abortion medications between 1 May
of nonthreatened ones. Researchers worry from ongoing trials of vaccines against an and 8 May, the week the U.S. Supreme
the lack of threatened trees in collections earlier Omicron strain, BA.1, and FDA says Court’s Dobbs v. Jackson Women’s
might hinder conservation and restoration it expects to consider these data in evaluat- Health Organization draft ruling was
programs aimed at growing and preserving ing fall boosters. leaked. (JAMA Internal Medicine)

IN DEP TH
Demonstrators gathered outside the Supreme Court the morning after the court overturned Roe v. Wade, effectively outlawing abortion in many states.
WORKFORCE
Scientists decry reversal of U.S. abortion rights

For some, the ruling limits professional mobility and conference attendance
By Katie Langin become pregnant. “I really worry that it’s go- “you don’t have a ton of choice about where
ing to affect people’s ability to write their own you end up in the first place, and now there’s
W
hen the U.S. Supreme Court over- destiny,” says an M.D.-Ph.D. student in Texas even less choice if you cut out half the states
turned Roe v. Wade on 24 June, who requested anonymity. and say, ‘Well, I’m not willing to live there.”
eliminating the constitutional The Supreme Court’s reversal will likely be But ultimately, she doesn’t think she’d feel
right to an abortion and hand- felt most strongly in groups that are already comfortable recruiting trainees to join her in
ing decisions about access to state underrepresented in science, says Nicole a state that doesn’t grant them reproductive
legislators, the response across Williams, the outreach director for the non- autonomy. As a second-year Ph.D. student,
the polarized country was swift and divided. profit 500 Women Scientists. “Being an she became pregnant unexpectedly after
Abortion opponents cheered the culmination African-American woman, and just knowing her contraception failed. Her subsequent
of a long quest to restrict the practice. But the stats—that Black birthing persons already decision to end the pregnancy was relatively
large swaths of the public, including many experience high levels of pregnancy-related easy, she says, because she wanted to focus
scientists, decried the decision as a poten- mortality—the overturning of Roe versus on her education—and she wants that same
tially deadly violation of human rights. “Abor- Wade is a death sentence for Black women freedom for others. “Being forced to carry an
tion bans will kill people in lots of different scientists and birthing persons,” including unwanted pregnancy to term during gradu-
horrible ways,” tweeted Amanda Stevenson transgender men and nonbinary people. ate school would have posed a significant
of the University of Colorado, Boulder, a re- These concerns have some scientists recon- burden on me and potentially derailed my
PHOTO: T.J. KIRKPATRICK/THE NEW YORK TIMES/REDUX
searcher who studies abortion. sidering their career plans and where they’re career plans,” she says.
Some scientists also worried about direct willing to live and work. The Texas M.D.- Faculty members already established in af-
effects on the research community. “It’s go- Ph.D. student, for example, might leave the fected states are wrestling with similar ques-
ing to really negatively impact science … if we state once she finishes her program because tions. “I had to take a mental health day just
have scientists actively avoiding half of this of Texas’s strict abortion laws. “It’s really to process everything that happened and deal
country, or all of it entirely,” says Rosa Lafer- hard. … I love Texas,” she says. Researchers with the emotions,” says an assistant profes-
Sousa, a neuroscience postdoc in the Wash- across genders and career stages shared simi- sor in a biomedical field who is based in a
ington, D.C., area who is considering how the lar stories on social media, saying they would southern state where abortion is now illegal
ruling will influence her upcoming faculty be leaving or not pursuing professional opin nearly all cases. Speaking with Science on
job search. She and others fear that the lack portunities in abortion-restricting states. the condition she remain anonymous, she
of an abortion option in many states will cre- Those seeking faculty positions face par- says she’s particularly concerned about stu-
ate hardship for some aspiring scientists who ticular challenges, Lafer-Sousa says, because dents at her university who come from dis-

N E WS
advantaged backgrounds and may not be MICROBIOLOGY

able to afford to travel to another state if they
need an abortion. “Will I ever have to … send
a student of mine to a ‘conference’ in Califor-
nia?” she wonders. “Is that something that I
Dengue and zika viruses turn
need to start thinking about?”
She’s not sure she wants to stick around to
find out, but whether to leave is a hard deci-
people into mosquito bait
sion. She’s the only racial or ethnic minority To spread, pathogens drive mice, people to make odorant
in her department and she feels she could
have more impact where she is than in more
liberal states—“both in terms of my votes, but By Mitch Leslie quitoes preferred the bouquet of dengue.
also in the mentoring and support I can pro- By analyzing molecules emanating from
T
vide the underrepresented minority trainees he viruses that cause Zika and dengue infected rodents, the researchers identified
that exist here,” she says. “I’m Latina. I’m a fever can’t get from person to person the ingredients in eau de Zika or dengue.
woman. And in science, both of those identi- on their own—they need to hitchhike Mice gave off larger quantities of 11 poten-
ties are not well represented.” But come fall, inside a mosquito. A study last week tial odorants when they got sick, and one,
she’ll likely apply for faculty positions else- in Cell suggests they hail these rides acetophenone, was a mosquito attractant.
where. “The fear for my safety and my well- by making their victims smell more at- Rodents that were ill exuded about 10 times
being—and even more so than that, that of tractive to the blood-suckers. more acetophenone than their uninfected
my students—it weighs really heavily on me.” It’s “a big advance,” says mosquito neuro- counterparts. Dengue patients also emitted
Some universities have issued statements scientist Laura Duvall of Columbia Univer- more of the molecule than healthy people.
expressing support for their students’ and sity. The work shows that “infection with Certain bacteria that dwell on the skin
employees’ reproductive rights. The M.D.- these mosquito-borne viruses can alter the are the main source of acetophenone. Skin
Ph.D. student, for instance, says her school way some people smell … to make them more cells normally keep their numbers in check
administrators sent out an email assuring likely to be bitten.” with a protein called RELMa that kills the
trainees and faculty they will be supported Malaria parasites were already known to microbes. However, Cheng and colleagues
to the extent possible. But other universities make human hosts smell irresistible to mos- found that mice infected with the Zika or
have remained silent, frustrating those who quitoes. Whether the viruses dengue viruses cranked out
want to know their employer is paying at-
tention. The southern professor, for example,
that cause Zika fever and den-
gue fever, which together infect
“These much less RELMa, which
might allow the bacteria to
hasn’t received any emails or statements from up to 400 million people every mosquito-borne proliferate and change the an-
her university. It “pisses me off,” she says. year, also meddle with a per- imals’ scent. And mosquitoes
Some scientists have also called on their son’s odor was unclear, however. viruses can alter were less fond of mice that
professional societies to take action, par-
ticularly regarding conference locations,
These pathogens travel from
person to person in Aedes ae-
the way some had consumed isotretinoin, a
derivative of vitamin A that
with some proposing boycotting meetings in gypti mosquitoes, which trans- people smell.” increases RELMa synthesis
states that ban abortion. Others have pushed mit other viruses as well. “It’s a very compelling pa-
Laura Duvall,
back on that idea, saying it will only harm sci- To determine whether the per,” says Ring Cardé of the
Columbia University
entists in those states. But safety of attendees insects are partial to individu- University of California, Riv-
is at stake, boycott advocates say. “Any person als with Zika or dengue, microbiologist Gong erside, who studies chemical ecology and
capable of childbearing could have an emer- Cheng of Tsinghua University and colleagues insect behavior. But he cautions that other
gency related to pregnancy at your confer- set up three interconnected cages for a mouse research teams have uncovered numerous
ence,” Northwestern University neuroscience experiment. Into one cage they piped air that odor molecules that draw A. aegypti mos-
postdoc Ana Vlasits tweeted. “Your event had blown across mice that were sick with quitoes to their victims, including lactic acid
should not be held in a place where your the Zika virus. A second cage received air that and ammonia. “It’s not clear how this com-
childbearing colleagues might be put at risk.” had flowed over healthy mice. The team then pound fits in with the known attractants.”
Those concerns make sense to Catherine added hungry mosquitoes to the third cage Still, the results could “revolutionize”
Alves, a social scientist based in Rhode Island and let them choose where to hang out. diagnosis of dengue and Zika, says James
who is currently pregnant. Last year, she had Seventy percent of the mosquitoes Logan of the London School of Hygiene
a miscarriage and took the drug misoprostol crowded into the cage receiving air from the & Tropical Medicine, who was part of the
to help her body expel the fetal tissue. Her Zika-infected rodents. The insects’ distribu- team that showed malaria parasites change
situation wasn’t an emergency, but other tion was equally lopsided when the air came people’s skin chemistry. Instead of a blood
pregnancy-related complications that are of- from rodents with dengue fever rather than test, an electronic “nose” that detects ace-
ten treated with abortion drugs, such as ec- Zika. However, the mosquitoes didn’t favor tophenone could provide a diagnosis much
topic pregnancies, can be fatal if they aren’t a particular cage when air from the infected more quickly and easily, Logan suggests. A
addressed quickly. It’s not clear whether animals’ cages went through a filtration ap- company he founded is already developing
those drugs will continue to be prescribed paratus that trapped chemicals, suggesting sensors to identify malaria from body odor.
in emergency situations in all states. “As a the odor of the sick mice was key. The findings suggest “a novel avenue” to
pregnant person,” Alves says, she wouldn’t Cheng’s team also wiped the armpits battle these diseases by reducing human
feel comfortable attending conferences in of healthy people and dengue fever pa- attractiveness to mosquitoes, Cheng says.
locations where, if something went awry, “I tients with an absorbent material, isolated For example, he and colleagues are test-
couldn’t get medical care that I needed and molecules that could become airborne, ing isotretinoin and related compounds in
that reflects my values.” j and dabbed them onto filter paper. Mos- dengue fever patients. j

NEWS | I N D E P T H
Lab studies show that one of the coronavirus’ protein-

cutting enzymes can mutate at several sites (orange)
that make it more resistant to the active ingredient in
Paxlovid (shown here as gray and colored spheres).
United States issued only 40,000 or fewer

prescriptions per week through mid-April.
Since then, prescriptions have surged to
more than 160,000 per week, according to
the latest numbers from the Centers for Dis-
ease Control and Prevention.
That rise creates selective pressure on
the virus, favoring mutations that help it
survive in the presence of the drug. And
because each infected person makes tril-
lions of copies of SARS-CoV-2, the virus has
plenty of opportunities to test out different
mutations as it replicates. So far, those mu-
tations don’t seem to have interfered with
Paxlovid’s effectiveness. Nirmatrelvir pre-
vents SARS-CoV-2’s main protease (Mpro)
from cutting a long precursor molecule
made by the virus into shorter active pro-
COVID-19 teins, an essential step in SARS-CoV-2’s re-
production. In February, Pfizer researchers
Bad news for Paxlovid? reported in JBC Accelerated Communica-

tions that nirmatrelvir remained effective
in halting the activity of Mpro in multiple
Resistance may be coming SARS-CoV-2 variants, including Alpha,

Beta, Delta, Gamma, Lambda, and Omi-
cron, as well as the original strain.
In lab studies, SARS-CoV-2 finds ways to evade key drug. But like many antiviral drugs, Paxlovid
Some of the viral mutations are already found in people may falter as the virus becomes resistant to
it. Preprints posted on bioRxiv on 7 June,
for example, show that SARS-CoV-2 grown
By Robert F. Service “Given the amount of infections out there, in the lab quickly gains the ability to avoid
it’s going to come,” Ho says. nirmatrelvir’s attack. Two research groups
P
rescriptions for Pfizer’s blockbuster The resistance studies come on the heels independently cultured the coronavirus with
drug Paxlovid have skyrocketed in of other recent concerns about Paxlovid, low levels of nirmatrelvir, killing some but
recent weeks. That’s good news for which in the United States remains re- not all of the virus. Such tests are meant to
many COVID-19 patients, as the pill stricted to use in people with risk factors simulate what might happen in an infected
has been proven to reduce severe making them more likely to develop severe person who doesn’t take the whole regimen
disease. But a bevy of new lab studies COVID-19. Confirming anecdotal reports of the drug or an immunocompromised pa-
show SARS-CoV-2 can mutate in ways that widely reported by media, several studies tient who has trouble clearing the virus.
make it less susceptible to the drug, by far have found a small percentage of infected One of those studies, led by Dirk
the more widely used of the two oral anti- people who receive the normal 5-day Jochmans, a virologist at KU Leuven in
viral drugs authorized to treat COVID-19 course initially feel better, only to have Belgium, found that after 12 rounds of
in the United States. Researchers have their symptoms rebound. And questions nirmatrelvir treatment, SARS-CoV-2 accu-
found some of those mutations in variants have grown about whether Paxlovid helps mulated three mutations—at positions 50,
already circulating in infected people, rais- those who aren’t at high risk of serious 166, and 167 in the string of amino acids
CREDITS: (GRAPHIC) V. ALTOUNIAN/SCIENCE; (DATA) RCSB PRO
ing fears that physicians could soon lose disease—Pfizer earlier this month halted that make up Mpro–that reduced the virus’
one of their best tools for fighting the virus. a large trial of the drug in standard risk susceptibility to nirmatrelvir 20-fold, as
Taken together, the studies show that COVID-19 patients because it was failing determined by the dose of drug required
“when you put pressure on the virus it es- to show statistically significant protection to kill half the virus in a sample. The other
capes,” says David Ho, a virologist at Co- against death or hospitalization. study, led by Judith Margarete Gottwein,
lumbia University who was among the first The U.S. Food and Drug Administration a virologist at the University of Copenha-
to document drug resistance mutations in (FDA) granted emergency use authorization gen, also spotted potential resistance-con-
HIV some 3 decades ago. Ho was not infor Paxlovid in December 2021. The drug con- ferring mutations at positions 50 and 166
volved with the new studies but is conduct- sists of nirmatrelvir, the active antiviral, and in Mpro. When those mutations occurred
ing similar work on SARS-CoV-2. Although ritonavir, a compound that slows the break- together, the virus was 80 times less sus-
the newly identified mutations have yet to down of nirmatrelvir in the body. Because of ceptible to nirmatrelvir. “This tells us what
become widespread, Ho and many other bottlenecks in manufacturing nirmatrelvir, mutations we should be looking for [in pa-
scientists think it’s only a matter of time. Paxlovid’s rollout was slow—doctors in the tients],” Gottwein says.

Indeed, some of these mutations are recur in COVID-19 patients on Paxlovid, ECOLOGY
already in coronavirus-infected people, which happens in just 2% or fewer of those
according to work by Adam Godzik, a
bioinformatics expert at the University of
California, Riverside. Godzik and his col-
who take the drug, the rebound does not
seem to be due to resistance mutations.
“It’s reassuring,” Shah says, but not proof
It takes a
leagues scoured the GISAID database, a
catalogue of more than 10 million SARS-
CoV-2 genomes sequenced from viruses
the virus won’t eventually find its way
around the drug.
Pfizer says its Paxlovid regimen may
(microbial)
isolated from infected individuals, search-
ing for amino acid changes at positions in
Mpro near where nirmatrelvir binds. In a
forestall resistance. Patients only take
the drug for a short period and typically
get a dose “manyfold higher” than that
village to make
bioRxiv preprint posted on 30 May, they
reported that mutations to amino acids
required to prevent the virus from repli-
cating in cells, thereby minimizing the op-
an algal bloom
166 and 167—two of the resistance muta- portunities for the virus to mutate, says Kit
tions flagged by the Belgian group—were Longley, a company spokesperson. More than nutrient levels
already in circulating viruses. Because Giving patients multiple antivirals could may drive toxic lake growths
these mutations occurred before wide- help prevent resistance by making it harder
spread use of Paxlovid, they likely occurred for the virus to evolve its way around differ-
randomly, Godzik says. However, he adds, ent compounds at the same time, a strategy By Elizabeth Pennisi
they reveal the enzyme has some flexibility that has proved highly effective in treating
E
at these positions that could help the virus other viruses, including HIV and hepatitis very summer, surges of toxic green
work around the drug. C, Ho says. Two other SARS- muck plague lakes worldwide, sicken-
And the list of potential
resistance mutations keeps
“When you put CoV-2 antivirals are autho-
rized in the United States,
ing hikers who fail to purify drinking
water, closing favorite swimming holes,
growing. In a paper posted pressure on the but they have drawbacks. The and killing fish. The most feared—and
last week on bioRxiv, Jun other oral drug, molnupiravir, studied—cause of these freshwater “al-
Wang, a medicinal chemist at virus, it escapes.” has proven considerably less gal” blooms is a genus of cyanobacterium
Rutgers University, and col- David Ho, effective than Paxlovid, and called Microcystis. Its explosive summer
leagues report 66 common Columbia University has raised safety concerns growth is thought to be spurred by rising lev-
mutations to Mpro near the because it induces random ge- els of phosphorus, nitrogen, and other nutri-
nirmatrelvir binding site. Like Godzik’s netic mutations in the virus—that typically ents, perhaps from fertilizer run off or other
team, they scanned the GISAID database stops it from replicating but could also pollution sources. But new research, driven
to find altered versions of the protease, spawn dangerous new variants, some sci- by advances in DNA sequencing, suggests
but then went a step further. Adding the entists caution. And remdesivir, which in- other types of microbes also play key roles in
gene for each of these variants of Mpro to terferes with the ability of the virus to copy these massive overgrowths.
Escherichia coli bacteria, they created its genome, is approved for use in patients According to one study, viruses killing off
supplies of the mutated enzymes for ad- with mild to moderate symptoms who are a main competitor of toxic Microcystis may
ditional tests: first to determine whether at risk of severe disease, but it must be de- help pave the way for blooms; another indi-
each variant still carried out the essential livered intravenously. A preprint posted on cates nitrogen fixation by other bacteria may
duties of cutting viral proteins, and sec- bioRxiv last week suggests combining mol- provide the needed boost. The results suggest
ond to determine whether the mutations nupiravir and nirmatrelvir is more effective that reducing nutrients may not be enough to
allowed Mpro to resist nirmatrelvir. Eleven in combating SARS-CoV-2 infections than stop these slimy explosions, some scientists
of the 66 variants retained the protease’s either antiviral given alone, at least in mice. say. That doesn’t mean curbing pollution is
function (it was impaired in most of the But the strategy has yet to be widely em- unimportant, they stress, but ecological fac-
others), and five of the 11 were resistant braced by doctors. tors must be considered.
to nirmatrelvir, requiring at least a 10-fold Meanwhile, pharmaceutical compa- “Interspecies biological interactions help
increase in the drug to kill half the virus nies are racing to complete clinical tri- determine blooms,” says Kevin Johnson,
in the sample. One of those variants had als on additional SARS-CoV-2 antivirals, a marine scientist at the Florida Institute
a previously seen resistance mutation, at some targeting Mpro at different sites. But of Technology who was not involved in the
position 166, but the other four had novel those aren’t available yet. And numerous work. “The more details we understand of
workarounds at positions 144, 165, 172, and researchers, including representatives of bloom creation, the better our knowledge of
192. The bottom line from all this work, the nonprofit Drugs for Neglected Diseases how they might be prevented or controlled.”
Wang says: “It’s just a matter of time be- Initiative, have complained that Pfizer has With the warming climate and continuing
fore we see resistance emerge.” not made Paxlovid easily available for tri- inflows of pollution, harmful algal blooms
So, why is Paxlovid still working? One als of combination therapies. The company are on the rise, becoming more frequent and
possibility is that not enough people have has said it plans to do those studies itself, longer lasting in ever more places across the
taken the drug yet to force the virus to mu- although some are skeptical. globe. They are “a pretty wicked problem,”
tate. Another explanation, Wang says, is Until more antiviral drugs become says Ariane Peralta, a microbial ecologist at
that it may take multiple mutations in Mpro available, Paxlovid will remain essentially Eastern Carolina University.
for the virus to get around Paxlovid while alone, raising fears that sooner or later In some lakes, reducing fertilizer runoff
remaining both fully functional and easily it will lose its punch. When pressed by a at first seemed to thwart blooms—then they
transmissible. Thus far, adds Aditya Shah, single antiviral, viruses usually find a way came back. Similar plans for bloom-choked
an infectious disease specialist at the Mayo around the drug, Gottwein says. “If it can Lake Erie might backfire, a team of academic
Clinic, studies show that when symptoms happen it will happen.” j microbiologists and water quality experts

cue activates them. Then, after the viruses
start slaying more and more picocyano-
bacteria, newly available nitrogen, phos-
phorus, and more light fuel a Microcystis
bloom, Krausfeldt suggests. The phages’
destruction of its hosts’ cells may release
even more nutrients, playing a key role in
enabling algal blooms, she concludes.
Sheik, who says he had not considered
phages as a factor in blooms but now
wants to explore such viral dynamics, em-
braces Krausfeldt’s ecosystem mindset. “By
taking a holistic approach, we can better
understand how supporting organisms can
help sustain blooms,” he says.
Sheik and his colleagues have also added
metagenomics, as well as gene activity assess-
ments, to his studies of several small lakes in
Minnesota. Those lakes, he reported at the
meeting, contain not only some Microcystis,
but also another bloom-forming cyanobacte-
Toxins produced by this algal bloom in a lake in Utah sickened several people. rium called Dolichospermum. In 2020 and
2021, when he and colleagues tracked the mi-
funded by the National Science Foundation environments, to reconstruct the microbial crobial dynamics in one lake throughout the
and other U.S. agencies reported in May. A ecosystem in Florida’s Lake Okeechobee. summer, they saw Dolichospermum become
2014 bloom there caused such severe short- The largest lake in the U.S. southeast, the most abundant microbe only to have its
ages of drinking water in the nearby city of Okeechobee’s annual summer blooms have population crash by July. Nitrogen levels in
Toledo, Ohio, that Canada and the United begun to spread down rivers and spill into the lake rose and fell in parallel with the mi-
States have agreed to cut phosphorus going the Gulf of Mexico and Atlantic Ocean, crobe, suggesting it was fixing nitrogen and
into the lake by 40%. forcing beaches to close. Between April boosting its concentration in the water.
But a simulation of that strategy, along and September in 2019, the bloom season, Nitrogen is usually quite scarce in these
with an analysis of more than 100 related Krausfeldt and her colleagues collected relatively pristine lakes, yet the nutrient is
scientific papers, led the team to conclude multiple water samples at 21 places across essential for the production of microcys-
that although limiting phosphorus might the lake. From the fragments of DNA iso- tin. That might explain why Sheik and his
shrink Lake Erie blooms, they could also lated from the samples and sequenced, colleagues saw levels of Microcystis and
grow more toxic: with lower overall growth they pieced together whole genomes be- its toxin rise after the bloom in nitrogen-
of microbes, any photosynthetic Micro- longing to specific species. fixing Dolichospermum. Microcystis must
cystis left would receive more sunlight and The analysis uncovered 30 kinds of rely on other members of the freshwater
have more nitrogen available, two condi- cyanobacteria never before detected in ecosystem to fix nitrogen or to recycle it by
tions that favor an increase in their pro- the lake, and in some cases new to sci- breaking down other life forms, Sheik says.
duction of microcystin, a substance that ence, including 13 that could potentially “I’m blown away” by the metagenomic
make the blooms toxic (Science, 26 May, cause blooms, she reported last month at work, says Benjamin Wolfe, a microbio-
p. 1001). They suggested the lake’s nitrogen Microbe 2022, the annual meeting of the logist at Tufts University, because it can il-
should also be curtailed. American Society for Microbiology. “I was luminate in great detail the lake’s micro-
That simuation hinted that other micro- surprised at the diversity,” Krausfeldt says. bial interactions.
bes can indirectly influence the impact When there was no bloom, the most The case of Dolichospermum illustrates
of Microcystis. But researchers studying common organisms were the picocyano- how complicated algal blooms can be. The
blooms have tended to overlook lakes’ many bacteria. But as the season progressed, good news, however, is that unlike in Eu-
microbial inhabitants, which can include DNA belonging to bacterial viruses, known rope, where this bacterium causes toxic
huge numbers of diatoms and other eu- as phages, that infect the picocyano- blooms, Dolichospermum species in the
karyotes, as well as viruses and various bacteria rose steeply. Shortly thereafter, the United States lack the genes to make tox-
types of bacteria, including smaller than concentration of toxic Microcystis began to ins—at least for now, says Sheik, who plans
PHOTO: RICK EGAN/THE SALT LAKE TRIBUNE/AP IMAGES
average ones called picocyanobacteria. skyrocket. An analysis of its genome sug- to keep watching for them in his metage-
“Everyone glosses over them as not of gested why: Microcystis contains several nomic studies.
managerial concern,” says Cody Sheik, a antiviral defenses, such as the system that How the microbial dynamics that drive
microbial ecologist at the University of spawned the genome editor CRISPR, that blooms can be interrupted is still un-
Minnesota, Duluth. picocyanobacterial lack. In addition, the known, and the picture is getting more
Part of the problem has been that it’s bloom-forming cyanobacterium has genes complicated all the time. “We are grappling
been difficult to sort out which microbes that enable it to store nitrogen, a key nu- with understanding what parts of complex
are doing what in a lake. But Lauren trient, which may provide another com- microbial communities are changing and
Krausfeldt, a microbiologist at Nova petitive advantage over the many lake what we can change to produce a different
Southeastern University, recently turned microbes that did not. outcome,” Peralta says. But she’s optimistic
to metagenomics, a strategy of sequencing Krausfeldt suspects the phages lie dor- that in time, “we can figure out what levers
all the DNA in samples of water and other mant until some unknown environmental we can move.” j

N E WS | I N D E P T H
EMPLOYMENT
Mass layoff looms for Japanese researchers

Thousands could see their jobs axed in the wake of labor law adopted a decade ago
By Dennis Normile fixed-term employees. Today RIKEN has in 2016 retroactively to those who have
programs in brain science, quantum com- already worked under contract for 10 or
T
housands of researchers at Japanese puting, and preventive medicine scattered more years, is “illegal,” says Yasuyuki Ka-
institutes and universities may see among 10 branches and campuses, and it nai, the executive committee chairman of
their jobs disappear by next spring, runs a powerful synchrotron and a peta- RIKEN’s labor union. He says the research-
an unintended result of labor legisla- scale supercomputer. But 77% of its 2893 ers have a right to continued employment.
tion adopted a decade ago that gave current researchers are fixed-term workers. Unhappy with the way RIKEN negotiated,
researchers who have worked under Legislation adopted in 2013 and amended the union on 20 June formally asked a gov-
fixed-term contracts for more than 10 years in 2014 gave most contract employees the ernmental labor relations board to order
the right to permanent employment. Japan’s right to request permanent employment the agency to bargain in good faith. With
science system has many such temporary after working for the same employer for union support, “researchers are now pre-
workers—but rather than fully hire them, in- 5 years; for researchers, the term was set to paring to take the matter to court,” Kanai
stitutions are terminating their jobs. 10 years. Many employers have responded says. The union notes that fresh cohorts
Scientists are trying to head off the lay- by making sure contract workers never ac- of several dozen fixed-term researchers
offs; the union for RIKEN, Japan’s net- cumulate that duration of service. annually will reach the term limit in the
work of nationally supported years ahead.
laboratories, filed a protest with Other institutions face similar
a Tokyo labor board last month problems, although few have as
and may take legal action. Re- many temporary contracts as
gardless of the outcome, the dis- RIKEN. Some are trying to find
pute could create more upheaval ways to retain their workers. The
in a research system whose National Institute of Advanced
global impact is already waning Industrial Science and Technol-
(Science, 27 May, p. 903.) “We ogy has converted to permanent
are on the verge of seeing a pos- status all 245 fixed-term con-
sible mass dismissal of research- tractors who applied for it, ac-
ers this year,” Tomoko Tamura, cording to press reports. Tohoku
a member of the legislature’s University is reportedly screen-
upper house, said during a May ing 275 fixed-term researchers
parliamentary question time on for possible permanent employ-
the issue. Tamura’s analysis of ment. At the University of To-
government data suggests up kyo, which has 588 fixed-term
to 4500 researchers are at risk, employees approaching 10 years
which “could have a serious long- SPring 8, one of the world’s most powerful synchrotrons, is among of service, some might be moved
term impact on Japan’s research the cutting-edge facilities RIKEN brought online during 3 decades of growth. to new projects, a spokesperson
and development,” she said. says, without providing details.
Japan’s R&D funding grew rapidly in RIKEN took that step in 2016, specifying The broader problem is a scarcity of op-
the 1990s and 2000s, but many newly that the count of years served starts in 2013. portunities for researchers to change jobs
recruited researchers were hired under That means contract researchers who have in Japan, says Eisuke Enoki, who heads
fixed-term contracts, which offer lower already worked for RIKEN for more than an Osaka-based organization that studies
pay, fewer benefits, and less job security 10 years may face termination next year. In science policy. “The originally envisioned
than permanent jobs. The scheme gave an email to Science, RIKEN says 203 fixed- ideal was for academics to become assis-
research institutions more flexibility—but term researchers will reach the end of their tant professors after one or two postdocs
in practice, most fixed-term contracts were final contracts before the end of March and to gain a permanent position if tenure
renewed indefinitely. 2023. The institute is currently screening is approved,” he says. But a tenure system
RIKEN is a prime example. Thirty years them and expects to make an unspecified has never taken hold, and there are few
ago, it had about 400 researchers, most of number permanent employees, but many permanent positions, even for team lead-
them permanent employees working on will have to leave. Among the vulner- ers with good track records, Enoki says.
basic physics and chemistry at the main able scientists are 42 team leaders whose A senior scientist at RIKEN who asked
campus near Tokyo. In the mid-1990s, Ja- groups will be disbanded if they go, which not to be identified agrees. His final con-
pan set out to roughly double government puts another 177 positions at risk. RIKEN tract is ending and it’s “very difficult to
spending on research within 5 years, but says it hopes those who are forced out “will find a new position,” he says: “If I get a job
the National Personnel Authority resisted be able to continue their research activi- in China, Korea, or Taiwan, I will move.”
PHOTO: RIKEN
increasing the number of employees on ties at universities, research institutes, and The crisis underscores that for young peo-
the national government payroll. Instead, private companies in Japan and overseas.” ple, in Japan “being a researcher is not an
RIKEN used project funds to hire many Applying an employment policy adopted attractive profession,” he adds. j

NEWS
FEATURES
PHOTO: CREDIT GOES HERE AS SHOWN; CREDIT GOES HERE AS SHOWN

A FUNGAL SAFARI
A new nonprofit has launched an ambitious
effort to raise the profile of often invisible soil fungi
By Gabriel Popkin, in Chile’s Villarrica National Park

Photography by Mateo Barrenengoa L.
A
s a motley medley of mycologists of the Chile expedition. “We have to figure
climbed the basalt slopes of the out where they are and what they’re doing.”
Lanín volcano earlier this year, the SPUN’s approach is bold—even bombas-
green foliage at lower elevations tic. The project launched last fall with a
gave way to autumnal golds and media-savvy campaign, including slick
reds. Chile’s famed Araucaria— videos, arguing that society must do more
commonly called monkey puzzle to study and protect fungi to safeguard
trees—soon appeared, their spiny biodiversity and curb climate change.
branches curving jauntily upward SPUN researchers describe themselves as
like so many cats’ tails. “myconauts” heading into the unknown.
Beneath the majestic trees, the scien- They wear customized blue jumpsuits
tists were focused on something far less emblazoned with “PROTECT THE UN-
glamorous—indeed, mostly invisible: mycor- DERGROUND” for publicity photos and
rhizal fungi, tiny organisms that intertwine when working in the field. (“I really love
with roots of the Araucaria and jumpsuits,” Kiers says.) A docu-
nearly all the other plants in this mentary video crew followed
forest. The multinational re- “Up to 50% SPUN scientists on their first
search team had come to collect
soil samples they hoped would,
of the living expedition, into Chile. Celebri-
ties such as primatologist Jane
with help from DNA testing, biomass of Goodall and best-selling author
reveal exactly which fungi live Michael Pollan have signed on
here, and how they support this soils are these as SPUN advisers.
complex assemblage of flora.
By the end of an exhausting
networks” Some researchers, however,
harbor doubts that SPUN’s
day that included bushwhack- Toby Kiers, mapping effort will have much
ing through heavy brush, the Free University of practical impact, noting that
fungi hunters had filled seven Amsterdam conservationists are already
small plastic sacks with dirt protecting forests and other
from different locations. “I wouldn’t be ecosystems that harbor fungi and store
surprised if there are 100 undescribed spe- planet-warming carbon. Others question
cies” of fungi in each bag, said mycologist whether the surveys will appreciably add
Giuliana Furci, founder of the Chilean non- to what scientists already know, in part be-
profit Fungi Foundation and one of the ex- cause SPUN is only studying one segment
pedition leaders. of the fungal community: those that form
The April ascent was also a road test of associations with plant roots.
sorts: the first of many surveys that the Mapping “a single class of microorganism
Society for the Protection of Underground seems to me too limited to come to an under-
Networks (SPUN), a new fungus-focused standing of the big picture,” says Heribert
nonprofit, hopes to conduct. It has raised Hirt, a plant scientist at the King Abdullah
some $3.5 million for an ambitious effort to University of Science and Technology. “I am
map the global distribution of mycorrhizal rather skeptical that we will really learn a
fungi, which can create subterranean net- lot from this big science project.”
works that are thought to play a key, but of- But SPUN’s effort to make soil fungi more
ten overlooked, role in shaping ecosystems. visible is being welcomed by most mycolo-
Mycologist Toby Kiers marks “Up to 50% of the living biomass of soils gists, who often feel as overlooked as the or-
a fungi sampling plot in are these networks,” says ecologist Toby ganisms they study. “I don’t think anything
Chile’s Valdivia National Reserve. Kiers of the Free University of Amsterdam, like this has ever happened before,” says
a co-founder of SPUN and one of the leaders Kabir Peay, a mycologist at Stanford Univer-

NEWS | F E AT U R E S
sity who advises SPUN. “It’s amazing there

are philanthropists that have the vision and
interest to support this type of activity.”
APPRECIATED OR NOT, fungi are integral to

Earth’s ecosystems. They evolved hundreds
of millions of years before land plants and
animals. By breaking down rock and free-
ing up nutrients, they helped plants colo-
nize land some 500 million years ago. To
this day, most land plants access water
and nutrients in part by partnering with
mycorrhizal fungi that grow on—and often
into—their roots. (Roughly speaking, “my-
corrhiza” means “fungal root.”) Some plants
get up to 80% of their phosphorus—a vital
nutrient—from fungi. And some fungi con-
struct intricate underground webs known
as mycelium that can stretch for kilo-
meters. “Wherever there are roots,” Kiers
says, “there are fungi.”
Despite their ubiquity and importance,
however, fungi challenge biological para-
digms and have defied easy description.
The mushrooms that many people think of
as classic fungi, for example, are just the
spore-forming appendages of larger organ-
isms that are typically hidden from view.
Early biologists lumped fungi with plants,
even though they don’t photosynthesize.
Only in 1969 did scientists recognize fungi
as a separate kingdom of life. (Furci pre-
fers “queendom.”) It’s also hard to define
an individual fungus: “One” mycelium can
hold many cell nuclei that don’t always
share the same DNA. So is the mycelium
an individual, or is each nucleus?
Fungi are often relegated to second-class
status within global scientific and conser-
vation agendas. Whereas visually striking
and charismatic species such as tigers,
whales, and orchids have grabbed atten-
tion, fewer than 600 fungal species have
had their conservation status assessed.
“Fungi are seen as the subservient group”
to plants, says Greg Mueller, chief scientist
at the Chicago Botanic Garden. But some
researchers have responded by flipping the
script: Maybe, Mueller says, a bit tongue-
in-cheek, “plants just exist to feed fungi.”
Kiers, for one, has adopted a fungi-first
worldview. She grew up in the United
States and fell for fungi during a stay at the
Smithsonian Institution’s tropical research
station in Panama. In her lab in Amster-
dam, she uses tools such as microscopy and
fluorescence to reveal how nutrients flow
PHOTOS: MATEO BARRENENGOA L.
through mycelial networks. In one highly

cited paper, Kiers’ team showed one type
of mycorrhizal fungus could reward indi-
vidual plants that provided it with plenti-
ful sugars by directing other nutrients to
those plants’ roots, while “punishing” stin- Three kinds of ectomycorrhizal fungi found in Chile hint at the immense diversity of these organisms,
gier plants by withholding nutrients. Such which associate with trees. From top to bottom: Ramaria flava, Cortinarius lebre, and Laccaria tetraspora.

results, Kiers says, demonstrate that fungi A vital partnership
can wield real power and agency. Most land plants team with mycorrhizal
But such experiments drastically simplify fungi to survive. This ancient partnership has
the unruly networks that shuttle water and helped both plants and fungi thrive over much
nutrients through natural ecosystems. That of Earth. Fungi receive carbon that plants
disconnect frustrated Kiers. “You’re think- fix through photosynthesis, while plants
ing: ‘My God, how different is that from gain access to nutrients and water.
what’s happening in the real world?’” Scientists know much more about
That question was in the air when, in the aboveground world than the
September 2020, Kiers met American subterranean ecosystem.
eco-logist Colin Averill over Zoom. Averill
works in the Zürich-based lab of ecologist Arbuscular
mycorrhizal fungi
Thomas Crowther, which specializes in These tiny organisms (blue)
mapping the global distributions of organ- penetrate the walls of plant
isms such as trees, nematodes, and mycor- root cell walls to supply
rhizal fungi, using computer algorithms to nutrients and receive
carbohydrates in return.
“fill in” areas lacking field data (Science,
25 October 2019, p. 412). Kiers was keen
to connect insights from her idealized lab
experiments to the global scale at which N
Averill works. The two eventually pitched P
the Boston-based Grantham Foundation
on a global effort to discover and map
mycorrhizal fungi, and harness them as Spo
Spores
res C
a climate solution (see sidebar, p. 147). In Maple tree
November 2021, the foundation gave the
Roott
R
researchers $3.5 million to launch SPUN. hair
Pine tree
THE GROUP’S TASK is daunting. Vast realms
of the underground world have hardly been
sampled, and scientists estimate that fewer
than 10% of fungal species have been for-
mally described—indeed, it could be as little Ectomycorrhizal fungi Give and take
These larger, mushroom- Fruiting body Mycorrhizal fungi
as 1%, according to a recent paper in the
forming fungi (purple) can N P (Sporome) give plants up
journal Fungal Diversity. build extensive networks to 80% of the
In part, that’s because studying fungi is through the soil. They Mantle phosphorus they
really hard. They live mostly underground, provide plant roots with need. In return,
and many are microscopic. Their most water and nutrients, C the fungi get up
especially nitrogen, and to 20% of the
visible and familiar manifestation—the can help plants survive carbon that plants
mushroom—is only made by certain types stressful conditions. fix into the soil.
of fungi. By contrast, arbuscular mycorrhi-
zal fungi, which associate with most of the
world’s plant species, nestle within the cell advises SPUN. He subsequently launched no systematic baseline” data on the global
walls of plant roots. They are so hard to find a Global Soil Mycobiome consortium and distribution and diversity of soil fungi. “We
and identify that scientists have described started to email colleagues, begging them want to build that baseline.”
fewer than 300 kinds. “We can’t even talk to dig up fungi, dry them, and send them To guide the sampling, Kiers and Averill
about diversity in a way that makes sense to him. The result was a paper published have harnessed machine learning al-
yet,” Kiers says. late last year that analyzed samples from gorithms developed by Johan van den
Some researchers have been chipping 3200 sites holding more than 700,000 “op- Hoogen, a researcher in the Crowther lab.
away at the problem. In 2014, a team led by erational taxonomic units”—DNA sequences The software uses some 10,000 existing
mycologist Leho Tedersoo of the University that could represent fungal species. fungal records and a bevy of environmen-
of Tarfu in Estonia reported in Science on Even with that accomplishment, Peay tal data sets to tease out subtle correlations
an analysis of soil samples from 365 sites says, “If you took the total volume of soil between where fungi live and variables such
on every continent except Antarctica—at sampled by all fungal ecologists, it’s still as aboveground vegetation, temperature,
the time, an unprecedented sampling effort teeny.” SPUN aims to paint a fuller picture and rainfall. Those correlations can identify
that one news outlet described as “stagger- by quickly doubling the number of fungal places with conditions favorable to fungi
ing.” The research revealed, among other samples collected from documented loca- that have not been surveyed using modern
things, that fungal diversity does not al- tions. The group’s leaders will collect some DNA analysis.
GRAPHIC: N. DESAI/SCIENCE
ways mirror that of plants. In other words, of the material themselves, but they also The potential hot spots identified by
protecting just the richest aboveground plan to fund and train a far-flung network SPUN include places such as the high
ecosystems might fail to safeguard the full of myconauts to sample their own regions. steppes of Mongolia and the lowlands of the
diversity of belowground life. The SPUN data, which will be combined Congo River Basin. The modeling also high-
That survey effort “was just the very tip with Tedersoo’s in an open repository, will lighted another candidate: the old-growth
of the iceberg,” says Tedersoo, who now help fill a key gap, Averill says: “There’s forests of Chile.

Mycolgist Giuliana Furci, here with a long-lived Alerce tree in Chile, helped make the South American nation one of the first in the world to legally protect fungi.
SPUN LEADERS CHOSE the South American exotic species imported from Australia— with a rubber mallet, then extracted it.
nation for their inaugural expedition for that hid their destination. Every step re- Once the dirt core was safely deposited
two reasons. One is that Chile holds ancient quired negotiating thickets of bamboo and into a plastic ziplock bag, he and Kiers
and globally unique forests. The Villarrica thorny blackberries that caught on hair trekked further to collect eight additional
foothills, for example, are dominated by and clothes. “She emerges from the deep,” samples, forming a three-by-three grid
a mix of Araucaria and southern beech Furci joked as Kiers used a GPS to navigate 30 meters on a side.
trees, whose roots could host unique as- through a particularly treacherous stretch. Once analyzed, these samples could con-
semblages of mycorrhizal fungi. “F*** the computer!” Kiers retorted. firm or refute a hypothesis based on the
The other is that Chile is home to Furci, To gather their treasure, the research- computer modeling: that the eucalyptus
who has extensively surveyed the country’s ers donned blue plastic gloves to avoid grove is a “cold spot” with relatively few
mushrooms. She is also the founder of the contaminating the soil and swung into fungal species. But the researchers were
Fungi Foundation, one of the world’s first action. Merlin pounded a roughly quarter- eager to collect the soil anyway, because
charities devoted to protecting the under- meter-long metal cylinder into the ground they wanted to see whether the roots of the
ground kingdom. “A sense of in- tall, almost comically skinny
justice” spurred her to start the eucalyptus trees hosted for-
organization in 2012, Furci says. eign fungi that had hitchhiked
“At the time, there was nobody in with the Australian trees.
advocating for fungi in Chile.” Understanding how human-
For the trip, Kiers assem- altered ecosystems like this
bled a team that included one function is as important
Furci, independent mycologist as understanding more natural
Merlin Sheldrake, and his forests, Kiers says.
brother Cosmo Sheldrake, The expedition featured
a U.K. musician and sound some unconventional epi-
recorder. (Furci and Merlin sodes. Before taking samples
are both on SPUN’s advisory on the volcano, for example,
board.) The morning after the the researchers asked Cosmo
researchers trekked across the Sheldrake to play a song, in-
volcano, they clambered into spired by an Indigenous elder
PHOTOS: MATEO BARRENENGOA L.
a van and bumped up a dirt they had met the day before
road toward a constellation of who advised offering music to
sampling spots identified by fungi. The group fell quiet as
SPUN’s modeling. Pulling over he produced a penny whistle
near a gap in a barbed wire and played a haunting, ethe-
fence, the team plunged into a Mycologists Toby Kiers and Merlin Sheldrake real melody followed by Cool-
grove of eucalyptus trees—an sample along the rocky coast of Chile earlier this year. ey’s Reel, a popular Celtic tune.

N E WS | F E AT U R E S
At another stop, Furci charmed a couple

into allowing sampling in their front yard.
By the time Kiers and Merlin were done, By aiding trees, fungi might help curb warming
Furci was handing out licorice and hugs
C
were exchanged all around. She even of- ould adding certain kinds of fungi to soil help curb climate change by enabling
fered the couple a handful of dirt to smell. trees to grow faster and suck more carbon dioxide from the atmosphere? That’s a
“I love it,” the man said rapturously. question that researchers with the new Society for the Protection of Underground
The next morning, the team headed Networks (SPUN) are trying to answer.
west toward the Pacific coast to sample In the spring of 2021, in an abandoned pasture in southwestern Wales, a
in Alerce Costero National Park, home to forestry company planted 25,000 trees over 11 hectares for an experiment designed
another old-growth forest dominated by a by mycologist Colin Averill, a co-founder of SPUN (see main story, p. 142). The planta-
rare tree species, the alerce. In all, the re- tions include sitka spruce, a common timber tree in the United Kingdom, and a mix of
searchers collected some 30 soil samples in native deciduous trees. To half of the seedlings’ roots, researchers added mycorrhizal
just over a week of fieldwork. They handed fungi—microorganisms that help provide plants with nutrients by associating their
them over to César Marín, a mycologist at roots—that were sourced from mature forests of the same type. Now, the researchers
Santo Tomás University in Chile, for DNA are waiting to see whether the treated trees grow faster and absorb more carbon than
analysis. The results—including some from those that didn’t get the treatment. A similar trial is ongoing in Yucatán, and a third will
additional samples collected by Marín— be planted this fall in Ireland.
will be fed back into SPUN’s modeling in The effort builds on research suggesting the right fungi can give arboreal growth
order to improve its predictions. a powerful boost. In a January paper published in The ISME Journal, Averill and col-
leagues reported growth rates of trees in forests across Europe vary by up to a factor
SPUN’S EFFORTS to map fungi in understud- of three depending on their fungal partners.
ied places reflect a “pragmatic” approach But just how much fungi can supercharge saplings is an open question. Companies
that could help scientists better under- have long sold mycorrhizae that tree planters can dab onto roots. But such commer-
stand how complex ecosystems like for- cial mixes are not adapted to specific tree species or locations, Averill says, and there
ests work, says mycologist Justine Karst of is little evidence they help. For success, he believes “you need to get the right organ-
the University of Alberta, Edmonton. She isms in the right place.”
also lauds the team’s efforts to excite the The Wales experiment is testing that hypothesis. Averill visited the site in April to
public—something she says most scientists measure the trees and found the inoculated saplings are growing faster than ones
don’t prioritize. “I’m looking forward to grown without added fungi. He’s awaiting a second year of data before publishing
seeing what they produce.” them, “but the effect sizes are large enough that I’d be surprised if they go away.”
SPUN, meanwhile, hopes its analyses Even if bespoke fungi enhance tree growth in field studies, however, it’s not clear
will also help bolster emerging efforts to whether they can be deployed cheaply and conveniently enough to broadly appeal to
protect fungi—and Furci’s advocacy work forest owners. Averill will investigate that this fall through a company he’s created, Funga,
in Chile could provide a template. Largely that will inoculate pines being commercially planted in the southeastern United States.
because of her lobbying, Chilean lawmak- The idea of tailoring fungi to boost tree growth “is something many of us write
ers a decade ago passed the world’s first about in our grants,” says Kabir Peay, a mycologist at Stanford University who advises
law formally protecting fungi nationwide. SPUN. “It’s nice to see that someone’s actually out there trying. —G.P.
Elsewhere, fungi are also beginning to
capture official attention. Several Euro-
pean countries have taken steps to protect protecting fungi. And although SPUN “is mapping and demonstrating patterns of
endangered fungal species, and Estonia not the only game in town” when it comes biodiversity, and demonstrating a connec-
has created a small preserve where several to lobbying for fungi, Mueller believes its tion to conservation,” Peay says.
red-listed fungi live. (In the United States, well-funded public outreach could aid such Still, SPUN’s leaders believe they can
by contrast, fungi have relatively low sta- efforts. “Having this bright light shining on add a new, subterranean dimension to
tus; just two fungi—both lichens—are pro- the issue,” he says, “will move a lot of ini- global conservation efforts. Kiers points
tected under the Endangered Species Act.) tiatives forward.” to recent research suggesting that, over
Last fall, the International Union for Some influential groups, however, have nearly one-third of Earth’s land area,
the Conservation of Nature (IUCN), which yet to afford fungi as much consideration aboveground biological diversity doesn’t
maintains the global “red list” of threat- as flora and fauna. The Nature Conser- match what’s found beneath the surface.
ened and endangered species, explicitly vancy—one of the world’s largest conserva- That means habitats often seen as rela-
called for fungi to be given the same con- tion groups—“does not specifically target tively species-poor—such as boreal forests
sideration as plants and animals, as did the fungi” when prioritizing ecosystems for and drylands—could harbor far greater un-
conservation organization Re:wild. And in protection, says David Banks, the group’s derground diversity than is currently rec-
December 2021, the United Nations’s Food chief conservation officer. “But because ognized. Better maps of fungal diversity,
and Agriculture Organization launched an we’re working in these bigger [ecosys- she adds, could aid forestry, farming, and
International Network on Soil Biodiversity tems], we can capture them.” (The conser- efforts to curb climate change.
that includes research on fungi. vancy is nevertheless funding a SPUN-led Kiers recognizes, however, that the onus
Mueller, who chairs IUCN’s fungal con- trip to sample mycorrhizal networks on a is now on SPUN to show its value before its
servation committee, and other advocates remote Pacific island.) first flush of funding runs out. “There has
are now working to have the Convention Even SPUN supporters acknowledge to be a demonstrable benefit to building
on Biological Diversity—a multinational that a more complete picture of fungi an organization dedicated to protecting
pact that helps set the global conserva- won’t automatically lead to better real- underground ecosystems,” she says. “The
tion agenda—explicitly embrace the goal of world outcomes. “There is a gap between clock is ticking.” j

N E WS | F E AT U R E S
At another stop, Furci charmed a couple

into allowing sampling in their front yard.
By the time Kiers and Merlin were done, By aiding trees, fungi might help curb warming
Furci was handing out licorice and hugs
C
were exchanged all around. She even of- ould adding certain kinds of fungi to soil help curb climate change by enabling
fered the couple a handful of dirt to smell. trees to grow faster and suck more carbon dioxide from the atmosphere? That’s a
“I love it,” the man said rapturously. question that researchers with the new Society for the Protection of Underground
The next morning, the team headed Networks (SPUN) are trying to answer.
west toward the Pacific coast to sample In the spring of 2021, in an abandoned pasture in southwestern Wales, a
in Alerce Costero National Park, home to forestry company planted 25,000 trees over 11 hectares for an experiment designed
another old-growth forest dominated by a by mycologist Colin Averill, a co-founder of SPUN (see main story, p. 142). The planta-
rare tree species, the alerce. In all, the re- tions include sitka spruce, a common timber tree in the United Kingdom, and a mix of
searchers collected some 30 soil samples in native deciduous trees. To half of the seedlings’ roots, researchers added mycorrhizal
just over a week of fieldwork. They handed fungi—microorganisms that help provide plants with nutrients by associating their
them over to César Marín, a mycologist at roots—that were sourced from mature forests of the same type. Now, the researchers
Santo Tomás University in Chile, for DNA are waiting to see whether the treated trees grow faster and absorb more carbon than
analysis. The results—including some from those that didn’t get the treatment. A similar trial is ongoing in Yucatán, and a third will
additional samples collected by Marín— be planted this fall in Ireland.
will be fed back into SPUN’s modeling in The effort builds on research suggesting the right fungi can give arboreal growth
order to improve its predictions. a powerful boost. In a January paper published in The ISME Journal, Averill and col-
leagues reported growth rates of trees in forests across Europe vary by up to a factor
SPUN’S EFFORTS to map fungi in understud- of three depending on their fungal partners.
ied places reflect a “pragmatic” approach But just how much fungi can supercharge saplings is an open question. Companies
that could help scientists better under- have long sold mycorrhizae that tree planters can dab onto roots. But such commer-
stand how complex ecosystems like for- cial mixes are not adapted to specific tree species or locations, Averill says, and there
ests work, says mycologist Justine Karst of is little evidence they help. For success, he believes “you need to get the right organ-
the University of Alberta, Edmonton. She isms in the right place.”
also lauds the team’s efforts to excite the The Wales experiment is testing that hypothesis. Averill visited the site in April to
public—something she says most scientists measure the trees and found the inoculated saplings are growing faster than ones
don’t prioritize. “I’m looking forward to grown without added fungi. He’s awaiting a second year of data before publishing
seeing what they produce.” them, “but the effect sizes are large enough that I’d be surprised if they go away.”
SPUN, meanwhile, hopes its analyses Even if bespoke fungi enhance tree growth in field studies, however, it’s not clear
will also help bolster emerging efforts to whether they can be deployed cheaply and conveniently enough to broadly appeal to
protect fungi—and Furci’s advocacy work forest owners. Averill will investigate that this fall through a company he’s created, Funga,
in Chile could provide a template. Largely that will inoculate pines being commercially planted in the southeastern United States.
because of her lobbying, Chilean lawmak- The idea of tailoring fungi to boost tree growth “is something many of us write
ers a decade ago passed the world’s first about in our grants,” says Kabir Peay, a mycologist at Stanford University who advises
law formally protecting fungi nationwide. SPUN. “It’s nice to see that someone’s actually out there trying. —G.P.
Elsewhere, fungi are also beginning to
capture official attention. Several Euro-
pean countries have taken steps to protect protecting fungi. And although SPUN “is mapping and demonstrating patterns of
endangered fungal species, and Estonia not the only game in town” when it comes biodiversity, and demonstrating a connec-
has created a small preserve where several to lobbying for fungi, Mueller believes its tion to conservation,” Peay says.
red-listed fungi live. (In the United States, well-funded public outreach could aid such Still, SPUN’s leaders believe they can
by contrast, fungi have relatively low sta- efforts. “Having this bright light shining on add a new, subterranean dimension to
tus; just two fungi—both lichens—are pro- the issue,” he says, “will move a lot of ini- global conservation efforts. Kiers points
tected under the Endangered Species Act.) tiatives forward.” to recent research suggesting that, over
Last fall, the International Union for Some influential groups, however, have nearly one-third of Earth’s land area,
the Conservation of Nature (IUCN), which yet to afford fungi as much consideration aboveground biological diversity doesn’t
maintains the global “red list” of threat- as flora and fauna. The Nature Conser- match what’s found beneath the surface.
ened and endangered species, explicitly vancy—one of the world’s largest conserva- That means habitats often seen as rela-
called for fungi to be given the same con- tion groups—“does not specifically target tively species-poor—such as boreal forests
sideration as plants and animals, as did the fungi” when prioritizing ecosystems for and drylands—could harbor far greater un-
conservation organization Re:wild. And in protection, says David Banks, the group’s derground diversity than is currently rec-
December 2021, the United Nations’s Food chief conservation officer. “But because ognized. Better maps of fungal diversity,
and Agriculture Organization launched an we’re working in these bigger [ecosys- she adds, could aid forestry, farming, and
International Network on Soil Biodiversity tems], we can capture them.” (The conser- efforts to curb climate change.
that includes research on fungi. vancy is nevertheless funding a SPUN-led Kiers recognizes, however, that the onus
Mueller, who chairs IUCN’s fungal con- trip to sample mycorrhizal networks on a is now on SPUN to show its value before its
servation committee, and other advocates remote Pacific island.) first flush of funding runs out. “There has
are now working to have the Convention Even SPUN supporters acknowledge to be a demonstrable benefit to building
on Biological Diversity—a multinational that a more complete picture of fungi an organization dedicated to protecting
pact that helps set the global conserva- won’t automatically lead to better real- underground ecosystems,” she says. “The
tion agenda—explicitly embrace the goal of world outcomes. “There is a gap between clock is ticking.” j

INSIGHTS
A model of a full-sized
human heart was created
using specific
arrangements of gelatin
microfibers.
PERSPECTIVES
PER SPECTIVES
BIOENGINEERING
Hearts by design
Scalable biofabrication of heart helical
tissue pattern augments pumping function
By Michael V. Sefton1 and Craig A. Simmons2 100 mmHg. Cardiomyocytes are organized Regenerative medicine approaches to fix
as helical fibers that enwrap the ventricles. damaged hearts are promising (1), but design
T
he pumping action of the heart comes Orchestrated, electrically paced contraction principles are needed to recapitulate native
from cardiomyocytes, the micrometer- of the helical fibers causes constriction of the
sized muscle cells of the heart, which heart chamber to eject blood, but also twist- 1
Medicine by Design, Institute of Biomedical Engineering,
generate centimeter-sized mechani- ing (torsion). From a mechanical perspective, University of Toronto, Toronto, ON, Canada. 2Ted Rogers
Centre for Heart Research, Institute of Biomedical
cal contraction and pumping of 70 ml heart failure is a consequence of inadequate Engineering, University of Toronto, Toronto, ON, Canada.
from a ventricle against a pressure of pumping or the heart being too stiff to fill. Email: michael.sefton@utoronto.ca

Building a heart fiber by fiber circumferentially aligned gelatin
Gelatin microfibers created with focused rotary jet spinning (FRJS) are formed circumferentially or helically in heart fibers. Fiber orientations were
ventricle–shaped molds. These are seeded with cardiomyocytes, which align along the spun fibers and produce immature based on an analytical model
beating heart muscle. Testing revealed the benefit of torsion created by the helical ventricle structure on ejection fraction (EF) that predicted greater ejec-
and cardiac output (CO), and this arrangement was used to create a full-sized dual ventricle model with three layers. tion fractions (fraction of blood
ejected from the ventricle with
Microfiber Circumferential Helical Dual ventricle model each beat) with helical fibers (7).
~1 mm ~1 cm ~1 cm ~10 cm Cardiomyocytes from rat neo-
nates or derived from human
induced pluripotent stem cells
seeded on the scaffolds gener-
ally aligned with the spun fibers,
producing oriented, beating, yet
developing heart muscle. When
electrically paced, engineered
ventricles with helical fibers
Adult heart showed higher conduction ve-
CO = 0.7 ml/min CO = 1.5 ml/min CO = 6 liters/min Helical locities along the ventricle long
EF = 1.6% EF = 3.3% EF = 50 to 75% Circumferential axis, more uniform deformation,
greater twist, and increased ven-
heart function and address the global heart of the mold, complex organ-scale 3D scaf- tricle output compared with those with cir-
failure epidemic. On page 180 of this issue, folds with specified micrometer-scale fiber cumferential fibers.
Chang et al. (2) introduce a new manufac- orientations were rapidly formed and cells The authors also fabricated dual-ventricle
turing process, focused rotary jet spinning then seeded on the structures. Because FRJS models with three layers, each with distinct
(FRJS), to create 3-dimensional (3D) human requires molds with convex surfaces, a full helical arrangements that better represent
heart structures with prescribed microscale human heart could not be produced as a sin- the native human heart than do single-layer
polymer fiber alignment. gle structure but required assembly and fu- ellipsoid models (see the figure). Although
CREDITS: (PHOTO, OPPOSITE PAGE) CHANG ET AL. (2); (RIGHT) MICHAEL ROSNACH, JOHN F. ZIMMERMAN, HUIBIN CHANG, MICHAEL M. PETERS; (GRAPHIC) N. DEESAI/SCIENCE
As early as the mid-17th century, the func- sion of four individual chambers (see the first electromechanical coupling, which is nec-
tional importance of torsion to the heart’s photo). Monolithic hearts can be bioprinted essary to avoid arrhythmias, was not dem-
pumping efficiency (analogous to wringing by means of 3D extrusion (6), but this does onstrated, the dual-ventricle models seeded
of a wet cloth) was speculated (3). Further not offer the throughput or spatial resolution with cardiomyocytes contracted and dem-
circumstantial evidence for the importance of FRJS. FRJS also improves on electrospin- onstrated calcium wave propagation in the
of the helical structure comes from failing ning and other fiber-spinning methods that direction of the surface fibers. This experi-
hearts in which structural abnormalities due cannot create 3D structures with the same mentally demonstrates the dependency of
to congenital defects, hypertension, ischemia, geometric complexity or fabrication speed. electrical wave propagation and mechanical
or fibrosis alter left ventricle twist and are Chang et al. capitalized on the distinct contraction on myocardial fiber orientation
associated with hindered pumping capac- capabilities of FRJS to revisit whether heli- that was predicted and incorporated into
ity (4). However, it is difficult to specifically cal muscle architecture improves ventricle computational models of the heart (8).
assess the extent to which the heart’s heli- function. They fabricated ellipsoid-shaped Although the study of Chang et al. con-
cal structure contributes to its function (or ventricles with a single layer of helically or tributes important advances to tissue biofab-
dysfunction) because heart fail- rication and understanding of
ure is a multifaceted disease, cardiac structure-function rela-
and medical treatments manage tionships, these proof-of-concept
symptoms rather than fixing models fall short of native heart
underlying deficits. The limited function. In particular, ejection
ability of the heart to regenerate fractions of the engineered sin-
leads to tissue engineering solu- gle ventricles are much less (at
tions. To that end, understanding best, ~6% of native) than those of
and replicating the heart’s helical healthy human hearts. Efficient
structure-function relationship is heart contraction involves in-
thought to be an important step. terplay between the multiscale
FRJS is based on rotary jet architecture of the heart, the me-
spinning, in which centrifugal chanical properties of the tissue,
force expels polymer solution and the contraction of embed-
through a fine nozzle to form ded cardiomyocytes, all of which
free-floating fibers that solidify were likely limiting factors in
as the polymer solvent evapo- this study. Despite their helical
rates (5). In an innovation re- architecture, the engineered ven-
ported by Chang et al., air jet tricles twisted and shortened less
streams align and focus the ex- than in native hearts. Greater
truded fibers to deposit them torsional contraction may be
on collecting structures (molds) possible with further optimiza-
(see the second photo). By con- Microfibers are precisely deposited onto a rotating tion of the fiber architecture,
trolling the shape and rotation heart ventricle mold by using focused rotary jet spinning. guided by greater understanding

I NS I GHTS | P E R S P E C T I V E S
of native ventricular microanatomy from dif-

fusion tensor magnetic resonance and other
imaging methods (9).
The mechanical behavior of the spun scaf-
folds in this study also did not match that
of native myocardium. The myocardium
consists of muscle fibers sheathed in a criss-
cross network of crimped collagen fibers that
bear little load at low strains (10). As such,
myocardium is soft at low strains and stiff-
ens at high strains as the collagen fibers un-
coil (11). By contrast, FRJS produces straight,
bundled fibers that are constantly engaged
and stiff when loaded because they do not POLICING
replicate the nanoscale configuration of col-
lagen fibers. Stiff scaffolds both resist con-
striction and reduce the contractile forces
generated by cardiomyocytes (12). This ef-
Accessing justice for survivors
fect was exacerbated by using relatively few
immature cardiomyocytes. Emerging strat-
egies to mature cardiomyocytes through
of violence against women
specialized media (13) or culture substrates A police reform experiment in India shows mixed results
(14) would increase their contractility, but a
concomitant softening of the scaffold would
still likely be needed for physiological func- By Graeme Blair1 and Nirvikar Jassal2 norms and behaviors of officers and citizens
tion. Alternative bioprinting methods such by promoting contact between policemen
O
as melt electrowriting can produce soft tis- ne in three women has survived physi- and policewomen, as well as through a role-
sue–like mechanics through precise control cal or sexual violence in her lifetime. In model effect. On the separated end, govern-
of single-fiber microarchitecture (15), but not a wave of reforms designed in part to ments establish women-only police stations.
in complex 3D tissue geometries or with the increase women’s access to justice for This may empower policewomen by reducing
speed of FRJS. This remains a challenge: Can such crimes, governments around the contact with sexist policemen, and encourage
hierarchical organization over spatial scales world instituted gender quotas in po- complainants to report in spaces supposedly
beyond that demonstrated by Chang et al. be lice hiring (see the photo), policewomen-run removed from patriarchal norms. Such “en-
achieved while maintaining the versatility counseling centers, women-only stations, and claves” also imply occupational separation:
and throughput of FRJS? legal mandates that women officers exclu- Policewomen are tasked solely with the com-
To achieve a bioengineered heart, a blood- sively handle cases of gender-based violence. plaints of other women, often based on es-
compatible vasculature, an electrical conduc- On page 191 of this issue, Sukhtankar et al. sentialist assumptions that they are innately
tion system, and means for avoiding immune (1) report results of the first randomized trial suited for such roles. Separated institutions
responses are still needed. Large metaboli- on these reforms. The researchers partnered are an implicit acknowledgment of the limits
cally active, cell-dense organs need lots of with the state police in Madhya Pradesh, In- of integration: if gender norms are unlikely
cells (of different types) and lots of oxygen. dia, in the Hindi heartland with a reputation to change in the short term, the state must
The heart is more than a simple pump. for deep-rooted patriarchy, to randomize the resign itself to the segregation of women of-
Nonetheless, Chang et al. have taken a step introduction of “women’s help desks”—spaces ficers and complainants.
toward making biomechanically equivalent, within police stations where women officers Previous studies have examined the ex-
functional structures. j can interface with women complainants. tremes of the reform spectrum, with more
The results are mixed: More incident reports evidence about separation than integration.
REF ERENC ES AND NOTES
were filed and some police officer attitudes Research on separated institutions finds
1. R. Z. Zhuang, R. Lock, B. Liu, G. Vunjak-Novakovic, Nat.
Biomed. Eng. 6, 327 (2022). toward violence against women changed, but mixed results. Women-only police stations
2. H. Chang et al. Science 377, 180 (2022) women were no more likely to report crimes improved police legitimacy and attitudes to-
3. J. I. E. Hoffman, Physiol. Rep. 5, e13404 (2017). and the arrest rate was unaffected. ward violence against women in Brazil (2).
4. A. M. S. Omar, S. Vallabhajosyula, P. P. Sengupta, Circ.
Cardiovasc. Imaging. 8, e003029 (2015). Gender-based reforms in policing share “Women’s justice centers” in Peru that com-
5. M. R. Badrossamay, H. A. McIlwee, J. A. Goss, K. K. similar goals but differ in design and un- bine police services with legal and medical
Parker, Nano Lett. 10, 2257 (2010). derlying motivation. They can be viewed as aid increased police filings and arrests (3).
6. A. Lee et al., Science 365, 482 (2019).
7. E. A. Sallin, Biophys. J. 9, 954 (1969). falling on a spectrum from integration to However, women-only stations in India led
PHOTO: REBECCA CONWAY/AFP/GETTY IMAGES
8. P. H. M. Bovendeerd et al., J. Biomech. 25, 1129 (1992). separation. On the integrated end, quotas policemen in regular police stations to refer
9. S. Nielles-Vallespin et al., J. Magn. Reson. Imaging. 52, and affirmative action increase the repre- survivors to alternate sites, forcing victims
348 (2020).
10. R. Avazmohammadi et al., Annu. Rev. Biomed. Eng. 21,
sentation of women in the police. Typically, to travel longer distances to access justice
417 (2019). women officers are then assigned the same (4, 5). Women officers staffing these sites
11. G. Sommer et al., Acta Biomater. 24, 172 (2015). tasks and roles as men officers. This may shift may not necessarily be more accommodat-
12. J. G. Jacot, A. D. McCulloch, J. H. Omens, Biophys. J. 95, ing to women complainants either, because
3479 (2008).
13. D. A. M. Feyen et al., Cell Rep. 32, 107925 (2020).
1
Department of Political Science, University of California, policewomen are not immune from exhibit-
14. W. Dhahri et al., Circulation. 145, 1412 (2022). Los Angeles, Los Angeles, CA, USA. 2Department of ing gender bias (6, 7). By contrast, the inte-
15. N. T. Saidy et al., Small. 15, 1900873 (2019). Government, London School of Economics and Political
Science, London, UK. Email: graeme.blair@gmail.com; gration of women officers into mainstream
10.1126/science.add0829 njassal@stanford.edu law enforcement through affirmative action

of native ventricular microanatomy from dif-

fusion tensor magnetic resonance and other
imaging methods (9).
The mechanical behavior of the spun scaf-
folds in this study also did not match that
of native myocardium. The myocardium
consists of muscle fibers sheathed in a criss-
cross network of crimped collagen fibers that
bear little load at low strains (10). As such,
myocardium is soft at low strains and stiff-
ens at high strains as the collagen fibers un-
coil (11). By contrast, FRJS produces straight,
bundled fibers that are constantly engaged
and stiff when loaded because they do not POLICING
replicate the nanoscale configuration of col-
lagen fibers. Stiff scaffolds both resist con-
striction and reduce the contractile forces
generated by cardiomyocytes (12). This ef-
Accessing justice for survivors
fect was exacerbated by using relatively few
immature cardiomyocytes. Emerging strat-
egies to mature cardiomyocytes through
of violence against women
specialized media (13) or culture substrates A police reform experiment in India shows mixed results
(14) would increase their contractility, but a
concomitant softening of the scaffold would
still likely be needed for physiological func- By Graeme Blair1 and Nirvikar Jassal2 norms and behaviors of officers and citizens
tion. Alternative bioprinting methods such by promoting contact between policemen
O
as melt electrowriting can produce soft tis- ne in three women has survived physi- and policewomen, as well as through a role-
sue–like mechanics through precise control cal or sexual violence in her lifetime. In model effect. On the separated end, govern-
of single-fiber microarchitecture (15), but not a wave of reforms designed in part to ments establish women-only police stations.
in complex 3D tissue geometries or with the increase women’s access to justice for This may empower policewomen by reducing
speed of FRJS. This remains a challenge: Can such crimes, governments around the contact with sexist policemen, and encourage
hierarchical organization over spatial scales world instituted gender quotas in po- complainants to report in spaces supposedly
beyond that demonstrated by Chang et al. be lice hiring (see the photo), policewomen-run removed from patriarchal norms. Such “en-
achieved while maintaining the versatility counseling centers, women-only stations, and claves” also imply occupational separation:
and throughput of FRJS? legal mandates that women officers exclu- Policewomen are tasked solely with the com-
To achieve a bioengineered heart, a blood- sively handle cases of gender-based violence. plaints of other women, often based on es-
compatible vasculature, an electrical conduc- On page 191 of this issue, Sukhtankar et al. sentialist assumptions that they are innately
tion system, and means for avoiding immune (1) report results of the first randomized trial suited for such roles. Separated institutions
responses are still needed. Large metaboli- on these reforms. The researchers partnered are an implicit acknowledgment of the limits
cally active, cell-dense organs need lots of with the state police in Madhya Pradesh, In- of integration: if gender norms are unlikely
cells (of different types) and lots of oxygen. dia, in the Hindi heartland with a reputation to change in the short term, the state must
The heart is more than a simple pump. for deep-rooted patriarchy, to randomize the resign itself to the segregation of women of-
Nonetheless, Chang et al. have taken a step introduction of “women’s help desks”—spaces ficers and complainants.
toward making biomechanically equivalent, within police stations where women officers Previous studies have examined the ex-
functional structures. j can interface with women complainants. tremes of the reform spectrum, with more
The results are mixed: More incident reports evidence about separation than integration.
were filed and some police officer attitudes Research on separated institutions finds
1. R. Z. Zhuang, R. Lock, B. Liu, G. Vunjak-Novakovic, Nat.
Biomed. Eng. 6, 327 (2022). toward violence against women changed, but mixed results. Women-only police stations
2. H. Chang et al. Science 377, 180 (2022) women were no more likely to report crimes improved police legitimacy and attitudes to-
3. J. I. E. Hoffman, Physiol. Rep. 5, e13404 (2017). and the arrest rate was unaffected. ward violence against women in Brazil (2).
4. A. M. S. Omar, S. Vallabhajosyula, P. P. Sengupta, Circ.
Cardiovasc. Imaging. 8, e003029 (2015). Gender-based reforms in policing share “Women’s justice centers” in Peru that com-
5. M. R. Badrossamay, H. A. McIlwee, J. A. Goss, K. K. similar goals but differ in design and un- bine police services with legal and medical
Parker, Nano Lett. 10, 2257 (2010). derlying motivation. They can be viewed as aid increased police filings and arrests (3).
6. A. Lee et al., Science 365, 482 (2019).
7. E. A. Sallin, Biophys. J. 9, 954 (1969). falling on a spectrum from integration to However, women-only stations in India led
PHOTO: REBECCA CONWAY/AFP/GETTY IMAGES
8. P. H. M. Bovendeerd et al., J. Biomech. 25, 1129 (1992). separation. On the integrated end, quotas policemen in regular police stations to refer
9. S. Nielles-Vallespin et al., J. Magn. Reson. Imaging. 52, and affirmative action increase the repre- survivors to alternate sites, forcing victims
348 (2020).
10. R. Avazmohammadi et al., Annu. Rev. Biomed. Eng. 21,
sentation of women in the police. Typically, to travel longer distances to access justice
417 (2019). women officers are then assigned the same (4, 5). Women officers staffing these sites
11. G. Sommer et al., Acta Biomater. 24, 172 (2015). tasks and roles as men officers. This may shift may not necessarily be more accommodat-
12. J. G. Jacot, A. D. McCulloch, J. H. Omens, Biophys. J. 95, ing to women complainants either, because
3479 (2008).
13. D. A. M. Feyen et al., Cell Rep. 32, 107925 (2020).
1
Department of Political Science, University of California, policewomen are not immune from exhibit-
14. W. Dhahri et al., Circulation. 145, 1412 (2022). Los Angeles, Los Angeles, CA, USA. 2Department of ing gender bias (6, 7). By contrast, the inte-
15. N. T. Saidy et al., Small. 15, 1900873 (2019). Government, London School of Economics and Political
Science, London, UK. Email: graeme.blair@gmail.com; gration of women officers into mainstream
10.1126/science.add0829 njassal@stanford.edu law enforcement through affirmative action

Women police officers serve at a station house modes of justice, future researchers should type-casting, including among women com-
in Delhi. Several of India’s states and all its investigate whether such measures lead to plainants and police colleagues?
union territories, including Delhi, have installed a better outcomes for survivors or are merely Another important inquiry concerns the
33% gender quota in police hiring to increase window dressing. For instance, does officer- impact of police reforms on actual rates of
women’s comfort with law enforcement. mediated reconciliation of victims with violence against women. Early evidence is
abusers inside station-houses empower mixed (12). Subsequent scholarship should
or quotas not only increased registrations of complainants, or are such measures simply examine how institutional reforms operate
violence against women in the US but also a mechanism by which the police avoid tak- in conjunction with other tools, including
improved police efficacy (8). Measures that ing formal action against abuse? Punitive media and educational campaigns (13), in-
integrate women officers in security agen- justice may discourage reporting because creasing women’s representation in elected
cies by enabling them to carry out the same victims may not necessarily desire the ar- office (14), limiting behaviors such as alco-
tasks as men can diminish stereotypes (9). rest of intimate partners; then again, reha- hol abuse, or resource transfers targeted to
The Sukhtankar et al. experiment, one of bilitative justice may appear toothless and women (15). In addition, researchers should
the largest of its kind, provides important fail to deter violence. explore whether reforms targeted at women
new evidence on the middle of the theoreti- A key challenge in studies of access to can also affect gender-based violence against
cal reform spectrum. Women’s help desks justice is distinguishing whether effects are men and LGBTQ citizens and their access to
are a form of separation—within regular po- driven by an increase in citizens trying to re- justice for such crimes.
lice stations as opposed to fully segregated port or officers recording those complaints. Recent studies have introduced new
sites—that also retain occupational divides Sukhtankar et al. counted citizens entering standards for transparent ethical decision-
between administrators staffing the institu- police stations from video surveillance cam- making, including scrutinizing the human
tions and other officers. The authors found era footage to distinguish between report- rights records of police partners and moni-
that the intervention did not increase the ing and registration. Future research should toring harm “red lines” (11). Sukhtankar et
likelihood that women report violence to go further by linking police registrations to al. push further. For instance, to mitigate
the police, but it did increase the filing of court records, allowing cases to be traced a potentially unintended consequence of
incident reports that do not lead automati- across successive stages of investigation, trial, their scheme in assigning policewomen to
cally to criminal investigations. They found and verdict. Improvements in data availabil- women’s help desks, they worked with law
only weak effects on formal registrations ity in India and China now make this data enforcement to avoid reassigning women
that warrant mandatory investigation. The linkage a possibility (10). This would allow officers from other stations, which might
help desks did not increase arrests of abus- researchers to probe whether policies like have displaced personnel and undermined
ers. Although the intervention increased women’s help desks mostly affect the first access to justice for women in untreated ar-
officers’ knowledge of the law, it did not step of accessing justice (case filings) or also eas. Subsequent scholarship should follow
generally alter gender norms. Policewom- change later outcomes, such as convictions their example in transparently describing
en’s perceptions that women report false or of suspects. Linking multiple stages of the harm reduction practices and consider pre-
exaggerated claims fell slightly, but police- justice system would reveal where and when registering ethical decision-making criteria.
men’s attitudes were unchanged. discrimination creeps into the process. With these measures, researchers should
The null effect on arrests admits several The short-term outcome measurement make the case, as Sukhtankar et al. persua-
interpretations. For domestic violence in by Sukhtankar et al. leaves open the possi- sively do, that the risks of conducting ran-
particular, women may wish abuse to stop bility that police behavior does change but domized interventions must be balanced
but not for the perpetrator to be arrested. only in the long run, and conversely that the against the value of building more evidence
But because information on what police do increases in case filing may not endure. Ef- about which police reforms work—and
in response to a report was not captured, the fects may dissipate for many reasons. For which do not. j
null may also reflect police inaction, either example, when partnerships with academ-
RE FE RE N CES AN D N OT ES
because women request investigations and ics end, police leaders’ prioritization of the
1. S. Sukhtankar et al., Science 377, 191 (2022).
are ignored or because the police act but the reform may wane. Maintaining buy-in from 2. A. Córdova, H. Kras, Comp. Polit. Stud. 53, 775 (2020).
inquiries do not result in arrests. Police also police leaders may be a key driver of success- 3. M. M. Sviatschi, I. Trako, “Gender Violence, Enforcement,
and Human Capital: Evidence from Women’s Justice
may not have completed investigations by ful implementation (11). The duration of im- Centers in Peru,” Policy Research Working Papers, The
the time data were collected, 2 months after plementation (distinct from how long after World Bank, 12 April 2021.
the intervention activities ended. In addi- implementation measurement takes place) 4. N. Jassal, Am. Polit. Sci. Rev. 114, 1035 (2020).
5. S. Amaral, S. Bhalotra, N. Prakash, “Gender, Crime and
tion, law enforcement may have acted, but may also be important. Because the Sukhtan- Punishment: Evidence from Women Police Stations
not by making arrests. Women’s help desks kar et al. intervention lasted only 13 months, in India,” Working paper, ifo Center for Labour and
Demographic Economics, 2018.
may have provided restorative or rehabilita- new studies with longer time horizons might 6. S. Karim et al., Int. Stud. Q. 62, 618 (2018).
tive justice instead of standard police ser- find different effects. 7. C. Santos, Latin Am. Res. Rev. 39, 29 (2004).
vices, e.g., informal couples’ counseling for More research is needed to understand 8. A. Miller, C. Segal, Rev. Econ. Stud. 86, 2220 (2019).
9. G. B. Dahl et al., Q. J. Econ. 136, 987 (2021).
crimes such as dowry-based harassment the implications of gender-based police re- 10. E. Ash et al., “Measuring Gender and Religious Bias in
and marital rape. form on personnel dynamics and gender re- the Indian Judiciary,” Center for Law and Economics
Working Paper Series 3, 2021.
Future work should disaggregate the lations within police forces. Future scholar- 11. G. Blair et al., Science 374, eabd3446 (2021).
types and gradations of crime affected by ship could probe whether officers appreciate 12. E. Perova, S. A. Reynolds, Soc. Sci. Med. 174, 188 (2017).
police reforms. Institutions like help desks being deployed to institutions like women’s 13. D. P. Green et al., Comp. Polit. Stud. 53, 2283 (2020).
14. L. Iyer et al., Am. Econ. J. Appl. Econ. 4, 165 (2012).
may improve registrations of crimes like help desks, or whether they feel that that 15. M. Hidrobo et al., Am. Econ. J. Appl. Econ. 8, 284 (2016).
dowry-based harassment, but police may deployment prevents them from handling
A CK N OW LE D G M E N TS
not take action on other crimes like gang diverse cases. If policewomen are dispropor-
We thank R. Blair, D. Hausman, M. Shah, and A. Wilke for help-
rape or acid attacks, which they may face tionately or exclusively tasked with inves- ful comments.
pressure to downplay in official statistics. tigating other women’s complaints, might
If these institutions emphasize alternative this diminish professionalization and induce 10.1126/science.abp9542

MEMBRANES transport to occur. As solvent concentration

increases and becomes a continuous fluid
Bridging membrane inside the swollen polymer, the mechanis-

tic view of polymer chain–gated diffusion
has been questioned because of the pres-
transport models ence of continuous solvent pathways. The

ability of the polymer to deform, however,
is an important extension embodied in the
Analysis of transport processes in swollen fluid-solid model presented by Hegde et al.
polymer networks bridges two classical models Recent experiments have supported the no-
tion that hydrated polymer segmental dy-
namics may control the selective transport
By Geoffrey M. Geise mechanistic differences associated with of water versus salt (8, 9). The combination
these two models for swollen polymers of polymer and solvent momentum bal-
T
he movement of small molecules and ultimately extend transport model- ances in the fluid-solid model and the con-
(e.g., water, organic solvents, gases) ing through an application of conservation tinuity equation permits the description of
through polymeric membrane mate- of mass and momentum for solvent and solvent transport, to an order of approxima-
rials is critical for a wide range of ap- the polymer membrane. This approach, tion that is reasonable for many membrane
plications including water desalina- coupled with perturbation analysis, satis- applications, in terms of either a solvent
tion and filtration, crude oil refining, fyingly results in a quantitative agreement concentration or pressure gradient.
and some controlled-release devices (1–4). between the two modeling approaches and Hegde et al. also address the different
For decades, researchers have used two resolves apparent mechanistic discrepan- meanings of “pressure” in the two classi-
classical models to describe fluid passage cies between the two classical models. cal models. The pore-flow model considers
through membranes: solution diffusion and One of the core differences between the pressure to be the hydrodynamic pressure
pore flow (1). Although each model is gen- solution-diffusion and pore-flow models is of the solvent phase. The solution-diffusion
erally accepted to describe certain materi- the treatment of concentration and pres- model, by contrast, considers the solvent
als, the mechanism of how solvent passes sure within the membrane. The solution- and polymer mixture as a single phase.
through materials whose structure falls Therefore, pressure in the solution-diffu-
between the two models, such as swollen sion context includes all of the mechani-
polymers, has been debated since the 1960s. “…the solution-diffusion model cal stresses associated with the swollen
On page 186 of this issue, Hegde et al. (5)
present a two-phase fluid-solid model and
and the pore-flow material. Mechanical equilibrium in the
fluid-solid model is consistent with the ap-
perturbation analysis to demonstrate that
both the solution-diffusion model and the
model ultimately describe propriate analogs in the classical models,
and the extension provided by Hegde et al.
pore-flow model ultimately describe the the same driving provides clarity to inform the usage of the
same driving force for solvent passage in term “pressure.”
swollen polymers. force for solvent passage…” The debate over the mechanism of trans-
Transport in dense polymer gas separa- port in swollen polymers has been sug-
tion membranes can be described appro- diffusion model is based on the existence gested to have implications for the design
priately using the solution-diffusion model, of a concentration gradient, whereas the of advanced materials. The realization that
where gas molecules enter the material pore-flow model is based on a pressure both the solution-diffusion and pore-flow
on one side of the membrane and diffuse gradient. The pressure difference–induced approaches are complementary through the
through the polymer before emerging on concentration gradient, suggested by the extension provided by the fluid-solid model
the other side. Alternatively, micro- or ultra- solution-diffusion model, was experimen- suggests interesting new ways to character-
filtration membranes are often porous poly- tally observed decades ago (6), but the ize and understand materials. This advance
mers, where the polymer supports a rigid application of this observation to highly in modeling transport in swollen polymers
tortuous network of pores that works more swollen polymers has been questioned. An supports and expands a more holistic view
like a sieve, and this mechanism of trans- argument can be made that highly swollen of materials design with a more productive
port can be described using the pore-flow polymers (e.g., hydrogels) have continuous focus on uncovering structure–property re-
model. solvent pathways through which transport lationships needed to engineer advanced
An intermediate case arises when the can occur by means of hydraulic pressure– membranes. j
membrane is a swollen polymer. Central driven flow, in the absence of a solvent
to the debate of how to model these inter- concentration gradient. This argument is
1. J. G. Wijmans, R. W. Baker, J. Membr. Sci. 107, 1 (1995).
mediate cases are the different mecha- resolved by Hegde et al. by recognizing 2. G. M. Geise, D. R. Paul, B. D. Freeman, Prog. Polym. Sci.
nisms described by each model for how that the use of either model does not ne- 39, 1 (2014).
molecules pass through the material. The gate the other and that the resulting sol- 3. K. P. Bye, M. Galizia, J. Membr. Sci. 603, 118020 (2020).
4. R. W. Baker, Membrane Technology and Applications
pore-flow model describes solvent passage vent flux relationships are equivalent. (Wiley, ed. 3, 2012).
as driven by a hydraulic pressure, whereas Solution-diffusion transport in gas sepa- 5. V. H. Hegde, M. F. Doherty, T. M. Squires, Science 377,
the solution-diffusion model describes ration membranes is considered to be gated 186 (2022).
6. S. Rosenbaum, O. Cotton, J. Polym. Sci. A1 7, 101 (1969).
solvent passage as driven by a diffusive by polymer chain motions needed for gases 7. B. D. Freeman, Macromolecules 32, 375 (1999).
process. Hegde et al. resolve the apparent to execute diffusional jumps—that is, the 8. K. Chang et al., Macromolecules 51, 9222 (2018).
steps that a molecule takes while navigat- 9. A. G. Korovich, K. Chang, G. M. Geise, L. A. Madsen,
Macromolecules 54, 11187 (2021).
Department of Chemical Engineering, University of Virginia, ing through a material (7). In other words,
Charlottesville, VA, USA. Email: geise@virginia.edu the polymer must move out of the way for 10.1126/science.abn5485

SOFT MATTER
Less is more when forming gels by dilution

Molecular self-assembly yields soft materials arising from the liquid state when diluted
By Matthew J. Webber association. Surfactants have broad applications between the surfactant and BTA-EG4
tions as detergents and interface stabilizers are weakened, allowing BTA-EG4 to resume
S
oft materials that change form or func- (4) and in lipid nanoparticle vaccines. its preferred filamentous assembly. The sol
tion in response to environmental or When molecules prepared from differ- and the gel phases both exist at the same
user-applied stimuli have a wide range ent motifs are mixed, self-assembly typically relative composition of these two orthogonal
of biomedical applications (1). Gels can obeys a thermodynamic preference for self- motifs and require only a change in the total
form in water from weakly interacting sorting, meaning that each species forms its overall concentration for the system to oscil-
molecules but can return to the state own segregated structures. However, realiz- late between its sol and gel states. This sol-
of a flowing liquid suspension, known as a ing pure self-sorting can be a challenge be- to-gel transition proved to be a robust phe-
sol, upon changes in the concentration of the cause of common driving forces shared by nomenon, arising when combining a suite of
molecules or the applied temperature. This the different molecular motifs (5). Interesting different common surfactants with BTA-EG4.
behavior is known as a reentrant phase tran- phenomena emerge from multicomponent The gel phase that forms upon dilution is a
sition. A gel-to-sol phase transition typically co-assembly when underlying driving forces, transient state. Further dilution crosses the
arises from a reduction in concentration, such as hydrophobic association, are con- critical gelation threshold of the BTA-EG4 su-
meaning that a gel becomes a sol upon dilu- served between different motifs. In this case, pramolecular filaments and restores the sol.
tion and a sol becomes a gel with increased minor adjustments in the strength of the in- In demonstrating even more complex gel-
concentration. On page 213 of this issue, Su teractions between molecules can dictate the -sol-gel-sol dilution cascades, Su et al. pre-
et al. (2) demonstrate a system that exhibits emergence of different self-assembled struc- pared two different protocols based on mul-
a sol-to-gel transition when diluted, inverting tures (6). Because these same interactions are ticomponent and orthogonal self-assemblies.
the common behavior of gels. Their observa- governed by parameters such as concentra- The system prepared from BTA-EG4 and sur-
tions offer insight into systems that undergo tion, temperature, and salt or pH levels of factant can begin from the state of a robust
reentrant phase transitions in biology. the solution, the resulting assemblies can be gel at a very high starting concentration at
Self-assembly is a process in which mol- tuned by controlling these parameters to cre- a molar ratio of 2 to 3 for supramolecular
ecules form specific structures either as a ate responsive material platforms. motif to surfactant. At this high initial con-
consequence of their properties or the con- The concentration-dependent interaction centration, spherical assemblies consisting of
ditions to which they are exposed. Several of two self-assembly motifs—here being those both motifs become fused to form a network.
self-assembly motifs have been used to pre- that assemble through disparate mechanisms Upon dilution, a sol emerges that is com-
pare soft materials. One such motif leverages of supramolecular polymerization or surfac- posed of similar spherical assemblies but at
molecules with certain supramolecular inter- tant micelle formation—is key to the phe- a concentration below that necessary to form
actions, such as directional hydrogen bond- nomena reported by Su et al. (see the figure). the fused network. Continued dilution from
ing units, to form supramolecular polymers. The primary supramolecular polymer motif this point restores the gel state through for-
This type of filamentous assembly is reminis- is a disk-like molecule known as BTA-EG4 (7), mation of the supramolecular filaments pre-
cent of high persistence-length polymers (3). which is prone to stack in water into elon- ferred by the BTA-EG4 motif. Yet additional
Another common motif is that of simple sur- gated nanoscale filaments that can entangle dilution reduces entanglement of these fila-
factants, which assemble into spherical ob- to form a gel. However, when this supramo- ments below the threshold for gel formation,
jects known as micelles through hydrophobic lecular motif is combined with a surfactant, restoring the sol once again.
the formation of supramolecular filaments is An alternate protocol enabling a different
inhibited as the surfactant disrupts the stack- gel-sol-gel-sol cascade is also demonstrated
Department of Chemical and Biomolecular Engineering,
University of Notre Dame, Notre Dame, IN, USA. ing. Upon dilution, however, the unexpected by Su et al. by including a third molecule
Email: mwebber@nd.edu sol-to-gel transition is observed as interac- capable of forming another orthogonal su-
A dilution-responsive supramolecular gel-sol-gel-sol cascade Orthogonal self-assembly motifs

GRAPHIC: KELLIE HOLOSKI/SCIENCE BASED ON M. J. WEBBER
Balancing the interactions of two different self-assembly motifs, Su et al. designed a mixture that Supramolecular polymer Surfactant
goes from gel to sol to gel to sol upon continued dilution.
Gel Gel
Dilute Dilute Dilute
Concentrate Concentrate Concentrate
Sol Sol
At high concentration, the two motifs Upon dilution, the network of spherical Upon further dilution, altered interaction Upon even further dilution, the supra-
interact to form spherical assemblies assemblies is disrupted to produce a of the two motifs form supramolecular molecular polymer network passes its
that fuse into a gel network. flowing colloidal suspension (sol). polymers that entangle into a gel network. gel point to restore the sol.

pramolecular network. This component com- EVOLUTION

prises a ureidopyrimidinone (UPy) supramo-
lecular motif that forms long fiber bundles
(8). Although the UPy motif also interacts
with the surfactant, the assembled UPy fiber
“Arch”-etyping vertebrates
bundles are not evidently disturbed by the Cellular details of gill arches in Cambrian fossils reignite
presence of surfactant. By this approach, the
system of BTA-EG4 and surfactant, at concen-
a centuries-old debate
trations that previously formed a sol consist-
ing of spherical assemblies, can instead form By Tetsuto Miyashita page 218 of this issue, Tian et al. (7) present
a gel by adding a high concentration of UPy compelling evidence in yunnanozoans for an
S
fibrillar structures. Dilution from this ini- cientists have long been searching unmistakable vertebrate trait—a pharyngeal
tial point surpasses the gelation capacity of for fossils of distant vertebrate an- skeleton made of cellular cartilage.
the UPy network while not yet reaching the cestors. In the 1990s, mysterious Interpreting organic stains on a shale slab
point of inducing BTA-EG4 filament forma- fishlike forms—now known as yun- is both a science and an art. Wielding scan-
tion, yielding a sol. Continued dilution acti- nanozoans—were discovered at a ning electron microscopy and computed mi-
vates BTA-EG4 filament formation, which in 520-million-year-old Cambrian fossil crotomography scans to yield unprecedented
combination with the remaining UPy fiber site in the Yunnan province of China (1–3). details, Tian et al. reveal cellular and subcel-
bundles restores a gel state comprising two More fishlike forms (e.g., Haikouichthys and lular structures of the skeletal bars that best
orthogonal networks of supramolecular poly- Myllokunmingia) were reported from the compare to cartilaginous gill arches of mod-
mers: the BTA-EG4 and UPy networks. Upon same locality shortly thereafter (4, 5), while ern vertebrates. These bars in yunnanozoans
further dilution, the system transitions back the 508-million-year-old Burgess Shale in the are patterned in a series, each associated with
to a sol once again. Canadian Rockies yielded Metaspriggina (6). gill filaments and connected by horizontal
The tunable nature of molecular-scale self- Having eyes and a brain at the front end of an rods. The morphology closely approximates
assembly in these materials offers simple otherwise wormlike soft body, these animals various predictions for vertebrate ancestors.
synthetic analogues of more complex phe- appear to have branched off the phylogenetic Of all the internal structures of the earli-
nomena observed in nature. For example, tree before the last common ancestor of all est vertebrates, pharyngeal skeletons perhaps
membraneless organelles—distinct compart- living vertebrates. However, there is ongoing stand the best chance for fossilization given
ments within a cell that are not enclosed by controversy about precisely how close to their robustness. Nonetheless, the complex
a traditional lipid membrane—are thought vertebrates these Cambrian forms were. On evolutionary history of the pharynx has
to arise from liquid-liquid phase separation
because of concentration gradients of as-
sociating multicomponent systems forming Of gills and jaws
these assemblies in a water environment (9). Cambrian vertebrates each evolved distinct pharyngeal anatomy with a series of gill-supporting skeletons.
The roles of membraneless organelles in bio- The yunnanozoan has a cartilaginous basket and the Haikouichthys has unjoined bars, whereas the
logical signaling during both normal and dis- Metaspriggina has upper and lower rods. However, it remains an open question whether their gill anatomies
eased states are increasingly appreciated (10). represent any evolutionary link to the jaws of modern vertebrates.
The behavior of the simple systems described
by Su et al. is therefore reminiscent of more Pharyngeal skeleton Gill filaments Myomeres
complex self-assembly phenomena in biol-
Yunnanozoan
ogy, illuminating the importance of subtle
thermodynamic driving forces that give rise C-shaped
to concentration-dependent phase separa- element
tion. This new paradigm in self-assembled
materials consisting of highly adaptive and Pharynx
dilution-triggered hydrogels may further-
more lead to the design of stimuli-responsive
material platforms for in situ modulation of
Haikouichthys
function in therapeutic biomedicine. j
1. M. A. Stuart et al., Nat. Mater. 9, 101 (2010).
2. L. Su et al., Science 377, 213 (2022).
3. T. Aida, E. W. Meijer, S. I. Stupp, Science 335, 813 (2012).
4. M. J. Rosen, J. T. Kunjappu, Surfactants and Interfacial
Phenomena (Wiley, ed. 4, 2012).
5. K. L. Morris et al., Nat. Commun. 4, 1480 (2013).
6. W. M. Jacobs, D. Frenkel, Biophys. J. 112, 683 (2017).
7. C. M. Leenders et al., Chem. Commun. 49, 1963 (2013).
8. S. I. S. Hendrikse et al., Chem. Commun. 53, 2279 (2017). Metaspriggina
GRAPHIC: V. ALTOUNIAN/SCIENCE
9. J. A. Riback et al., Nature 581, 209 (2020).

10. Y. Shin, C. P. Brangwynne, Science 357, eaaf4382 (2017).
ACK N OW LED GMENTS

M.J.W. acknowledges funding from the National Institutes of
Health (R35GM137987) and the National Science Foundation
(BMAT, 1944875).
10.1126/science.abo7656

strained attempts to interpret fossil imprints. Rather, its identity is predicated on the pres- NEUROSCIENCE
Pharyngeal slits and skeletons long precede ence of a specific stream of neural crest cells,
the origins of vertebrates, and cellular car-
tilage has been found in the lips of a devel-
oping invertebrate chordate (8). Vertebrates
a fifth cranial nerve, and specialized mouth
structures for pumping water and feeding.
Without these markers, and with variations
Sounding
develop the entire pharyngeal skeleton from
cellular cartilage in the embryonic pharyn-
geal arches. This cartilage originates in the
observed among different vertebrate lineages,
overall positions of the gills help little to de-
termine arch homology in yunnanozoans.
out pain
vertebrate-specific cell lineage called the neu- Tian et al. offer an emerging scope of A circuit for sound-induced
ral crest. To complicate matters further, the
pharynx is regionally and phylogenetically
diversity in pharyngeal anatomy of early
vertebrates (see the figure). Among other
analgesia has been found in
differentiated among vertebrates. Depending Cambrian fishlike forms, Haikouichthys the mouse brain
on whether similarities or differences are seems to have skeletal rods that support
emphasized, the pharyngeal arch I in which the gills, whereas only the gill pouches By Rohini Kuner1 and Thomas Kuner2
jaws develop, that is, the mandibular arch, are described for Myllokunmingia (4).
T
may (9–11) or may not (12–14) share an evolu- Metaspriggina has skeletal bars segmented he perception of physical pain is sub-
tionary origin with gill-supporting skeletons. into upper and lower halves (6). Hagfish and ject to variation depending on the
New information from yunnanozoans pre- lampreys evolved from the common ances- context and which other sensory in-
sents an opportunity to clarify these issues. tor with a cartilaginous basket around the puts are being received, including
Tian et al. avoided jumping to conclusions, gill pouches and a specialized oral skeleton sound. The emerging field of music
as they did not explicitly identify which gill (15). Similar pharyngeal skeletons also occur therapy (1)—which is applied to con-
bar in a yunnanozoan corresponds to which in successive out-groups of jawed vertebrates trol postoperative, pediatric, postpartum,
arch in a modern vertebrate. But they did (12). And jawed vertebrates have a series of and cancer pain and is being increasingly
signal their favored interpretation that yun- jointed skeletal arches, the first of which dif- tested in chronic pain disorders—capital-
nanozoans, with each gill bar identical to the ferentiate as jaws. These different patterns izes on the interactions between sound and
next, represent the ancestral vertebrate con- are as anatomically disconnected from each pain perception to attenuate pain. Given
dition. This view is in line with the belief that other as they are phylogenetically distant. that music and natural sounds can posi-
all pharyngeal arches originally supported Given such a complex distribution of char- tively affect mood, relieve stress, and relax
gills and that one of them evolved into a jaw acters, it seems premature to assume any sin- the body, it is not unreasonable to think
(9–11). At face value, yunnanozoans could gle form as the ancestral phenotype on a lin- that these factors underlie pain relief. On
serve as long-awaited evidence for the gill- ear path toward modern vertebrates. In the page 198 of this issue, Zhou et al. (2) demon-
arch hypothesis of jaw origins. phylogenetic analysis by Tian et al., yunnano- strate that pain relief by sound is not purely
However, this jaw-origin narrative relies zoans are an out-group to all other vertebrate attributable to stress reduction and distrac-
solely on how one chooses to identify the gill branches. This suggests differential evolution tion. They interrogate neural circuits to un-
bars of yunnanozoans—that is, whether the of pharyngeal patterning among early ver- ravel a specific pathway for sound-induced
first gill bar in yunnanozoans corresponds tebrate lineages. By the time yunnanozoans analgesia in the brains of mice.
to the mandibular arch in jawed vertebrates. were sloshing about in the Cambrian sea, Using rodents to study how music and
Historically, one or more additional arches other primitive fishes had evolved to slurp sound are related to pain presents major
were postulated in front of the mandibu- food differently with their uniquely derived challenges, not least because it is unknown
lar arch for vertebrate ancestors (11). Other pharyngeal anatomy. Although evolutionary how animals perceive music. Zhou et al. car-
views posit no such extra arches and consider biologists have been busy chasing the mythi- ried out behavioral tests addressing pain
the mandibular arch as distinct from the gill cal ancestor that explains everything about sensitivity and found that mice did not show
arches (12–14). Yunnanozoan morphology the vertebrate body plan, perhaps the oppo- differential responses to melodic classical
excludes none of these ideas. Their first gill site is a sensible approach. In other words, music (consonant sounds), dissonant mu-
bar could be the elusive arch that was later the meandering journey toward modern sic, or white noise. Notably, they found that
lost in vertebrates, or, conversely, the man- vertebrates may be best understood by popu- the decisive factor in eliciting pain relief is
dibular region might not have formed a full lating the family tree with divergent and dis- a 5-dB increase in sound intensity in any of
skeletal arch. Consistent with the latter sce- continuous anatomical forms, guided by phy- these three types of sound relative to ambi-
nario, the snout and lips either appear di- logenetic inference rather than by theory. j ent sound levels, whereas 10-, 15-, or 20-dB
minutive or are absent in yunnanozoans and increases were ineffective. In mouse mod-
R EFER ENCES AN D N OT ES
other Cambrian forms (5–7). Specifically, one els, a 5-dB increase in sound intensity led
1. J.-Y. Chen et al., Nature 377, 720 (1995).
C-shaped oral structure, as identified by Tian 2. J.-Y. Chen et al., Nature 402, 518 (1999). to inhibition of both sensory-discriminative
et al., may represent what might be a slim 3. D. Shu et al., Science 299, 1380 (2003). aspects of pain, such as evoked responses
4. D.-G. Shu et al., Nature 402, 42 (1999).
mandibular arch of yunnanozoans. If this is 5. D.-G. Shu et al., Nature 421, 526 (2003). aiding escape from noxious stimuli (nocicep-
true, then a prominent snout and lips, which 6. S. Conway Morris, J.-B. Caron, Nature 512, 419 (2014). tion), and affective behaviors that are linked
arise from the neural crest, are a later innova- 7. Q. Tian et al., Science 377, 218 (2022). to suffering and negative emotions associated
8. D. Jandzik et al., Nature 518, 534 (2015).
tion of the living vertebrate group. 9. J. A. Gillis et al., Nat. Commun. 4, 1436 (2013). with acute and chronic pain. Therapeutically
To discriminate between different hypoth- 10. C. Hirschberger et al., Mol. Biol. Evol. 38, 4187 (2021). relevant findings were that repetitive ap-
11. J. Mallatt, Zoolog. J. Linn. Soc. 117, 329 (2008).
eses, unequivocal correlates of arch identities 12. P. Janvier, Early Vertebrates (Oxford Monographs on plication of 5-dB sound over ambient levels
are needed. A mandibular arch is not defined Geology and Geophysics, Clarendon Press, 1996).
by being the most anterior pharyngeal arch. 13. S. Kuratani, Evol. Dev. 14, 76 (2012). 1
Institute of Pharmacology, Heidelberg University,
14. T. Miyashita, Biol. Rev. Camb. Philos. Soc. 91, 611 (2016). Heidelberg, Germany. 2Department of Functional
15. T. Miyashita, Can. J. Zool. 98, 850 (2020). Neuroanatomy, Institute for Anatomy and Cell Biology,
Canadian Museum of Nature, Ottawa, Ontario K1P 6P4, Heidelberg University, Heidelberg, Germany.
Canada. Email: tmiyashita@nature.ca 10.1126/science.adc9198 Email: rohini.kuner@pharma.uni-heidelberg.de

led to analgesia lasting multiple days How can sound affect pain perception? liculus, or association areas where
and that ongoing pain and inflam- Sounds reach the brain via the auditory pathway, signaling through the gustatory (taste), olfactory, visual,
matory hypersensitivity were sup- thalamus into the auditory cortex (A1; light red arrows). External pain auditory, and somatosensory inputs
pressed. Although this experimental stimuli activate a subset of peripheral sensory neurons, creating signals converge, such as the insula cortex.
paradigm is not equivalent to using that are processed in, and travel along, the spinal cord and thalamic Previous concepts on using mu-
music and pleasant sounds to evoke nuclei before being conveyed to neocortical areas that generate the sic and sounds for pain relief at-
analgesia in humans, the study of perception of pain (blue arrows). Zhou et al. identified a circuit (red tributed their effects to distraction-
Zhou et al. opens up new directions arrow) that mediates sound-induced analgesia. Other possible circuitry associated analgesia and reduction
for research on sound-induced anal- is shown using dashed arrows. of anxiety (8). Although sound likely
gesia by creating a model in which contributes to distraction, the study
the mechanistic underpinnings can Neocortical level by Zhou et al. reveals that sound-
be investigated. induced analgesia is a specific
Prefrontal, S1 and S2 Association A1
Neuroimaging studies on the hu- cingulate somatosensory cortices auditory mechanistic entity in its own right,
man brain suggest that a variety supported by the observation that
of pain-associated regions show suppression of pain outlasted the
changes in activity during music or application of sound by several days.
sound therapy, such as the prefron- Thalamic level Similarly, neither a 5-dB change over
tal, cingulate, and insular cortices; Mediodorsal
Posterior and Medial ambient noise nor manipulations of
the thalamic relay nuclei; and the ventral posterior geniculate the auditory cortex–thalamic path-
nucleus
nuclei nucleus
brainstem centers gating descending way led to any changes in anxiety or
pathways (3). Moreover, recent stud- stress levels. The lack of redundancy
Sound
ies point to a key role for the insular of sound-induced analgesia with
and prefrontal cortices in integrat- Somatic mechanisms of analgesia elicited by
noxious Brainstem level
ing auditory inputs with pleasant inputs Spinal cord/
distraction, stress, or anxiolysis, as
or negative affective emotions (4). Superior Auditory well as by opioids, have important
trigeminal
colliculus nuclei
Analyses of how sound is encoded nucleus clinical implications. In particular,
have recently revealed a rapid tran- sound-induced analgesia could be
sition of brain activity from the pri- implemented in parallel to these
mary auditory cortex to higher-order cortical of the auditory cortex to the somatosensory approaches to additively, and perhaps even
association areas—including the insula cor- thalamus is surprising. Furthermore, the synergistically, enhance pain relief and to
tex—within 220 ms of the onset of the audi- involvement of the higher-order PO, which reduce opioid dosage. These attractive pos-
tory stimulus, suggesting direct connectivity receives a strong corticothalamic input sibilities need rigorous experimental testing
(5). However, contrary to the expected corti- from the primary somatosensory cortex (6), and translation to humans.
cocortical interaction, Zhou et al. found that is unexpected because it mostly contributes The interaction between sound and pain
a corticothalamic pathway mediates sound- to corticothalamocortical processing and, to is a double-edged sword, so caution in in-
induced analgesia (see the figure). This was a lesser extent, receives input from the spi- terpreting these data is warranted. Certain
achieved using high-resolution tracing of nal cord (6). However, the auditory cortex sounds can trigger pain, such as headaches,
anatomical connections originating from could influence pain processing by modu- or intensify the pain experience, such as pho-
the auditory cortex and live monitoring of lating the relay of nociceptive signals from nophobia, which is common to migraines.
cellular activity in their projection areas in the periphery as well as by affecting corti- Nonpreferred music associated with in-
awake mice. When Zhou et al. systematically cothalamocortical processing of somato- creased pain ratings has been reported to
manipulated efferent projections from the sensory signals, a system that is currently further enhance gamma band neural oscil-
auditory cortex to diverse brain regions using poorly understood. Burst and tonic firing lations that are evoked by painful stimuli in
light-induced inhibition of neuronal activity modes of thalamic relay neurons are im- the somatosensory cortex, whereas this was
(optogenetic silencing), they found that only portant in the modulation of pain, and de- not true for preferred music (9). The study
the connectivity between the auditory cortex ciphering the cellular mechanisms through by Zhou et al. lays a path for testing the neu-
and the thalamic posterior nucleus (PO) and which auditory inputs affect these will help robiological basis of these interesting asso-
ventral posterior nucleus (VP) is function- further elucidate how sound-induced anal- ciations in rodent circuitry and manipulating
ally important for sound-induced analgesia. gesia comes about. them for optimizing pain relief, which may
Conversely, selective activation of this path- What is special about sound that is 5 dB be translated to humans. j
way in the absence of sound was sufficient to above ambient noise? This is an intriguing
recapitulate sound-induced analgesia. question that might be best studied in an
1. L. C. Cole, G. LoBiondo-Wood, Pain Manag. Nurs. 15, 406
What makes the thalamus such an impor- interdisciplinary manner. Detecting a small (2014).
tant target for audio-somatosensory interac- increase over ambient noise would permit 2. W. Zhuo et al., Science 377, 198 (2022).
3. C. E. Dobek et al., J. Pain 15, 1057 (2014).
tions? All somatosensory inputs—i.e., those animals to engage innate defense behav- 4. M. E. Sachs et al., Neuroimage 218, 116512 (2020).
leading to the perception of touch, pressure, iors, which have been shown to be driven 5. L. Bonetti et al., Neuroimage 245, 118735 (2021).
6. R. Kuner, T. Kuner, Physiol. Rev. 101, 213 (2021).
GRAPHIC: KELLIE HOLOSKI/SCIENCE
vibration, temperature, and pain—get re- through corticofugal projections arising

7. H. Wang et al., PLOS Biol. 17, e3000417 (2019).
layed to the cortex through excitatory pro- from the auditory cortex (7). From an evo- 8. C. R. Millet, L. F. Gooding, J. Music Ther. 54, 460 (2018).
jection neurons of the VP, rendering them a lutionary perspective, it makes sense that 9. M. Hauck et al., Pain 154, 539 (2013).
prime site to modulate the magnitude of the sound-driven defense behaviors are acutely ACK N OW LE D G M E N TS
forwarded nociceptive signal (6). In general, accompanied by pain suppression. Along The authors acknowledge funding by Deutsche
such modulatory influences on thalamic re- these lines, it will be insightful to study ar- Forschungsgemeinschaft (SFB1158, project nos. B01, B06,
lay neurons are known, but a descending eas in the brain that play a role in multisen- and B08).
corticothalamic input from layer 5 neurons sory integration, such as the superior col- 10.1126/science.add0640

channel at ~1-km depth. Here, water tempera-
P OLICY FORUM ture, salinity, and pressure create a minimum
in sound speed. Propagating sound bends to-
MARINE POLICY ward, and may become trapped in, low-speed
water layers and thus efficiently travel over
Noise from deep-sea mining great distances. Our model predicts that min-
ing noise from a single operation will exceed
ambient noise levels experienced in gentle
may span vast ocean areas weather conditions (~94 dB re 1 µPa) to a

range of ~500 km (9) (see the figure).
If each contractor operated one mining
Potential harm is understudied and largely overlooked system, a radius of 4 to 6 km about each
mine would exceed 120 dB re 1 µPa, a US
National Marine Fisheries Service thresh-
By Rob Williams1, Christine Erbe2, through, e.g., acoustic masking, behavior dis- old denoting the possibility of behavioral
Alec Duncan2, Kimberly Nielsen1, ruption, stress, and hearing loss (4). Here we impacts, generically, to marine mammals.
Travis Washburn3, Craig Smith4 focus on extraction of polymetallic nodules Roughly 5.5 million km2 would be ensonified
in the Clarion-Clipperton Zone (CCZ), the above gentle-weather ambient conditions
D
espite increasing interest in deep-sea abyssal region where Nauru’s and 16 other (see the figure) (9). Some acoustically sen-
mining, there are long-standing con- nodule-mining exploration contract areas are sitive whales show behavioral responses to
cerns about environmental impacts concentrated, to illustrate the broad issue of low-frequency continuous noise <120 dB re
on vulnerable and poorly understood potential ocean noise from mining. Many ce- 1 µPa (10). Most species in the deep sea have
ecosystems (1–3). These concerns took tacean species occur throughout the CCZ, in- yet to be described, and sensitivities to noise
on new urgency in June 2021, when the cluding endangered migratory baleen whales have not been studied for even the most
Republic of Nauru notified the International and deep-diving toothed whales (6). common deep-sea species, so no informa-
Seabed Authority [ISA, the intergovernmen- In the absence of empirical data on sound- tion is available on their responses to noise.
tal body erected by the UN Convention on source characteristics, we used reasonable However, without sunlight, many species use
the Law of the Sea (UNCLOS) responsible for analogs to estimate potential noise levels sound and vibrations in their ecology and
managing seabed resources in areas beyond from nodule mining equipment. For mining thus are likely to be relatively vulnerable to
national jurisdiction] of intent to sponsor tools that remove nodules from the seabed, noise from human activities (4, 11).
an exploitation application for polymetallic we used coastal dredges that remove seafloor Our prediction that noise from one mine
nodule mining in the Pacific in 2 years. This gravel and sediment (181 dB re 1 µPa m, 20 will exceed gentle-weather background levels
triggered a provision in the 1994 Agreement Hz to 20 kHz) (7). Noise generated by each to distances of hundreds of kilometers ap-
related to UNCLOS, leaving 2 years for the booster pump along the riser was approxi- pears to be problematic under the require-
ISA to adopt its first set of exploitation regu- mated by pump-out operations from coastal ments of UNCLOS and for the ISA’s draft ex-
lations or, failing that, consider Nauru’s ap- dredging (183 dB re 1 µPa m, 20 Hz to 20 ploitation guidelines to assess environmental
plication under existing international law. kHz) (7). For the mining vessel, we took re- impacts of nodule mining. The UNCLOS
We argue that a critical source of potential cordings from floating production storage framework requires sufficient protection
environmental harm is understudied and and offloading (FPSO) vessels used by the oil against harmful effects through precaution-
largely overlooked: underwater noise gener- and gas industry (188 dB re 1 µPa m, 20 Hz to ary management. Under ISA guidelines,
ated by mining activities, which can disrupt 2 kHz) (8). We then applied common sound ecological metrics in areas affected by min-
ecosystems in many ways. propagation models (see supplementary ma- ing will be compared to similar metrics in
Environmental concerns about deep-sea terials) to predict how noisy full-scale opera- zones within contractor areas required to be
mining include removal and burial of critical tions may be. Our predictions are expected unimpacted, called preservation reference
seafloor habitat and large suspended sedi- to be underestimates. Shallow-water sur- zones (PRZs). However, to fit within contrac-
ment plumes, but underwater noise gener- rogates likely missed acoustic energy below tor areas, PRZs are expected to be within 100
ated by mining activities has received little 100 Hz that will be particularly important for to 300 km of mining sites, causing PRZs to
study, with limited modeling and measure- heavier machinery required by seabed min- be ensonified by mining activities and thus
ment in environmental impact assessments ing. Sonars, support vessels, and additional of limited value as controls for understand-
(4). For example, although sound was con- pipes and pumps for discharge of dewatering ing impacts of mining noise (3, 12) (see the
sidered in the environmental impact state- plumes represent additional sound sources figure). To be within ambient levels, it may
ment for the Solwara 1 project (5), informa- that we have not yet considered. be necessary to place PRZs outside the entire
tion was so scarce 15 years ago that only the Our predictive model shows that the CCZ; however, this would likely cause the
topside vessel was modeled. Sound is the pri- acoustic environment surrounding a mining PRZs to be ecologically unrepresentative of
mary modality for many marine organisms operation in the CCZ will likely be substan- the areas mined (13).
to probe their environment and to communi- tially altered by even the incomplete list of Given the costs and logistical challenges
cate. Noise from human activities can affect activities modeled. Noise will be distributed to answering basic questions about sound
organisms ranging from plankton to whales throughout the water column from ships at impacts on deep-sea ecosystems, it seems
the surface, risers with pumps every ~1 km unlikely that adequate data to assess eco-
1
Oceans Initiative, Seattle, WA, USA. 2Centre for Marine in depth, and mining tools on the seabed, ul- logical risks from mining noise will be col-
Science and Technology, Curtin University, Perth, WA, timately creating a cylinder of sound gradu- lected within ~18 months, or even within
Australia. 3National Institute of Advanced Industrial Science ally decreasing in amplitude while propagat- a 10-year mining moratorium proposed in
and Technology, Tsukuba, Ibaraki, Japan. 4Department of
Oceanography, University of Hawai’i at Mānoa, Honolulu, HI, ing outward. Mid-water pumps will couple 2019 (12, 14). Risk assessments will be even
USA. Email: rob@oceansinitiative.org with the Sound Fixing And Ranging (SOFAR) more challenging without access to empiri-

I NS I GHTS | P O L I C Y F O RU M
0
Behavioral
Deep-sea mining noise from surface to seabed
A depth section along a line due north from one mine operating in the western Clarion-Clipperton Zone (CCZ)
1000 threshold
for cetaceans
shows the model-derived root-mean-square sound pressure level (SPL) over 20 Hz to 20 kHz (left).
(120 dB re 1 µPa) A map shows the maximum SPL (20 Hz to 1 kHz) over all water depths for the combined sound field from
2000 17 simultaneously operating mines, with one in the area of each exploration contract most likely to be mined
Depth (m)
first (i.e., area of highest nodule density) (bottom). The maximum modeled range is 1000 km from each
3000 source. The 120 dB re 1 µPa behavioral threshold for cetaceans corresponds to the yellow line. Gentle ambient
noise of 94 dB re 1 µPa is exceeded within the white line.
4000
60 70 80 90 100 110 120 130 140
5000
Noise level (dB re 1 µPa)
0 2 4 6 8 10
Distance from source (km) Pacific Ocean MEXICO
Noise exceeds ambient

background level (94 dB)
Areas reserved
for future mining
Exploration areas
of current mining
interest
0 1000
Areas of Particular Environmental Interest Behavioral threshold for cetaceans (120 dB re 1 µPa)
km
cal data on sound-source characteristics of plications within the next 18 months without be coupled with intensive, independent stud-
nodule-mining equipment. Testing of re- data transparency and rigorous standards ies of regional baselines and environmental
duced-scale, incomplete, prototype mining and guidelines in place would represent the impacts, as well as with the ability to alter
machines is underway; however, contrac- start of a large-scale, uncontrolled experi- or halt operations quickly if newly acquired
tors, thus far, have not published sound- ment. If this industry does proceed through data indicate notable unexpected effects. j
source characteristics from internal risk Nauru’s claim or elsewhere, we recommend
assessments or pilot studies. We urge con- application of the precautionary principle (in
1. J. Halfar, R. M. Fujita, Science 316, 987 (2007).
tractors to release in a timely manner in- keeping with the ISA’s duty to provide for ef- 2. K. J. Mengerink et al., Science 344, 696 (2014).
formation on sound-source characteristics fective environmental protection) such that 3. C. R. Smith et al., Trends Ecol. Evol. 35, 853 (2020).
4. C. Martin, L. Weilgart, D. J. Amon, J. Müller, Deep-Sea
of all seabed-mining components (e.g., from the ISA allows only one or two mining opera- Mining: A noisy affair (2021); https://www.oceancare.
lifting pumps on riser pipes, and rattling of tions to proceed until the long-term cumula- org/wp-content/uploads/2021/11/DeepSeaMining_a-
nodules in lifting pipes and on barges). tive environmental impacts from noise (and noisy-affair_report_OceanCare_2021.pdf.
5. Coffey Natural Systems: Environmental Impact
Transparency from industry, agencies, other stressors) are well documented (13). Statement: Nautilus Minerals Niugini Limited, Solwara
and scientists related to the acoustic energy The ISA is expected to require that noise 1 Project, Vol. A, Coffey Natural Systems, Queensland,
Australia (2008).
produced from mining is urgent. We need impact be considered in environmental im- 6. R. Williams et al., Ecography 37, 212 (2014).
to learn about mining systems and their pact assessments for future exploitation con- 7. K. J. Reine, D. Clarke, C. Dickerson, J. Acoust. Soc. Am.
impacts iteratively, through real-world ex- tracts, but noise impacts will be very difficult 135, 3280 (2014).
8. C. Erbe et al., J. Acoust. Soc. Am. 133, EL465 (2013).
perience (15). As such, the regulatory regime to assess without substantially more data on 9. F. Niu et al., Mar. Pollut. Bull. 173, 113057 (2021).
should start with cautious thresholds for mining noise and ecosystem sensitivities. 10. B. L. Southall et al., Aquat. Mamm. 33, 411 (2007).
11. M. A. Wale, R. A. Briers, K. Diele, Mar. Pollut. Bull. 173,
noise, enable rapid management responses The ISA is tasked with protecting the marine 112958 (2021).
should thresholds be approached, and in- environment from “serious harm” from sea- 12. D. J. Amon et al., Mar. Policy 138, 105006 (2022).
GRAPHIC: K. FRANKLIN/SCIENCE
clude regular review of thresholds as more bed mining, but it has yet to define what con- 13. C. R. Smith, M. R. Clark, E. Goetze, A. G. Glover, K. L.
Howell, Front. Mar. Sci. 8, 797516 (2021).
data about noise impacts are reported. Our stitutes serious harm, including unaccept- 14. A. Kung et al., J. Environ. Manage. 279, 111593 (2021).
first approximation of noise generated by able noise levels. Thresholds should be set at 15. C. S. Holling, Adaptive Environmental Assessment and
Management (Wiley, 1978).
polymetallic nodule mining highlights both precautionary levels, and contractors should
the potential extent and uncertainty of this be required to collect environmental and ACK N OW LE D G M E N TS
noise and its ecological impacts. mining noise data and make them publicly R.W. and C.E. contributed equally to this work.
Moving forward with nodule mining ap- available. We recommend that mining should 10.1126/science.abo2804

Robotic spacecraft like the Mars 2020 rover
are still reliant on human operators.
spaceflight, but not all. Astronauts have per-

formed or participated in scientific research,
including the 2018 production of Bose-
Einstein condensate in the ISS Cold Atom
Lab. But they do so at great expense and
at great risk to themselves. Having humans
embark on longer missions farther into the
Solar System only increases cost and risk and
brings diminishing scientific returns, main-
tain Goldsmith and Rees.
The authors expand their argument over
the course of nine chapters, exploring the
challenges, hazards, and expenses of sending
humans into orbit or to the Moon, Mars, as-
teroids, or space colonies. Humans, they ar-
gue, may be able to do more science quickly,
B O OKS et al . but robotic exploration will, over time, yield
more science per dollar and with less risk to
human life. And the robots we use will only
PLANETARY STUDIES become progressively more capable.
Computer technologies have grown not
Interrogating the final only more capable and reliable in recent

years but also much smaller and lighter—a
phenomenon that pervades our daily lives as
frontier from afar much as it does spacecraft design. The same

is true for imaging technologies and elec-
tronics in general. Rovers have gotten larger
Robots are better equipped for space exploration over time, but the instruments they carry use
than humans, argue a pair of authors increasingly miniaturized technologies.
For a book that is ostensibly designed
to sell the reader on a future of robotic ex-
By Matt Shindell mercial crew capsules that have delivered ploration, there is a surprising absence of
astronauts and cosmonauts to the station. any detailed history of the development of
A
t the dawn of the space age, it was With China now operating a station of its robotic exploration, systems engineering,
assumed that humans would be own, and commercial firms vying to build the experts who make robotics possible, or
necessary for planetary exploration. new stations, these activities show no sign the sites where this expertise is produced
Computers were large and limited, of slowing. NASA now plans to return hu- or practiced. While the authors do provide
robots stiff and clunky. Humans, mans to the Moon, as do other examples of robotic spacecraft
meanwhile, were versatile and cre- national programs. Even so, the accomplishments, they offer only
ative manipulators of their environment history of space exploration has vague projections about how
whose eyes and brains would be needed largely left humans safely in con- advancements in artificial intel-
to make discoveries in situ. Early popular trol rooms on Earth, operating ligence (AI) will continue to im-
depictions of space exploration featured the robotic craft that explore the prove robotic capabilities.
ships filled with people performing space Solar System. Another notable absence is any
heroics. Robots, in television shows like That humans should leave the discussion of the human work be-
Star Trek, were few and far between; even exploration of space to robots is The End of Astronauts hind robotic exploration. Aside
the starship Enterprise’s computer, as pow- the central argument of science Donald Goldsmith from the design and production of
erful as it was, required a human interface. writer Donald Goldsmith and as- and Martin Rees spacecraft, teams of humans oper-
These visions of spaceflight may yet tronomer Martin Rees in The End Belknap Press, 2022. ate these craft, while other teams
192 pp.
come true—humans may eventually start of Astronauts. We may want to set science goals and analyze re-
taking to space in large numbers, building see astronauts exploring new worlds, and we turned data. The AI integrated into robotic
larger ships and space stations and living may wish to be among them, they acknowl- spacecraft today is minimal. People operate
entire lifetimes in space. We have continu- edge, but humans—at least physically—do spacecraft, even as spacecraft have facilitated
ously occupied low Earth orbit through- not make ideal space explorers. The human the emergence of the interdisciplinary fields
out the entire 21st century, thanks to the body is far too vulnerable to the hazards of of planetary science and robotic exploration.
PHOTO: NASA/JPL-CALTECH
International Space Station (ISS), the space space and the traumas of spaceflight. And In leaving the people out, the authors
shuttle, and the various Russian and com- as much as spacecraft and space suits have make it seem as though the spacecraft can—
improved, they remain insufficient to keep or someday will—do everything on their
humans safe during long journeys. own. But space exploration, no matter how
The reviewer is a space history curator at the
Smithsonian’s National Air and Space Museum, Much of the research done by humans in it is conducted, is human exploration. j
Washington, DC, USA. Email: shindellm@si.edu space thus far has been related to human 10.1126/science.abq7394

I N S I G HT S
PSYCHOLOGY
The cultural landscape of emotions

Social context plays an important role in shaping how we feel
By Asifa Majid herself ultimately concludes that Western- Between Us:

ers have OURS emotions. How Cultures
Create Emotions
I
n the animated film Inside Out, 11-year- Taken as a whole, however, the book
Batja Mesquita
old Riley’s emotions are personified as contributes much to the discussion of the Norton, 2022. 304 pp.
brightly colored internal figures that origins of emotions, presenting a dazzling
drive her behaviors. The same five emo- array of cross-cultural studies interspersed
tions—anger, fear, disgust, sadness, and with personal anecdotes about immigrants’
joy—appear in every other character’s struggles to reconcile diverse emotional leaves hanging some questions. If emotional
head as well, functioning in much the same and social worlds. In chapter 8, for exam- worlds are truly constructed on the fly ac-
way in each individual. In Western cultures, ple, Mesquita describes an incident where cording to individual histories and identities,
this is the case, argues psychologist Batja she—a Dutch native living in the United for example, then how is intersubjectivity
Mesquita in Between Us. Emotions in such States—bumped into the eminent American possible? Mesquita claims that emotion lexi-
contexts, she writes, are considered “MINE,” psychologist Hazel Markus at a conference cons differ radically and are not amenable to
or “Mental, INside the person, and Essen- Markus helped organize. Wishing to express direct translation, so language cannot serve
tialist,” the latter defined in the book as al- empathy with Markus’s workload, Mesquita as a bridge between cultures.
ways having the same properties. declared “You look a little tired.” The re- But differences in words do not necessar-
This conception of emotion is not uni- mark appeared to fluster Markus but was ily entail differences in emotion concepts.
versal, however. Emotions else- There can be shared prototypes
where, she argues, are thought of despite variations around the
as “OURS”—“OUtside the person, periphery, distinct nonlinguistic
Relational, and Situated.” Using concepts represented by a single
this distinction, Mesquita sets label, and mental content where
about contrasting emotions in there are lexical gaps. Mesquita’s
“the West,” where the individual is theory also fails to account for the
the top concern, with “the Rest,” possibility that we can be wrong
where community is prioritized. about our own emotions—for ex-
Mesquita describes amae as ample, I thought I was scared, but
a central emotion in Japanese now I realize I was actually anx-
culture, where it builds interde- ious—because, in her view, there
pendence by fostering nurturing is no emotional truth to be known,
indulgence, and hasham—which only one to be constructed. And it
encompasses shame, embarrass- is unclear how to incorporate ev-
ment, and social respectability— idence from prelinguistic infants
as a fundamental emotion for or nonhuman primates into this
Egyptian Bedouins. Such obser- particular theory of emotions, or
vations provide a backdrop for how to account for the recurrent
her to explore a range of issues, emotional motifs that appear
including childhood socializa- Sociocultural influences underlie our emotions, argues Mesquita. across cultures—grief at the loss
tion, the nature of friendship, the of a loved one, for example, or dis-
role of language in shaping emotions, and intended as an expression of sympathy— gust upon smelling something rancid.
cross-cultural communication in a glo- to sympathize in Dutch is to acknowledge Sociocultural and biological factors both
balized world. suffering, not offer a silver lining as in the contribute to emotions, but exactly how
Despite Mesquita’s emphasis on US. Navigating emotional worlds across cul- large a role is played by either remains
cross-cultural emotions, there is little tures is full of hidden pitfalls, we learn. controversial. Mesquita emphasizes the
discussion of whether the MINE-OURS The book’s take-home message is radical: importance of the larger social context
PHOTO: GALINA ZHIGALOVA/ALAMY STOCK PHOTO
dichotomy accurately articulates global There are no natural emotions, no innate in shaping emotions and urges empathy
cultural variation. Other scholars have emotions, no universal emotions. Mesquita and understanding of cultural differences.
noted, for example, that hunter-gatherer argues that emotions are “meaning mak- These are valuable lessons, whatever the-
societies simultaneously emphasize both ing” and “a preparation for action” and that ory of emotion one subscribes to. j
individual autonomy and social coopera- the idea of “emotions as inner states” is a
tion (1, 2). And in an apparent contradic- Western construct. Instead, she suggests
1. K. M. Endicott, in Anarchic Solidarity: Autonomy,
tion to her earlier arguments, Mesquita that emotions are a “dance” cocreated be- Equality, and Fellowship in Southeast Asia, T. Gibson,
tween people who live in a specific cultural K. Sillander, Eds. (Yale Univ. Southeast Asia Studies,
context at a particular historical moment. 2011), pp. 62–87.
The reviewer is at the Department of Experimental 2. E. Wnuk, Y. Ito, Cogn. Linguist. 32, 195 (2021).
Psychology, University of Oxford, Oxford OX3 9DU, UK. By pushing a strongly social construction-
Email: asifa.majid@psy.ox.ac.uk ist account of emotions, Mesquita raises but 10.1126/science.abq8098

of Natural History, SE 10405 Stockholm, Sweden.
7
Missoula, MT 59802, USA. 8Department of Biology,
Lund University, SE 22362 Lund, Sweden. 9School
of Life Sciences, Arizona State University, Tempe,
AZ 85287, USA. 10Department of Marine Sciences,
University of Gothenburg, Tjärnö, Sweden. 11National
Marine Fisheries Service, Northwest Fisheries
Science Center, Seattle, WA 98112, USA. 12College
of Forest Resources & Environmental Science,
Michigan Technological University, Houghton,
MI 49931, USA. 13Turner Endangered Species
Fund, Bozeman, MT 59718, USA. 14Department of
Environmental Research and Monitoring, Swedish
Museum of Natural History, SE 10405 Stockholm,
Sweden. 15Conservation and Evolutionary Genetics
Group, Estación Biológica de Doñana–Consejo
Superior de Investigaciones Científicas, 41092
Sevilla, Spain. 16Forest Ecology Unit, Research and
Innovation Centre–Fondazione Edmund Mach, San
Michele all’Adige 38010 (TN), Italy.
*Corresponding author.
Email: linda.laikre@popgen.su.se
1. Sveriges Riksdag,“Committees want to see changes in
species protection and in the management of wolves”
(2022); www.riksdagen.se/sv/aktuellt/2022/maj/5/
LET TERS utskott-vill-se-andringar-i-artskyddet-och-i-
forvaltningen-av-varg/ [in Swedish].
2. M. M. Hansen et al.,“Evaluation of the conservation
genetic basis of management of grey wolves in Sweden”
Immigration and population growth are crucial to making Sweden’s inbred wolf population sustainable. (Swedish Government Investigation SOU 2011:37, 2011).
3. L. Laikre et al., Heredity 117, 279 (2016).
4. O. Liberg et al., Biol. Lett. 1, 17 (2005).
5. M. Åkesson et al., Conserv. Genet. 23, 359 (2022).
Edited by Jennifer Sills population (6). The genetic base is thus 6. M. Åkesson, L. Svensson,“The pedigree of the
extremely narrow, and genomic erosion Scandinavian wolf population up until 2021: Report from
Planned cull endangers has been confirmed (7, 8). The average
level of inbreeding is similar to the level
the Swedish Wildlife Damage Centre 2022–3” (Swedish
University of Agricultural Sciences, 2022)
[in Swedish].
Swedish wolf population found in the offspring of two full siblings
(6). Inbreeding in this population has
7. M. Kardos et al., Nat. Ecol. Evol. 2, 124 (2018).
8. A. Viluma et al., Genome Res. 32, 1 (2022).
9. J. Räikkönen et al., PLOS ONE 8, e67218 (2013).
In May, the Swedish Parliament announced been shown to reduce litter size (4). Also, 10. Swedish National Veterinary Institute (SVA),“License hun-
a goal to reduce the Swedish wolf popula- high frequencies of anatomical defects (9) ting of wolves 2021, report 65/2021” (2021) [in Swedish].
tion from about 400 to about 200 individu- and male reproductive disorders (10) have 11. P. W. Hedrick et al., Anim. Conserv. 22, 302 (2019).
12. CBD,“Recommendation adopted by the Subsidiary Body
als (1). This action further threatens this been observed. on Scientific, Technical, and Technological Advice” (2021),
highly endangered population, which is To make this population viable, popula- p. 10, Headline Indicator A.0.4; www.cbd.int/doc/
genetically isolated and inbred. Scientific tion size and immigration must increase. So recommendations/sbstta-24/sbstta-24-rec-02-en.pdf.
advice for improvements has not been far, the population has been too small, and COM P E T IN G IN T E RESTS
implemented (2, 3). limited immigration followed by inbreeding This work is the view of the authors and not necessarily that of
The Swedish Parliament proposed this could lead to extinction, similar to the Isle the agencies they represent.
drastic cull at a time when biodiversity is a Royale wolf population (11). The goal should 10.1126/science.add5299
global focus. The 50-year anniversary of the be to recreate a well-connected metapopula-
first UN conference on the environment was
celebrated in June, and the UN Convention
tion spanning Scandinavia and Finland with
a genetically effective population size of over
End animal testing for
on Biological Diversity (CBD) will soon final-
ize its global biodiversity framework for
500, in line with the proposed CBD indicator
(12). Considerably more genetic exchange
biosimilar approval
2020 to 2050. Sweden’s actions are inconsis- than the current one-migrant-per-generation Drug toxicology testing in animals has long
tent with the country’s obligations under the aim is needed (3). been a standard requirement for establish-
CBD and European Union law. Linda Laikre1*, Fred W. Allendorf2, Jouni ing the safety of both new drugs (1) and
Few wild populations are as well studied Aspi3, Carlos Carroll4, Love Dalén1,5,6, Richard copies of biological drugs coming off pat-
as the Scandinavian wolf. Genetic monitor- Fredrickson7, Christina Hansen Wheat8, Philip ent, known as biosimilars (2). Recently, the
ing has provided a full pedigree since the Hedrick9, Kerstin Johannesson10, Marty Kardos11, international community has acknowledged
PHOTO: RONALD KAMPHIUS/MINDEN PICTURES
population was reestablished in the 1980s Rolf O. Peterson12, Mike Phillips13, Nils Ryman1, that this type of test may not be necessary
after extinction, and the data confirm per- Jannikke Räikkönen14, Carles Vilà15, Christopher W. or useful. Although policies for new drug
sisting genetic isolation (4–6). Hunting, Wheat1, Cristiano Vernesi16, John A.Vucetich12 approval are in the process of changing,
1
Department of Zoology, Stockholm University, SE
conducted both legally and illegally, has pre- biosimilar approval policies have been over-
10691 Stockholm, Sweden. 2Division of Biological
vented population expansion and the influx Sciences, University of Montana, Missoula, MT looked. Regulatory agencies should update
of genetic variation. 59812, USA. 3Ecology and Genetics Research these policies to streamline the biosimilars
Three founders comprised the popula- Unit, University of Oulu, FIN-90014 Oulu, Finland. approval process and to prevent unneces-
4
Klamath Center for Conservation Research,
tion’s genetic origin until 2007, and only sary, and thus unethical, animal testing.
Orleans, CA 95556, USA. 5Centre for Palaeogenetics,
three more wolves have subsequently SE 10691 Stockholm, Sweden. 6Department of Policies requiring animal toxicology stud-
contributed genetically to the present Bioinformatics and Genetics, Swedish Museum ies to test biosimilars often stipulate the

I N S I G HT S
use of a dose multiple times as high as the

human dose (3). This strategy fails to rec- LIFE IN SCIENCE
ognize that higher doses lead to nonlinear
responses, which invalidates the results. In
addition, the animal species used in many
Danger in the desert
studies do not have the binding receptors Deep in Chile’s Atacama Desert, a landscape often compared to Mars, our team
that the drugs target in humans, or they of four female microbiologists watched as a car full of curious men pulled up be-
have similar receptors that bind to drugs side us. Because we were strangers in a desolate place, our minds immediately
at concentrations different from those in jumped to ways we could protect ourselves. Defensively, our Chilean compatriot
humans. Given the differences in receptors, hefted the sturdy shovel she’d been using to dig up plant roots. The rest of us
either animals cannot respond or develop tried to look braver than we felt.
side effects to the drug, or their response We had come to this desert to conduct DNA studies on giant horsetails, fern-
lacks relevance to the human response (4). related plants that somehow flourish in one of Earth’s driest places. We were
The US Food and Drug Administration searching for plants in the most remote locations, where
(FDA) is discussing alternative models they would be unaffected by human activities such as min-
Call for submissions
of testing for new drugs (5, 6) but still Life in Science is an
ing and agriculture.
requires animal toxicology studies for occasional feature We’d been warned that the trip could be dangerous. Because
biosimilars (2). The requirement remains highlighting some we were traveling so far from fuel sources, we were told to
even though the FDA has discarded most of the humorous or take along a can of gasoline. Our destination was at the end
unusual day-to-day
of these studies, which are often conducted realities that face of a tortuous single-lane dirt road lined with burned-out vehi-
by the dozens for a single biosimilar drug our readers. Can you cles that had not successfully negotiated the steep descent.
(7, 8). The World Health Organization and top this? Submit Our sample site was near a village, and the people, we were
your story to www.
other regulatory agencies [e.g., (3)] also submit2science.org. told, might not respond positively to us. We were instructed
still require animal toxicology studies for to report our travel plans at the nearest police station so that
biosimilar approval (9). Only the European search parties would know where to look for us if we disappeared.
Medicines Agency (10) has updated its We had found the amazing plants; their bright green stems towered over our
policy to reflect that animal pharmacology heads, evoking thoughts of ancient Carboniferous swamp vegetation. The men
and toxicology studies are irrelevant to the approached as we finished collecting our samples. We waited tensely as a man
evaluation of biosimilars. exited the car and walked toward us. To our surprise and relief, he politely invited
More than 120 biosimilars have been us to visit their village—they wanted to show us a lovely church of which they
approved in the United States and the were justifiably proud. That day, we learned about more than the microbiomes
European Union (7), and no results from the that help desert plants thrive. We also met a welcoming community who had
required animal studies have yielded useful likewise beautifully adapted to their challenging home.
information about responses or side effects Linda Graham
in humans. It is time for regulatory agen- Department of Botany, University of Wisconsin–Madison, Madison, WI 53706, USA.
cies to forbid animal testing for biosimilar Email: lkgraham@wisc.edu
approval for the sake of animal welfare,
cost, and time.
Sarfaraz K. Niazi
Department of Pharmaceutical Sciences, College The author’s team
of Pharmacy, University of Illinois, Chicago, IL negotiated remote
60612, USA. Email: sniazi3@uic.edu roads deep in Chile’s
REF ERENC ES AND NOTES Atacama Desert.
1. S. W. Junod,“FDA and clinical drug trials: A short history”
(FDA, 2014); www.fda.gov/media/110437/download.
2. C. B. Rangel,“H.R. 3590—111th Congress (2009–2010):
Patient Protection and Affordable Care Act” (2010), www.
congress.gov/bill/111th-congress/house-bill/3590.
3. “Guidelines on similar biologics: Regulatory
requirements for marketing authorization in India”
(2016); https://cdsco.gov.in/opencms/resources/
UploadCDSCOWeb/2018/UploadAlertsFiles/
BiosimilarGuideline2016.pdf.
4. P. van Meer et al., Drug Discov. Today 20, 483 (2015).
5. FDA,“Publications co-authored by FDA on
alternative methods” (2022); www.fda.gov/
science-research/about-science-research-fda/
publications-co-authored-fda-alternative-methods.
6. A. M. Avila et al., Reg. Toxicol. Pharm. 114, 104662 (2020).
7. S. K. Niazi, Biologics 2, 107 (2022).
8. S. K. Niazi, Expert Opin. Biol. Ther. 22, 149 (2022).
9. World Health Organization (WHO),“Guidelines on
evaluation of biosimilars” (2022); https://cdn.who.
int/media/docs/default-source/biologicals/who-
guidelines-on-evaluation-of-biosimilars---4-nov-2021.
pdf?sfvrsn=f17799ae_5.
PHOTO: MARIE TREST
10. European Medicines Agency,“Tailored scientific advice for

biosimilar development” (2021); www.ema.europa.eu/en/
documents/report/tailored-scientific-advice-
biosimilar-development-report-experience-
pilot-2017-2020_en.pdf.

RESEARCH
IN S CIENCE JOU R NAL S
Edited by Michael Funk
ANTHROPOLOGY
The emergence of
cumulative culture
E
xperimental research has found that individ-
uals with no prior knowledge spontaneously
eously
produce the types of stone tools associated
ciated
with early hominins. One of the hallmarksrks
of modern humans is the capacity for
cumulative culture: the ability for knowledgege to
accumulate across generations through cultural
ultural
transmission. Archaeologists have thoughtt that
this capacity first emerged with the appear- r-
ance of the Oldowan stone tool industry about
bout
2.6 million years ago. Snyder et al. tested this
his
supposition by presenting naïve individuals s with
a puzzle that could be solved by making simple
mple
core and flake tools, finding that individuals
s
taught themselves to make artifacts that match
Oldowan industry examples. These results sug-
gest that the capacity for cumulative culturere is
more recent than the Oldowan industry. —SGO
Sci. Adv. 10.1126/sciadv.abo2894 (2022).
Experiments with modern humans suggest that

specialized training was not necessary for
early hominins to craft Oldowan industry tools.
NEUROSCIENCE outgrowth a and

nd arborization, as preparation of the reactants ACTUATORS
well as the expression
e ression
exp i off PPARa
PPAR
PPAR using
i procedures
procedure
d s ffoundd iin
n th
tthe
he
Boosting brain power target genes that are associ- literature. Automating the most
D
Designing
i i better
b tt
with omega-3s ated with synaptic plasticity. common such procedures ways to stretch
Activation of the nuclear receptor Supplementing the diet of young could save the research com- Dielectric elastomers are a type
peroxisome proliferator–acti- male rats with DHA increased munity a great deal of time while of electroactive polymer that
vated receptor a (PPARa) in mice the amount of 7(S)-HDHA in the also facilitating crowd-sourced can transform electrical energy
is neuroprotective and enhances cortex. —LKF optimization. Rohrbach et al. into mechanical work. They
PHOTO: APPIO STUDIOS/SHUTTERSTOCK
memory. Liu et al. found an Sci. Signal. 15, eabo1857 (2022). translated more than 100 popular are often made from silicone
endogenous ligand for PPARa in protocols into formats execut- or acrylic elastomers. Silicone
the brain that may promote these able on a synthesis machine, materials are easier to process
effects. 7(S)-HDHA, a hydroxyl- ORGANIC CHEMISTRY half of which they then validated but have lower strains, whereas
ated derivative of the omega-3 experimentally. The authors also acrylics require prestretching
fatty acid DHA, bound to PPARa
Toward automation of set up an open database to which to achieve high strains. Shi et al.
with high affinity. Treating common chemistry new executable protocols can be designed a dielectric elastomer
primary cortical neurons with Most experiments in modern added. —JSY based on acrylates in which
7(S)-HDHA promoted dendritic organic chemistry require prior Science, abo0058, this issue p. 172 the selection of monomers, the

use of both a short and long The authors found that the
IN OTHER JOURNALS Edited by Caroline Ash
cross-linker molecules, and intervention, which was com-
and Jesse Smith
the ability to control hydrogen posed of training, outreach, and
bonding enabled fine tuning of the assignment of female offi-
the stress-strain responses and cers across 180 police stations,
viscoelasticity. The authors fur- significantly increased police
ther developed a method to dry registration of cases of violence
stack the elastomeric sheets against women. These find-
without a loss of properties ings carry policy implications
that is more compatible with for efforts to improve police
current large-scale fabrication responsiveness to abused
processes. —MSL women more widely. —AMS
Science, abn0099, this issue p. 228 Science, abm7387, this issue p. 191;
see also abp9542, p. 150
STRUCTURAL BIOLOGY
CATALYSIS
Dynamic control of GPCRs
Arrestins are a group of proteins Better on the outside
that regulate signaling through One way to maximize the activ-
G protein–coupled receptors ity of scarce and costly noble
(GPCRs). They are best known metals in catalysis is to use
as an off switch in signaling small cluster or even single
through G proteins, but they atoms on metal oxide sup-
also coordinate G protein–inde- ports. Bifunctional platinum
pendent signaling. Kleist et al. and zeolite catalysts are used
took advantage of an intrinsi- industrially in alkane hydrocon-
cally b-arrestin–biased GPCR, version. Cheng et al. show that
atypical chemokine receptor 3 the amount of platinum used DEVELOPMENT
(ACKR3), to study b-arrestin could be reduced and isomer
recruitment. Nuclear mag- selectivity improved when the Tissue fusion aided by migration
netic resonance spectroscopy catalyst preparation created
D
uring development, tissues sometimes need to fuse to
experiments support a role for platinum clusters on the surface build structure. Failure of fusion in the human lip and
conformational selection. The of zeolite or its alumina binder, palate can result in cleft lip and/or palate, one of the most
inactive state shows confor- rather than in the interior of the common congenital abnormalities. The palate forms as
mational heterogeneity at the zeolite channels. Preparations precursor tissue grows to meet at the midline, above the
ligand-binding pocket. Ligand that would have formed tongue. Edges of the growing tissues are covered by squamous
binding can cause stabiliza- single-atom catalysts were less periderm cells that prevent premature fusion. Teng et al. show
tion of an active state that in active because the platinum that apoptosis plays a part in the removal of the periderm cells
turn tunes the dynamics at the ions adsorbed on the alumina but is dispensable; removal of fusion-resistant surface cells
intracellular region to allow surface were more difficult to depends crucially on epithelial migration. The epithelial surface
b-arrestin recruitment. —VV reduce. —PDS breaks up as the cells converge into migrating streams that
Science, abj4922, this issue p. 222 Science, abn8289, this issue p. 204 sweep away the surface of the developing palate precursors
before midline fusion. —PJH Development 149, dev200181 (2022).
THERMOELECTRICS
GENDER-BASED VIOLENCE Cell migration, rather than apoptosis, is crucial for correct antenatal
Tools to improve police Disordered but efficient closure of the palate.
Thermoelectric materials,
responsiveness many having relative simple
Violence against women is a compositions, convert heat
problem of global concern, and into electricity. However, Jiang CELL BIOLOGY interaction between the two
inadequate police responsive- et al. found that adding more actin-binding domains, ABD1 and
ness in many settings is a factor cations into a germanium
Organizing actin bundles ABD2. Initial binding to actin fila-
Plastins are actin-bundling pro-
in its persistence. Sukhtankar tellurium–based material ments weakens the interaction
teins that are involved in cellular
et al. undertook a randomized stabilized a phase with excellent between ABD1 and ABD2, and
processes from cell migration to
controlled trial of police reform thermoelectric properties. This this may be sufficient to allow
measures in India that was high-entropy material has low endocytosis. In the extensions cross-linking of aligned actin fila-
PHOTO: MECHIR/ALAMY STOCK PHOTO
designed to improve reporting thermal conductivity due to the of migrating cells called filipodia, ments. Phosphorylation further
rates of incidences of violence cation disordering but improved actin filaments form aligned weakens the ABD1–ABD2 inter-
against women in a context symmetry that helps to maintain bundles, but in the endocytic action, which may be required
where such violence is wide- good electrical properties. The machinery, they form branched to cross-link meshed networks
spread and where getting the material was used in several networks. Schwebach et al. of actin filaments. A structure of
police to take action has been devices that showed a high ther- suggest that the ability of plastin ABD1 bound to F actin supports
historically difficult (see the moelectric efficiency. —BG to remodel actin into a range of this model of a tunable interac-
Perspective by Blair and Jassal). Science, abq5815, this issue p. 208 structures comes from a tunable tion between ABD1 and ABD2

R E S EARC H | I N O T H E R J O U R NA L S
CHROMATIN
Histone history rewound
T
wo meters of DNA are packed within the nucleus of every Sebé-Pedrós et al. shed new light on histones by reconstructing
cell in our bodies. The DNA is first wrapped around nucleo- the evolutionary history of histone marks and their modify-
somes consisting of eight histone proteins, and together ing enzymes and regulators. Histone modifications do occur
these form chromatin. Nucleosomes also contain spac- in Archaea, but they are rare. Although the proteins that mark
ing information, and the tails of histones carry chemical histones have a pre-eukaryotic origin, the proteins that read
modifications including methylation, acetylation, and ubiquity- marks and chaperones are eukaryotic innovations. Further
lation. These modifications, or marks, instruct the cells about chromatin evolution has been characterized by diversification
what genes should be expressed and what genes should be kept among reader proteins, including by “capture” and transfer by
silent. Myriad proteins have evolved to write, read, and erase transposable elements and viruses among eukaryotes. —DJ
the histone marks, in concert with remodelers and chaperones. Nat. Ecol. Evol. 10.1038/s41559-022-01771-6 (2022).
Reconstructing the evolution of histone proteins reveals that they have a dual origin.
that allows plastins to modulate that this technique can be themselves at about $85,000, offers an efficient mechanism
the strength of actin cross- used as a source of heralded the assistance program provided that enables rapid cell expansion.
linking. —VV single photons in quantum $16 of value for every dollar spent —SMH
Nat. Struct. Mol. Biol. 29, 519 (2022). networks.—ISO by the government. —BW eLife 11, e75945 (2022).
Opt. Lett. 47, 2830 (2022). Q. J. Econ. 10.1093/qje/
qjac017 (2022).
QUANTUM OPTICS
ELECTRON MICROSCOPY
Designed for separation SOCIAL POLICY
PLANT BIOLOGY Nanoscale vibrational
Thin film lithium niobate is a
versatile platform across a
Welfare payments, Rapid cell expansion mapping
range of classical and quantum crime-reduction payback When root cells grow, they often Complex microscopic structures
optoelectronic applications. Removing low-income young need to extend rapidly. This is have been developed to modulate
One requirement for quan- adults with disabilities from a done by expanding the vacuole, thermal transport for a variety of
tum optics is the need for an US cash assistance program which means that the volume applications. However, vibrational
efficient source of heralded when they turned 18 increased of the cytosol does not need to mapping for understanding this
single photons. However, the number of criminal charges expand at the same rate. Dunser transport has spatial resolution
the spontaneous parametric against them over the next 20 et al. developed a small molecule limits. Gadre et al. show how to
down-conversion process that years by 20%. Deshpande and that inhibits vacuole membrane, circumvent this issue by using
occurs in lithium niobate typi- Mueller-Smith show that as a or tonoplast, expansion. The electron energy loss spectros-
cally results in the emission of result of those charges, primarily authors used this small molecule copy for phonon mapping of
entangled photon pairs, and for crimes to generate income, to show that endocytosis-derived silicon-germanium quantum dots.
untangling them compromises the annual likelihood of incarcera- vesicles contribute to the rapid The mapping allowed the authors
the purity of the single photons. tion in these subjects increased expansion of the tonoplast during to get a clear picture of how
Xin et al. demonstrate that by by 60%. The per-removal costs root cell expansion. Endocytosis atomic structure and interfaces
IMAGE: V. ALTOUNIAN/SCIENCE
engineering the dispersion of courts, police, and incar- is required for plasma mem- affect phonon dynamics and ther-
and phase-matching condi- ceration alone, about $41,000, brane and cell wall remodeling. mal transport. This strategy can
tions of the lithium niobate nearly eliminated the savings A potential change in membrane be used to better understand and
thin film, they could produce from removing these young recycling after endocytosis, as develop thermoelectrics, thermal
photon pairs that are spectrally adults from the program (about well as increased secretion of new barriers, phase-change memory,
separable. The high purity of $49,000). When also factoring in membrane during plasma mem- and other materials. —BG
the separated photons suggests the per-victim costs of the crimes brane and cell wall biosynthesis, Nature 606, 292 (2022).

RES EARCH
◥ epithelial-specific loss of HIF1a (Hif1aEKO)

RESEARCH ARTICLE SUMMARY phenocopied the wound-healing defect of
GFP-KI and Il17rcEKO mice. However, in con-
IMMUNOMETABOLISM trast to GFP-KI wounds, recombinant IL-17A
was unable to augment repair in Hif1aEKO mice,
Interleukin-17 governs hypoxic adaptation underscoring the necessity of this transcrip-
tional effector mediating re-epithelialization
of injured epithelium downstream of IL-17A signaling.
RORgt+ cell–deficient and control animals
Piotr Konieczny†, Yue Xing*†, Ikjot Sidhu, Ipsita Subudhi, Kody P. Mansfield, Brandon Hsieh, had comparable levels of wound hypoxia, indi-
Douglas E. Biancur, Samantha B. Larsen, Michael Cammer, Dongqing Li, Ning Xu Landén, cating that the failure of RORgt+ cell–deficient
Cynthia Loomis, Adriana Heguy, Anastasia N. Tikhonova, Aristotelis Tsirigos, Shruti Naik* animals to induce HIF1a was independent of
oxygen sensing. Instead, IL-17A was sufficient to
activate epithelial HIF1a both in vivo and in
INTRODUCTION: Injury compromises our epi- RESULTS: Using unbiased profiling of repair- epithelial organoids. IL-17RC signaling rap-
thelial barriers, leaving them vulnerable to associated lymphocytes, we uncovered the idly induced activation of both ERK and AKT,
external threats. Organismal survival thus im- enrichment of heterogeneous populations which augmented HIF1a protein and tran-
pinges on rapid barrier re-establishment after of type 17 lymphocytes. Homozygous Rorgt- script levels, but not stability, through mam-
damage. Immunocompromised individuals EGFP (GFI-KI) mice deficient in RORgt+ cells malian target of rapamycin (mTOR). Acute
and immunodeficient animals have profound exhibited a significant defect in epithelial hypoxia potently activated epithelial HIF1a
defects in epithelial repair. However, the precise migration and wound re-epithelialization. in vitro, and this response was further boosted
mechanisms and consequences of immune- After damage, we observed a rapid local ex- by IL-17A. By contrast, we uncovered a striking
epithelial cross-talk after injury remain incom- pansion of preexisting skin-resident RORgt+ loss of mTOR and HIF1a in chronic hypoxia.
pletely defined. cells that was sufficient to drive repair. Of Exogenous IL-17A rescued mTOR, and con-
the myriad wound RORgt+ populations, spa- sequently HIF1a, under conditions of chronic
RATIONALE: Epithelial repair requires cellular tial transcriptomics and functional studies hypoxia, indicating that IL-17A acts as a second
adaptation to the hypoxic wound microen- revealed that innate-like gd T cells directed signal to enable cell adaptation to low-oxygen
vironment through evolutionarily conserved tissue repair by localizing to the wound front environments.
hypoxia-inducible factors (HIFs). This response and boosting epithelial migration. These cells IL-17A induced a transcriptional and func-
has long been thought to depend upon epi- dominantly produced IL-17A/F in the wound tional program of glycolytic metabolism in
thelial cell-autonomous sensing of oxygen dep- microenvironment, which signaled directly to epithelia, which was dependent upon mTOR
rivation and responsiveness through HIFs. epithelia through IL-17RC to induce a migra- and HIF1a. The IL-17A–HIF1a-dependent pro-
However, whether and how supportive micro- tory program. gram of glycolysis fueled epithelial migration,
environmental signals from immune cells in- Comparative spatial transcriptomics of and pharmacological inhibition of glycolysis
tersect with ancient hypoxia responses during RORgt + cell-deficient and wild-type (WT) impaired wound re-epithelialization.
epithelial repair have yet to be explored. To wound edges revealed differences in HIF1a
address these questions, we were drawn to signaling. WT migrating epithelia had robust CONCLUSION: Our findings upend a long-
the skin, a primary epithelial barrier that is nuclear HIF1a activation that was notably held view that hypoxia is sufficient to cell-
surveilled by immune cells and has evolved missing from RORgt + cell– and epithelial autonomously induce HIF1a-mediated metabolic
sophisticated repair mechanisms. IL-17RC–deficient (Il17rcEKO) animals. Accordingly, remodeling. We illustrate that IL-17A supplied
by RORgt+ gd T cells is necessary for optimal
HIF1a activation in the wound-edge epithe-
IL-17A/F lium. The IL-17A–HIF1a axis directed the
Epidermis
Wound’s edge metabolic rewiring of damaged epithelium

IL-17RA IL-17RC toward a program of glycolysis to fuel migration.
Epithelial HIF1a and glycolysis are drivers of tumor pro-
cell P P gression and metastasis, raising the possibility
P AKT that IL-17A or other immune-derived signals
ERK could drive these pathways in cancer. IL-17A is
P Chronic hypoxia also central to the pathology of many auto-
Local IL-17A/F mTOR
immune conditions, including psoriasis and
proliferation
HIF-1α inflammatory bowel disease. Thus, the IL-17A–
HIF-1α HIF1a axis unveiled by our study may provide
Dermis
Hif1a therapeutic opportunities for a range of epithe-

RORγt+ γδ T cells
HIF-1α
Re-epithelialization
lial inflammatory and metastatic diseases.
▪
HRE Glycolysis
X genes The list of author affiliations is available in the full article online.
Systemic circulation *Corresponding author. Email: shruti.naik@nylangone.org
(S.N.); yue.xing@nyulangone.org (Y.X.)
†These authors contributed equally to this work.
IL-17 drives epithelial HIF1a to fuel wound repair through glycolysis. IL-17A/F supplied by expanding
Cite this article as P. Konieczny et al., Science 377,
skin-resident RORgt+ gd T cells is necessary for optimal HIF1a activation in the wound-edge epithelium, eabg9302 (2022). DOI: 10.1126/science.abg9302
even in the presence of hypoxia. IL-17A signaling through the IL-17RC receptor rapidly induces ERK/AKT/
mTOR, which augments HIF1a transcripts and protein. The IL-17–HIF1a axis directs a transcriptional READ THE FULL ARTICLE AT
and functional program of glycolysis to promote migration. https://doi.org/10.1126/science.abg9302
Konieczny et al., Science 377, 170 (2022) 8 July 2022 1 of 1

RES EARCH | R E S E A R C H A R T I C L E
A Control Day 3 Day 5 B Percent

Th/Tc 1 (767) Ifng
Type 1
2.5
Th/Tc 2 (286) 5.0
Th/Tc 3 (352) Tnfa 7.5
γδT/MAIT 1 (138) Il4 10.0
Type 2
γδT/MAIT 2 (476)
Il13 12.5
NK 1 (428)
NK 2/NKT (429) Il5
Average
Treg Type 17
ILCs (242) Il17a 1.2
Treg 1 (595)
UMAP-2
Il17f 0.9
Treg 2 (502)
Il22 0.6
Treg 3 (260)
0.3
UMAP-1 Treg 4 (431) Il10
Ctrl D3 D5
C Control skin Wound edge
600
0.0233 E D3 splinted wound
Human K14 Integrin α5 EdU DAPI
RORγt+ cells per mm2

1600 0.0006
K14 RORγt DAPI
Migrating tongue (μm)

400 WT
1200
0.0193
800
200
400
0 w.b. 0
Ctrl D1 D7
WT GFP-KI
D Control skin Wound edge 600
0.0101 D3 unsplinted wound
0.0040 GFP-KI 1600

Murine
GFP+ cells per mm2
400 <0.0001
K14 Rorc-GFP DAPI
200 1200
0.0034
100 800
ns
80
60 400
40 w.b.
20 0
WT GFP-KI
0
Ctrl D1 D3 D5 D7
F Percent
G K14 Rorc-GFP EdU DAPI H
1200

5 Wound
Mki67 A1 ns
10
Wound edge
15
Cdk1 FTY720 FTY720
800
Average A1
1.00
0.75
Top2a Day 3
0.50 Analysis 400
-D1 D0 D1 Veh FTY
Ctrl Wd 0.25
Fig. 1. Skin-resident RORgt cells direct wound re-epithelialization. (A) UMAP beginning and end of migrating tongue, respectively. White, EdU; blue, DAPI
visualization of individual samples (control, n = 1443 cells; D3, n = 1997 cells; nuclei; w.b., wound bed (n ≥ 7, N = 3). (F) Wound lymphocytes have a higher
D5, n = 1486 cells). (B) Dot plot of cytokine expression in control and wounded frequency (percentage) and expression of proliferation genes (Mki67, Cdk1, and
lymphocytes. Shown is the frequency of cells expressing gene (percentage) Top2a) than control skin. (G) Wound RORgt cells actively proliferate. Shown is an
and average expression per cluster (average). RORgt+ cells (yellow) are enriched immunofluorescence image of D3 wounded Rorc-EGFPTg mice after a 24-hour
at the edge of human (C) (n = 3, N = 1) and murine (D) (n ≥ 4, N = 3) acute EdU pulse. White, EdU; blue, DAPI nuclei; green, RORgt+ cells; red, keratin
wounds. Magenta, keratin (K)14; cyan, DAPI nuclei. White arrows mark RORgt (K)14 (n = 3, N = 3). (H) FTY720 treatment does not alter wound re-epithelialization
cells. (E) Impaired re-epithelialization in RORgt-deficient (GFP-KI) compared (n ≥ 7, N = 3). In (C), (D), and (G), white dashed lines denote dermo-epidermal
with WT animals. Shown are images and quantifications of integrin a5+ (green), borders and yellow boxes define magnified areas. Scale bars in (C), (D), and
K14+ (red) migrating epidermal tongue from full-thickness silicone-splinted (top) (E), 100 mm; scale bar in (G), 50 mm. Significance was determined using a
and unsplinted (bottom) D3 wounds. The dashed lines and arrows mark the two-tailed t test and a 95% confidence interval.
into the wound bed. By contrast, mice heal dent RORgt lymphocytes locally expand to RORgt cells or epithelial migration (Fig. 1H
by re-epithelialization and myofibroblast- fuel repair. Indeed, wound RORgt+ cells had and fig. S3G). Thus, the expansion of pre-
mediated dermal contraction (17). We there- heightened expression of cell-cycle genes existing skin RORgt+ cells is sufficient to drive
fore secured murine wounds with silicone (Mki67, Cdk1, and Top2a) and 5-ethynyl-2′- re-epithelialization.
splints to eliminate dermal contraction. Micro- deoxyuridine (EdU) incorporation relative to
scopic visualization revealed a specific impair- unwounded skin and to their systemic lymph IL-17RC–mediated gd T cell–epithelial cross-talk
ment of the keratin 14+ (K14+) integrin a5+ node counterparts (Fig. 1, F and G, and fig. controls re-epithelization
migrating epithelial tongue, but not epithelial S3E). To determine whether skin RORgt+ lym- We mapped the dynamics of skin RORgt sub-
proliferation, in GFP-KI splinted wounds (Fig. 1E phocytes were sufficient to instigate repair, sets over the course of repair using multi-
and fig. S3D). Similarly, epithelial migration was we treated mice with FTY720, a sphingosine-1 parameter flow cytometry. Innate-like gd
markedly compromised in unsplinted GFP-KI phosphate receptor agonist that inhibits lym- T cells, MAIT (MR1 tetramer+), and invariant
wounds relative to controls (Fig. 1E). phocyte egress into blood by “trapping” them NKT (iNKT) cells (CD1d tetramer+) expanded
Given the rapid enrichment of RORgt lym- within lymph nodes (fig. S3F) (18). FTY720 first at D1 after wounding (fig. S4, A to C).
phocytes in wounds, we postulated that resi- blockade did not alter the frequency of wound Adaptive TH 17, TC17, and RORgt+ Treg cells
Konieczny et al., Science 377, eabg9302 (2022) 8 July 2022 2 of 12

subsequently increased between D3 and D5. A C

Control Skin D3 Wounded Skin
is
Consistent with our CITE-seq analysis, the
is
derm
derm
5-H lium ge
4-W elium tal
proportions of RORgt+ type 3 innate lymph-
is dge
is
le
is
derm
epit und e
ollic
ular
deromund e
epit nd d
lary
oid cells (ILC3s) were unchanged by injury,
etic
air f
ypo
he
apil
u
h
o
suggesting that in the presence of adaptive
o
0-W
3-W
2-R
6-H
1-P
immunity, ILC3s may not play a dominant
role in repair (figs. S2D and S4, A to C). ILCs
Enrichment Score -log10 (P value)

8
effector lymphocyte clusters

To pinpoint which RORgt subset(s) was NK 2/NKT
positioned at the re-epithelializing wound NK 1
6
front, we turned to spatial transcriptomics (ST), B Control skin cluster annotation
γδT/MAIT 2
CITE-seq
a technology that enables spatially resolved 0-Reticular dermis 4
γδT/MAIT 1
transcriptional profiling of tissues at 50-mm 1-Hypodermis 1
2-Panniculus carnosus
resolution (8) (fig. S5, A to C, and table S4). 3-Epidermis/
2 Th/Tc 3
Papillary dermis
PCA and UMAP projections distinguished 4-Hypodermis 2 Th/Tc 2
seven clusters (tissue regions) in unwounded 5-Hair follicle
Th/Tc 1
UMAP-2
6-Epidermis
skin and six clusters in wounded skin (Fig. 2, 7-Fascia Day 3 wounded ST clusters
A and B). The identified clusters were con- UMAP-1
sistent with histological regions and genes D E F
that typified tissue architecture (Fig. 2, A Day 3 wounded skin cluster annotation TcrdCreER;Rorcfl/fl Mr1–/– Cd4Cre;Rorcfl/fl
and B, and fig. S5, D and E). Wound-edge epi- 1200 1200 1200
0.0093

ns
thelium was discernable by Krt14 and Krt17 0-Wound edge dermis
1000 1000
1-Papillary dermis 1000
expression (fig. S6A). Nevertheless, to confirm 2-Reticular dermis
ns
the identity of epidermal wound edge clusters 3-Wound distal epithelium 800 800 800
4-Wound edge epithelium
and validate our ST data, we performed multi- 5-Hair follicle
UMAP-2
600 600 600

modal intersectional analysis (MIA) (8), a 6-Hypodermis
method that can infer cell-type enrichment 400 400 400
UMAP-1
in tissue regions. Integrating an independent
l
c fl/f
/–
re
et
R re ER
l
c fl/f
r1 –
re
r1 H
l
c fl/f
or
C
or
M
Percentage
R
scRNA-seq dataset (19), which defined wounded
C
G
M
or
R
of total cells
epithelium as growth arrest cells and wound γδT/MAIT 1
0 H
10
distal epithelium as Col17a1+ cells, with our ST 20 Intradermal
1000 0.0369

Wound
data revealed a significant degree of overlap γδT/MAIT 2 rmIL-17A or PBS
30
between Col17a1+ cells and wound distal epi- ILCs Day 3
800
Average
thelial cluster 3 (P = 2.3 × 10−16) and growth Th/Tc 1 expression Migrating tongue
Th/Tc 2 per cluster D0 D1 D2 analysis
arrest cells and wound-edge epithelial cluster 4 Th/Tc 3 GFP-KI
600
(P = 5.4 × 10−35) (fig. S6, B and C). NK 1 6

NK 2/NKT 4
We next leveraged MIA to resolve the loca- 400
2 PBS rmIL-17A
7a
2
7f
tion of wound lymphocyte populations identi-

Il2
Il1
Il1
0
fied by our CITE-seq data (Fig. 1A). MIA of the
top 300 up-regulated genes (≥0.25 log FC)
I Il17rc Il17ra J 0.0015
1600
Control
from each effector lymphocyte CITE-seq cluster

and each ST cluster simultaneously located all 1200
of our lymphocyte populations in wounded
skin and revealed a specific enrichment of gd T/
Wound
800
MAIT cluster 1 (P = 7.9 × 10−6) and cluster 2
(P = 1.9 × 10−5) within wound-edge epithelial
400
cluster 4 (Fig. 2C and fig. S6D). 0 0.5 1.0 1.5 WT Il17rcEKO
0 0.4 0.8 1.6
Both gd T cells and MAIT cells have been
implicated in wound repair (20). However,
systematic comparisons between these and Fig. 2. RORgt+ gdT cells drive re-epithelialization through epidermal IL-17RC. (A) Hematoxylin and
other RORgt+ cell subsets in the context of eosin–stained tissue image (top) and ST plot of microarray spots (bottom) of unwounded and D3
re-epithelialization are lacking. We therefore wounded skin (unwounded, n = 1; wounded, n = 2). Scale bars, 200 mm. (B) UMAP plot and cluster
analyzed mice specifically lacking RORgt+ gd annotation of ST spots based on marker genes in fig. S5. (C) gd T/MAIT clusters are enriched in wound
T cells (TcrdCreER;Rorcfl/fl), MAIT cells (Mr1–/–), edge epithelium. P-value score table from MIA of D3 wounded ST clusters and CITE-seq effector
and TH17/TC17/MAIT/iNKT cells (Cd4Cre;Rorcfl/fl) lymphocyte clusters. Integrin a5+, K14+ migrating epidermal tongue length in unsplinted D3 wounds
(fig. S6E). After tamoxifen-induced depletion, from TcrdCreER;Rorcfl/fl and Rorcfl/fl (D) (n ≥ 7, N = 3). Mr1–/– and Mr1Het (E) (n ≥ 4, N = 3). Cd4Cre/–;
TcrdCreER;Rorcfl/fl mice exhibited a shorter Rorcfl/fl and Cd4Cre/– (F) (n = 5, N = 3). (G) gd T/MAIT cell clusters dominantly express Il17a and
migrating tongue than control animals, in- Il17f. Dot plot of genes from D3 wounded skin CITE-seq analysis. Frequency of cells expressing stated
dicating a nonredundant role for RORgt+ gd gene (percent) and average expression per cluster (average). (H) rmIL-17A administration augments the
T cells in wound re-epithelialization (Fig. re-epithelialization in GFP-KI mice. Experimental schematic (left) and quantification of D3 migrating
2D). Wounds from both Mr1–/– and Cd4Cre; tongue length of GFP-KI skin intradermally injected with rmIL-17A or PBS (right) (n = 4, N = 2). (I) Spatial
Rorcfl/fl mice had migrating tongues com- plots of Il17rc and Il17ra reveal an up-regulation of these receptors at the wound’s edge. (J) IL-17RC
parable to their respective controls, under- expression on epithelial cells drives re-epithelialization. Quantification of migrating tongue length from
scoring the unicity of RORgt+ gd T cells in splinted D3 wounds Krt14Cre; Il17rcfl/fl mice (n ≥ 5, N = 2). Significance was determined using a two-tailed
directing re-epithelialization (Fig. 2, E and F). t test and a 95% confidence interval.

RORgt+ gd T cells exert potent effector func- wound distal epithelium (fig. S8C). By con- associated cytokines [rmIL-1, rmIL-6, and
tion by producing the cytokines IL-17A and trast, WT and GFP-KI migrating tongue cells recombinant murine tumor necrosis factor a
IL-17F, which provide instructive signals to the had 1548 differentially expressed transcripts (rmTNFa)] did not activate HIF1a (fig. S10G).
surrounding parenchyma (14, 21). Of all the (adjusted P < 0.1) (fig. S8D). HIF1a signaling Epidermal growth factor also failed to augment
effector lymphocytes analyzed by CITE-seq, and associated glycolysis pathways were sig- HIF1a to the levels observed after rmIL-17A
gd T/MAIT cell clusters expressed the highest nificantly enriched in WT samples (fig. S8E). treatment, underscoring the specificity of
level of these cytokines at D3 after wounding, Consistent with the enrichment of Hif1a IL-17 in amplifying this pathway (fig. S10H).
hinting at their role in the RORgt+ gd T cell– transcripts, WT migrating epithelial tongue IL-17RA and IL-7RC were robustly expressed
epithelial cross-talk at the wound’s edge (Fig. robustly expressed nuclear HIF1a, which was in our epithelial organoids and wound-edge
2G). Accordingly, intradermal recombinant absent from GFP-KI wound fronts (Fig. 3E and integrin a5+ cells (fig. S10I) (14, 22). Epithelial-
murine IL-17A (rmIL-17A) administration re- fig. S9, A and B). GFP-KI mice did not com- specific loss of Il17rc abrogated the ability of
versed the re-epithelization defect of GFP-KI pensate for the loss of HIF1a by up-regulating rmIL-17A to activate HIF1a both in vivo and
mice (Fig. 2H). HIF2a (fig. S9D). Furthermore, Il17rcEKO mice in vitro, excluding a role for accessory cells in
IL-17A/F signaling is mediated by an IL-17 phenocopied the HIF1a defect observed in this cross-talk (Fig. 4, C and D, and fig. S10, J
receptor A (IL-17RA)–IL-17RC heterodimer, GFP-KI wounds (Fig. 3F). Nuclear HIF1a was and K). To delineate how epithelial IL-17RC
whereas IL-17F can also bind an IL-17RC/RC also robustly expressed in the migrating epi- signaling resulted in HIF1a activation, we re-
homodimeric complex (14, 22). ST plots showed thelial front of human wounds, suggesting a turned to our comparative GFP-KI and WT
increased Il17ra and Il17rc transcripts at the highly conserved role for this transcription wound transcriptomics (Fig. 3, A to C). mTOR
wound’s edge, with the most prominent ex- factor in re-epithelialization (fig. S9C). signaling, which is implicated in HIF1a reg-
pression being Il17rc (Fig. 2I). We therefore HIFs ubiquitously mediate cellular adapta- ulation, was salient as a key differentially
generated and wounded epithelial-specific tion to low-oxygen environments (23). There- expressed pathway (24, 25). We therefore
IL-17RC–deficient mice (Krt14Cre;Il17rcfl/fl la- fore, we measured hypoxia in WT and GFP-KI evaluated ribosomal protein S6S240/244 phos-
beled Il17rcEKO) (fig. S6F). Compared with wounds using both a fluorescent hypoxia phorylation, a major downstream substrate of
IL-17RC–sufficient mice, epithelial-specific loss marker, pimonidazole (PIM), and a fiber probe mTOR complex 1. rmIL-17A robustly increased
of Il17rc mirrored the re-epithelialization defect to directly measure tissue O2. WT and GFP-KI epithelial pS6S240/244 in WT but not Il17rcEKO
of GFP-KI and TcrdCreER;Rorcfl/fl mice (Fig. 2J). wounds had equivalent PIM signals and O2 mice and organoids (Fig. 4, E and F, and fig.
Thus, IL-17A/F supplied by RORgt+ gd T cells levels (~10 to 12 mmHg or 2% O2) (Fig. 3, G S10L). rmIL-17A–treated organoids also showed
directly engages epithelial IL-17RC to induce and H). Moreover, WT and GFP-KI epithelia enhanced activation of the upstream kinase
migration and re-epithelialization. expressed comparable levels of prolyl hydrox- phosphorylated S6 kinaseT389 (p-S6KT389) and
ylases (Egln1, Egln2, and Egln3) and Vhl phosphorylated mTORS2448 (p-mTORS2448)
HIF1a is a re-epithelialization factor controlled ubiquitin ligase (fig. S9D). Thus, wound-edge (Fig. 5, A and B). Inhibition of mTOR with
by RORgt+ cells epithelium failed to sustain HIF1a without rapamycin significantly reduced HIF1a protein
PCA and UMAP plots integrating wild-type RORgt+ cell–derived signals even in the pres- downstream of IL-17A (Fig. 5, A to C, and fig.
(WT) and GFP-KI tissue sections from D3 ence of tissue hypoxia and while having the S11, A and B). Additionally, Hif1aEKO and con-
wounds discerned 15 clusters with high re- necessary regulatory machinery. trol organoids had comparable sizes and mTOR
producibility between biological replicates We next generated and wounded epithelial- activation (p-S6S240/244) after rmIL-17A stimu-
(Fig. 3, A and B, and fig. S7, A and B). We specific HIF1a-deficient mice (Krt14Cre; Hif1afl/fl lation, confirming that mTOR is upstream of
focused on wound edge clusters 7 and 8 on labeled Hif1aEKO) (fig. S9E). Loss of epithe- HIF1a (fig. S11, C and D).
the basis of histology and expression of the lial HIF1a impaired the development of mi- In addition to augmenting HIF1a protein,
re-epithelialization factors Gjb2, Lamc2, Mmp9, grating tongue (Fig. 3I). Additionally, unlike rmIL-17A stimulation also induced Hif1a
Mmp13, and Stfa1 (fig. S7A). We further de- GFP-KI mice, rmIL-17A failed to rescue the re- gene expression in an mTOR-dependent man-
lineated cluster 7 as wound-edge basal epi- epithelialization defect of Hif1aEKO mice (Fig. ner (Fig. 5D). Increased Hif1a transcripts in
thelium (Krt5, Krt6, Krt14, and Krt17) and 2H and fig. S9F). Thus, epithelial HIF1a is nec- rmIL-17A–treated organoids were not caused
cluster 8 as wound-edge differentiated epithe- essary to respond to IL-17A signals and activate by enhanced mRNA stability, and HIF1a pro-
lium (Krt10, Lce1a1, and Lor). Pathway analy- a program of migration. tein was more rapidly degraded in the pres-
sis of WT and GFP-KI cluster 7 differentially ence of rmIL-17A (fig. S11, E and F). Thus,
expressed genes revealed enrichment of IL-17 IL-17RC signaling induces HIF1a through mTOR controls HIF1a transcription and pro-
signaling, skin/epidermal development, epi- extracellular signal–regulated kinase 1/2 tein levels, but not stability, downstream
dermal differentiation, epithelial cell/tissue (ERK 1/2)-AKT–mammalian target of IL-17A.
migration, and T cell activation pathways in of rapamycin (mTOR) To determine whether and how IL-17RC
WT (Fig. 3C and fig. S8A). HIF1 signaling, To test our hypothesis that gd T cell–derived proximal signaling links to mTOR, we next
which is typically associated with tissue hy- IL-17 serves as a second signal to activate measured early (≤1 hour) mTOR activation
poxia, also emerged as a differentially expressed HIF1a, we intradermally injected rmIL-17A, after rmIL-17A stimulation and found enrich-
pathway (Fig. 3C) (10), with Hif1a transcripts rmIL-17F, or rmIL-22 into unwounded skin. ment of p-S6KT389 and p-S6S240/244 (fig. S11G).
higher in WT compared with GFP-KI wound Administration of rmIL-17A and, to a lesser Working our way upstream, we detected ac-
edges (Fig. 3D). degree, rmIL-17F was sufficient to induce epi- tivation of p-ERK1/2 T202/Y204 , a canonical
To validate this finding, we analyzed the dermal HIF1a activation and hyperplasia (Fig. mediator of IL-17RC signaling, and protein
transcriptomes of integrin a5 and integrin 4A and fig. S10, A and B). IL-22 marginally kinase B (p-AKT S473), an upstream mTOR
a6 double-positive wound-edge epithelium increased epidermal thickness but did not kinase, after IL-17A stimulation (Fig. 5E). In-
(migrating tongue) and integrin a6 + -only up-regulate HIF1a (fig. S10, A and B). Skin hibition of either ERK1/2 (U0126) or AKT
wound distal epithelium, from WT and GFP-KI epithelial organoids cultured for 5 days in the (MK-2206) after long- and short-term rmIL-17A
wounds (fig. S8B). We observed very modest presence of rmIL-17A were significantly larger treatment (5 days and 30 min, respectively),
differences (62 differentially expressed genes, than controls and robustly expressed HIF1a diminished p-S6KT389, p-S6S240/244, and HIF1a
adjusted P < 0.1) between WT and GFP-KI (Fig. 4B and fig. S10, C to F). Other damage- expression (Fig. 5E and fig. S11H). However,

A B C KEGG Pathways enriched in WT wound’s edge

Cluster annotation ST cluster
WT 0- Reticular dermis
1- Wound bed –Log10 (P value)
2- Hypodermis 0 2 4 6 8
3- Hair follicle 1
Adherens junction
4- Distal dermis undefined 1
5- Distal dermis undefined 2 EGFR tyrosine kinase inhibitor resistance
6- Hair follicle 2
mTOR signaling pathway
7- Wound edge basal
GFP-KI 8- Wound edge differentiated IL-17 signaling pathway
9- Fascia
10- Panniculus carnosus 1 MAPK signaling pathway
11- Distal epidermis 1
UMAP-2
12- Distal epidermis 2
HIF-1 signaling pathway
13- Panniculus carnosus 2
14- Hair follicle 3 IL-17 signaling pathway: HIF-1 signaling pathway:
UMAP-1 Il17rc, S100a9, Mapk3, Ldha, Egfr, Stat3, Hk2,
Cebpb, Nfkb1, Jun Eif4e, Cdkn1b
D Hif1a E Migrating tongue
WT WT GFP-KI
80
0.0004 F 150
cells
cells
0.0011
K14 HIF1α DAPI
+
60
No. nuclear HIF1

100
No. nuclear HIF1

40
GFP-KI
50
20
0 0
WT GFP-KI WT Il17rcEKO
0 0.5 1.0 1.5 2.0 2.5 Low High

0.0039
H
G 10 ns 50 0.0120 Ctrl skin
PMI Staining intensity (A.U)

Wound
PIM Neg WT GFP-KI
8
* * * High
40
Pimonidazole
O 2 (mmHg)
6 30 ns
4 20
Low 2 10
0 0
WT GFP-KI Ctrl WT GFP-KI
I 1200 <0.0001
WT Hif1aEKO
K14 Integrin α5 DAPI

1000
800
600
w.b. w.b.
400
WT Hif1aEKO
Fig. 3. RORgt+ cells control wound edge epithelial HIF1a. ST spots (A) and PIM staining in D3 WT and GFP-KI wounds (n = 4, N = 2). (H) Fiber probe
UMAP plot (B) of annotated clusters (marker genes in fig. S7) from WT and measurement of unwounded and D3 wounded tissue oxygen (n ≥ 3, N = 2).
GFP-KI D3 wounded skin samples (WT, n = 2; GFP-KI, n = 3). (C) KEGG pathways (I) Impaired re-epithelialization in epithelial HIF1a-deficient animals (Hif1aEKO)
enriched in WT versus GFP-KI cluster 7 spots. (D) Spatial plots of Hif1a compared with WT. Confocal images and corresponding quantifications of
expression in D3 WT and GFP-KI wounds. (E) WT migrating tongues have higher integrin a5+ (yellow), K14+ (red) migrating epidermal tongue from splinted
HIF1a expression than GFP-KI migrating tongues. Red, top, K14; yellow, D3 wounds. The dashed line and arrows mark the beginning and end of migrating
top, HIF1a (pseudocolor fire, bottom); cyan, top, DAPI nuclei. Quantification of tongue, respectively. Cyan, DAPI nuclei; w.b., wound bed (n ≥ 5, N = 3). Scale
HIF1a in integrin a5+ cells from GFP-KI and WT D3 wounds (n = 6, N = 3). bars in (E), 50 mm; scale bars in (G) and (I), 100 mm. For (E), (G), and (I),
(F) Quantification of HIF1a in integrin a5+cells from WT and Krt14Cre; Il17rcfl/fl D3 two-tailed t test at a 95% confidence interval was used. For (H), one-way ANOVA
wounds (n = 5, N = 2). (G) Pseudocolor fire view images and quantification of multiple-comparisons test was used.
blocking ERK also resulted in a compensatory Given the prominence of hypoxia at the and in agreement with our in vivo wounding
increase in p-AKTS473. Dual inhibition of ERK wound edge, we next examined the IL-17A– data, p-S6S240/244 and HIF1a levels were mark-
and AKT ablated the mTOR-HIF1a response mTOR–HIF1a axis in organoids cultured under edly reduced by long-term hypoxic exposure
downstream of rmIL-17A stimulation (Fig. 5E acute (5 hours) or chronic (5 days) hypoxic and rescued by the presence of rmIL-17A (Fig.
and fig. S11H). Thus, IL-17RC–proximal AKT and conditions (2% O2). Acute hypoxia was suf- 5F). Rapamycin treatment abrogated the ability
ERK signaling directly activates mTOR and ficient to stabilize HIF1a, and IL-17A further of IL-17A to augment HIF1a in chronic hypoxic
consequently induces HIF1a. augmented this response (Fig. 5F). By contrast, cultures (Fig. 5G and fig. S11, I and J). Thus,

A B
PBS K14 HIF1α Phalloidin DAPI
* 100
<0.0001
6
cells
Ctrl rmIL-17A
K14 HIF1α DAPI
0.0355
Nuclear intensity
80
+
4
No. nuclear HIF1

60
rmIL-17A 40
2
20
0 0
PBS rmIL-17A Ctrl rmIL-17A
Low High
C D Organoids
E F rmIL-17A
WT PBS WT
* WT
Il17rcEKO PBS rmIL-17A
p-S6(S240/244) DAPI
p-S6(S240/244) DAPI
100 0.0030 4
cells
K14 HIF1α DAPI
0.0013 0.0015
Nuclear intensity
80
+
3
No. nuclear HIF1
60
Il17rcEKO 2 rmIL-17A Il17rcEKO
* 40 ns
1
20
0 0
rmIL-17A WT Il17rcEKO
Fig. 4. IL-17A induces HIF1a through epidermal IL-17RC. (A) Intradermal as in (A). (D) IL-17A–induced HIF1a in organoids is dependent on epidermal
injection of rmIL-17A in unwounded skin induces epidermal hyperplasia IL-17RC (n = 4, N = 2). (E) Epidermal response to rmIL-17A is accompanied by
and HIF1a. Magenta, K14; yellow, HIF1a; cyan, DAPI nuclei (n = 5, N = 3). increased phosphorylation of S6 (p-S6Ser240/244, green) (n = 3, N = 2).
(B) Epidermal organoids cultured for 6 days at 21% O2, ± rmIL-17A stimulation (F) rmIL-17A–induced p-S6Ser240/244 (green) requires epidermal IL-17RC (n = 3,
for 5 days (see experimental schematic in fig. S10C) robustly up-regulate N = 2). In (E) and (F), blue, DAPI nuclei. In (A) and (C), white asterisks label
HIF1a. Quantification of HIF1a staining normalized to controls. Red, K14; green, autofluorescence. In (A) to (C) and (E) and (F), white dashed lines demarcate
HIF1a; white, phalloidin F actin; blue, DAPI nuclei (n = 3, N = 3). (C) rmIL-17A dermo-epidermal junction or organoid boundaries. Scale bars, 50 mm. For
induced nuclear HIF1a in WT mice but not epidermal IL-17RC–deficient (A) to (C), two-tailed t test at a 95% confidence interval was used. For (D),
(Il17rcEKO) mice (n = 3, N = 2). Staining and color scheme are the same two-way ANOVA multiple-comparisons test was used.
IL-17A directly and potently induces mTOR to and E, and fig. S12, E and F). GKP-KI, Il17rcEKO, Discussion
up-regulate HIF1a under both normoxic and and Hif1aEKO wound-edge epithelia expressed Collectively, our results illustrate that IL-17A
hypoxic conditions (Fig. 5H). significantly lower glucose transporter 1 levels supplied by RORgt+ gd T cells is necessary for
than their respective controls (Fig. 6, F and G, optimal HIF1a activation in the wound-edge
IL-17A–dependent metabolic reprogramming is and fig. S12G). rmIL-17A administration res- epithelium. HIFs are an ancient class of tran-
vital for re-epithelialization cued glucose transporter 1 levels in GKP-KI scription factors that are present in all meta-
Migration, a cardinal feature of re-epithelialization, wounds (fig. S12H). Thus, the IL-17A–HIF1a zoan cells and are stabilized in response to
requires cells to expend energy. However, exactly axis induces a program of glycolytic metab- low microenvironmental oxygen (27). Previ-
how migrating epithelial cells meet their energy olism in epithelial cells. ous studies attributed the increased expres-
demands in a hypoxic wound microenviron- Our findings raised the possibility that the sion of HIF1a in injured skin and intestinal
ment is unclear (Fig. 3, G and H). Glycolysis and IL-17A–HIF1a–dependent program of glucose epithelium to tissue hypoxia (28, 29). Thus,
oxidative phosphorylation are the two major metabolism fuels epithelial migration. We our discovery that the chronic hypoxic envi-
cellular energy–producing pathways (26). Of topically treated mice with 2-deoxy-D-glucose ronment of the wound edge is insufficient to
these, HIF1a is known to control a transcrip- (2-DG), a glucose analog that inhibits glycolysis. activate HIF1a prompts investigation into IL-17,
tional program of glycolysis (10). Accordingly, 2-DG–treated mice had significantly shorter and other secondary activators of HIF1a, in
the WT wound front expressed higher levels integrin a5+ migrating tongues than their con- nonhealing wounds and conditions involving
of glycolysis-associated genes [hexokinase 2 trol counterparts (fig. S12I). Blocking glycol- chronic hypoxia.
(hk2), phosphoglycerate kinase 1 (Pgk1), and ysis did not alter epithelial cell proliferation, Consistent with published reports, we found
glucose transporter protein 1 (Slc2a1)] than indicating that glycolysis is dispensable for that acute hypoxia robustly induces HIF1a
did that of GFP-KI (Fig. 6A). expanding the epithelial cell pool at the wound (25). By contrast, exposure to chronic hypoxia
rmIL-17A treatment significantly increased front (fig. S12J). 2-DG treatment also led to a (reminiscent of the wound microenvironment)
glucose transporter 1 (Slc2a1) and glycolytic reduction in wound-edge type 17 cells (fig. S12K). leads to a reduction in mTOR and consequently
enzyme expression in epithelial organoids We therefore turned to an in vitro epithelial HIF1a. The reasons for down-modulation of
(Fig. 6B and fig. S12, A and B). rmIL-17A–treated scratch assay to control for the immune effects mTOR in chronic hypoxia are unclear and
organoids had heightened glycolysis activity of topical 2-DG treatment. 2-DG or HIF1a in- warrant further investigation. Nevertheless,
and glycolytic capacity and secreted higher hibitor (BAY872243) hindered the ability of IL-17A acts as a “second signal” to activate
levels of lactate, an end product of glycolysis, human epithelial cells to migrate and seal the mTOR in this context through ERK1/2 and
compared with controls (Fig. 6C and fig. S12, in vitro “wound” both in the presence and ab- AKT kinases. We found that IL-17A–induced
C and D). Consistent with a role for mTOR sence of recombinant human IL-17A (rhIL-17A) mTOR controls HIF1a gene transcription and
in regulating HIF1a, rapamycin-treated or (Fig. 6H). Thus, IL-17A augments glycolysis, an protein levels but not stability. Thus, our results
Hif1aEKO organoids expressed lower levels of essential bioenergetic pathway powering epi- suggest that adaptation to hypoxia in com-
glycolysis enzymes than did controls (Fig. 6, D thelial cell migration, to expedite healing. plex tissue microenvironments requires second

A Epidermal cell isolation E Ctrl rmIL-17A

AKT/ERK1/2 inhibitor MK-2206 - + - + - + - + kDa
rm
rmIL-17A Day 5.5 for 3 hours U0126 - - + + - - + +
44
p-ERK1/2(T202/Y204) 42
Day 0 Day 1 Rap inhibitor Day 6 0.53 0.61 0.55 0.55 2.19 1.89 0.65 0.61
for 12 hours
ERK1/2 44
B C 4 <0.0001 42
<0.0001 1.34 1.26 1.20 1.24 1.30 1.35 1.38 1.34
Nuclear intensity
Ctrl rmIL-17A
3 p-AKT(S473)
Veh Rapa Veh Rapa kDa 60
0.52 0.33 1.51 0.50 0.92 0.41 1.68 0.52
p-mTOR(S2448) 180 2 ns AKT(Pan) 60
0.35 0.24 1.03 0.63
1 0.79 0.75 0.75 0.81 0.89 0.83 0.97 1.03
mTOR 180
Veh p-S6K(T389)
0.73 0.46 0.95 0.76 Rapa (12hrs) 70
0
p-S6K(T389) 70 Ctrl rmIL-17A 0.86 0.70 1.65 1.13 1.59 1.18 1.42 0.72
0.87 0.51 1.24 0.42 D 4

<0.0001
<0.0001
S6K 70
Hif1a mRNA expression

S6K 70 0.67 0.56 0.74 0.66 0.75 0.78 1.03 1.11
1.15 0.89 1.14 1.24 3 p-S6(S240/244) 35
p-S6(S240/244) 35
0.95 0.21 0.75 0.24 2.00 1.28 1.73 0.57
0.42 0.07 1.56 0.12 2
ns S6 35
S6 35
1 1.13 0.79 1.08 0.88 1.53 1.49 1.61 1.33
1.48 1.29 1.69 1.42
Veh
β-actin Rapa (12hrs) HIF1α 120
40 0
Ctrl rmIL-17A 0.32 0.28 0.39 0.29 0.96 0.49 1.12 0.51
β-actin 40
F 21%O2 2%O2
G 48 hours 5 days
Rapa - + - + - + - + IL-17A
Time 0 5hrs 5d 5hrs 5d rmIL-17A - - + + - - + + kDa H
rmIL-17A - + + - + - + kDa HIF1α IL-17RC
120
p-S6(S240/244) 35
0.78 0.30 0.82 0.60 0.17 0.02 0.33 0.20
1.89 2.20 2.69 0.74 1.27 0.60 3.37
β-actin 40
P P
S6 35 ERK AKT P
P
2.5 0.0037
1.78 1.81 2.50 1.87 2.10 0.88 1.81 0.0082
Nuclear intensity
β-actin 40 2.0
Chronic Hypoxia mTORC1
1.5
ns HIF1α
HIF1α 120
1.0
0.40 0.43 0.84 2.13 2.46 1.01 1.93
0.5
β-actin 40 Veh
Rapa (12hrs)
0.0 Hif1a
Ctrl rmIL-17A
Fig. 5. ERK/AKT signaling downstream of IL-17RC controls mTOR and HIF1a. p-S6 S240/244, and HIF1a (N = 2). (F) IL-17A stimulates p-S6 S240/244 and
(A) Experimental schematic of inhibitor treatment in control and rmIL-17A– HIF1a in organoids cultured under normoxic or hypoxic (2% O2 ) conditions
stimulated organoids. (B) rmIL-17A–treated organoids (D5) have increased (N = 3). (G) Rapamycin abrogates IL-17A–mediated HIF1a expression
pmTORS2448, p-S6KT389, and p-S6S240/244 that is diminished after 12 hours of in hypoxia (n = 4, N = 3). (H) Schematic of proposed mechanism illustrating
rapamycin (Rapa) treatment. Vehicle (Veh), N = 2. (C) Rapamycin abrogates that AKT and ERK activation proximal of IL-17RC induces mTOR and HIFa.
rmIL-17A–mediated HIF1a expression in normoxia (21% O2). Quantification Chronic hypoxia inhibits mTOR and consequently HIF1a. Significance was
of HIF1a staining normalized to controls (n ≥ 5, N = 4). (D) IL-17A up-regulates determined using a two-way ANOVA multiple-comparisons test. In (B),
Hif1a transcripts through mTOR (n = 5, N = 2). (E) Inhibiting AKT (MK-2206) (E), (F), and (G), protein quantifications are relative to the presented internal
and/or ERK (U0126) for 3 hours blocks rmIL-17A–induced p-S6KT389, b-actin controls.
signals from accessory cells in addition to cell- conditions, including psoriasis, inflammatory Tcrdtm1.1(cre/ERT2)Zhu/J, B6(Cg)-Rorctm3Litt/J,
autonomous oxygen monitoring. bowel disease, and spondyloarthropathies (14). B6.129-Hif1atm3Rsjo/J. Mr1 tm1(KOMP)Vlcg mice
IL-17A directed the metabolic rewiring of In addition to glycolysis, other HIF1a tran- were generated from the KOMP ES cell line
wound-edge epithelium toward a program of scriptional targets may fuel IL-17–mediated Mr1tm1(KOMP)Vlcg, RRID:MMRRC_058774-UCD,
glycolysis. In contrast to recent work high- epithelial pathologies. Thus, the IL-17A–HIF1a obtained from the Mutant Mouse Resource
lighting a role for IL-17A–induced glycolysis in axis that we uncovered here may provide ther- and Research Center (MMRRC) at University
fibroblast survival and proliferation, wound- apeutic opportunities for a wide range of in- of California, Davis. Tg(Rorc-EGFP)1Ebe mice
edge epithelia did not require glycolysis to flammatory and metastatic diseases. were a gift from Dr. G. Eberl (Institut Pasteur).
proliferate (21). Instead, this mechanism of All animal studies were approved by the institu-
energy production was vital for epithelial mi- Materials and Methods tional animal care and use committee (IACUC).
gration at the wound front. HIF1a and glycol- Animals Mice were bred and maintained under specific-
ysis are drivers of tumor progression and The following mouse strains were purchased pathogen-free conditions at the Association for
metastasis, raising the possibility that IL-17A from The Jackson Laboratory: C57BL/6J, Assessment and Accreditation of Laboratory
or other immune-derived signals could drive B6.129P2(Cg)-Rorctm2Litt/J (Rorgt-EGFP), B6N.Cg- Animal Care (AAALAC)–accredited facilities
these pathways in cancer (30, 31). IL-17A is also Tg(KRT14-cre)1Amc/J, B6.Cg-Il17rctm1.1Koll/ at the New York University Langone Health
central to the pathology of many autoimmune J, B6.Cg-Tg(Cd4-cre)1Cwi/BfluJ, B6.129S- Center and housed in accordance with the

A Hk2 Pgk1 Slc2a1 B 3 Ctrl

rmIL-17A 0.0001
WT
0.0004 0.0003 0.0001
Expression (ddCT)
0.0002
2 0.0027
0.0011
3
2
1
GFP-KI 1
0
0
Slc2a1 Pfkl Aldoa Tpi1 Pgam1 Eno1 Ldha
D 4 Veh Veh+rmIL-17A
C 80
Ctrl rmIL-17A Glycolysis Glycolytic Capacity
Rapa Rapa+rmIL-17A
<0.0001
Expression (ddCT)
2-DG 50 80 <0.0001 3 <0.0001
Oligomycin <0.0001 <0.0001
ECAR (mpH/min)
ECAR (mpH/min)
ECAR (mpH/min)
60 40 <0.0001
60 <0.0001 <0.0001
Glucose 2 <0.0001
40 30
40 0.0054 <0.0001 0.0007 <0.0001 0.0024 0.0001
20
20 1
10 20
0 0 0 0
0 20 40 60 80 100 Ctrl rmIL-17A Ctrl rmIL-17A Slc2a1 Pfkl Aldoa Tpi1 Pgam1 Eno1 Ldha
Time (minutes)
E Migrating tongue
2.5 WT
Hif1aEKO
WT+rmIL-17A
Hif1aEKO+rmIL-17A <0.0001
F WT 1.5
G WT
Staining intensity (A.U)

1.5
Staining intensity (A.U)

Expression (ddCT)
2.0 0.0003
Glucose transporter 1
<0.0001 0.0003
<0.0001 0.0003
0.0001
1.5 ns
0.0004 1.0 1.0
0.0321 0.0017 ns
0.0001 0.0008 0.0486 0.0113
1.0 Il17rcEKO 0.5 Hif1aEKO 0.5
0.5
0.0 0.0
0.0 WT Il17rcEKO WT Hif1aEKO
Slc2a1 Pfkl Aldoa Tpi1 Pgam1 Eno1 Ldha Low High Low High
0.0003
H Veh rhIL-17A 2-DG 2-DG+rhIL-17A BAY872243 BAY872243+rhIL-17A
100 <0.0001ns
ns
0 hours
Scrach area (%)

80 0.0157
60
40
20
24 hours
24 hours
0
Veh 2-DG
rhIL-17A 2-DG+rhIL-17A
BAY872243 BAY872243+rhIL-17A
Fig. 6. IL-17A–HIF1a-mediated metabolic reprograming fuels re-epithelialization. transporter 1 staining in D3 WT and Il17rcEKO (n ≥ 5, N = 2) (F) and WT and
(A) ST plots of hk2, Pgk1, and Slc2a1 expression in GFP-KI (n = 3) and WT Hif1aEKO (n ≥ 4, N = 2) (G) migrating tongue. Scale bars, 100 mm. (H) Inhibition of
(n = 2) wounds. (B) Increased expression of glycolytic enzymes after rmIL-17A HIF1a (BAY872243) and glycolysis (2-DG) impairs the migration of primary
treatment (normalized to control) (n = 6, N = 4). (C) rmIL-17A stimulation human keratinocytes in the presence and absence of rhIL-17A. Yellow dashed line
results in functional enhancement of glycolysis. Representative ECAR of marks scratch wound edges. Scratched areas were quantified as a percentage
organoids cultured in the presence or absence of rmIL-17A. There were relative to the start area at 0 hours (n = 5, N = 3). Scale bars, 500 mm. In (B),
22 technical replicates. (D and E) IL-17A–induced expression of glycolytic (C), (F), and (G), two-tailed t test at a 95% confidence interval was used. In (D)
enzymes is abrogated in rapamycin-treated (D) and Hif1aEKO (E) organoids and (E), two-way ANOVA multiple-comparisons test was used. In (H), one-way
(n ≥ 5, N = 2). (F and G) pseudocolor fire images and quantifications of glucose ANOVA was used.
procedures outlined in the National Institutes a 27-year-old male) at Karolinska University and conducted according to the Declaration
of Health’s Guide for the Care and Use of Hospital, Stockholm, Sweden. After local lido- of Helsinki’s principles.
Laboratory Animals. Experiments were per- caine injection, two full-thickness excisional
formed with IACUC-approved protocols. Age- wounds were created using a 4-mm biopsy Skin wound-healing models
and sex-matched controls were used in all punch at the upper buttock area of each donor. Full-thickness wounds
experiments presented. Central excised skin was used as unwounded Seven- to 8-week-old mice in the telogen
control. Wound-edge skin was collected using (resting) phase of the hair cycle were used for
Human wound samples a 6-mm biopsy punch 1 and 7 days later. All wounding studies. After hair removal, a 4-mm
Normal and wounded human skin samples donors consented to the collection and use of biopsy punch (Miltex) was used to make full-
were collected from three healthy donors (a clinical samples. The study was approved by thickness wounds on dorsal skin. Wound clo-
48-year-old female, a 24-year-old female, and the Stockholm Regional Ethics Committee sure was assessed macroscopically with an

engineer’s caliper, and the wound-healing shown with the pseudocolor Fire from ImageJ Recombinant cytokines
area and rate were calculated as previously after they were contrasted equally and back- The wound edge was intradermally injected
described (9). ground was uniformly removed. with either rmIL-17A (500 ng in 20 ml of PBS,
Epithelial organoids were released from carrier-free; PeproTech) or PBS control im-
Splinted full-thickness wounds basement membrane matrix (BME) with or- mediately after a 4-mm biopsy. The following
Four-millimeter full-thickness wounds were ganoid harvesting solution (Cultrex), fixed in cytokine concentrations were used for daily
stented open by a 10-mm-diameter silicone 4% paraformaldehyde for 2 hours at 4°C, and intradermal injections for 7 days in unwounded
ring and secured with skin glue and eight then incubated in acetone at –20°C. Organoids ear pinnae and compared with the PBS-only
simple, interrupted 6-0 sutures. Wounds were were stained with conditions similar to tissue vehicle control: IL-17A (500 ng in 20 ml of PBS,
dressed immediately after splinting and dress- sections and embedded in agarose for imaging. carrier-free; PeproTech), IL-17F (1 mg in 20 ml
ings were changed every other day thereafter. Data were analyzed using ImageJ. Organoid of PBS, carrier-free; PeproTech), or IL-22 (500 ng
Animals were singly housed to limit disrup- HIF1a intensity was determined using the in 20 ml of PBS, carrier-free; PeproTech).
tion to splint. Single Target Translocation Assay from HSC
studio (Thermo Fisher). For the complete list 2-DG
Skin tissue digest of immunofluorescence antibodies and con- 2-DG (40 mg/ml; Cayman Chemical) was first
Unwounded or 0.25 mm of skin around the centrations used, see table S2. dissolved in PBS and then diluted in a water-
wound edge was excised and digested in based gel. This gel was topically applied to
Liberase TL (Sigma-Aldrich) as previously Microscopy and image processing dorsal skin 2 days before wounding and then
described (9). Images were acquired with a Zeiss Axioplan2 at the wound perimeter daily until analysis.
using a Plan-Apochromat 20×/0.8 air objec-
Flow cytometry and fluorescence-activated cell tive or Zeiss LSM 710 laser scanning confocal CITE-seq
sorting (FACS) microscope using Plan-Apochromat 20×/0.75 FACS-purified live CD45.2+CD90.2+ TCRVg3–
Single-cell suspensions were preincubated air and Plan-Apochromat 63×/1.40 oil objec- cells from unwounded skin at D3 and D5
with anti-CD16/32 before staining with sur- tives. Tiled and stitched images of sagittal sec- after wounding were prelabeled with surface-
face fluorescent conjugated antibodies and/or tions were collected using a Plan-Apochromat epitope-marking oligo-tagged antibodies and
oligo-tagged antibodies (table S1) at predeter- 20×/0.75 air objective, controlled by Zen soft- sample-specific oligo-tagged Totalseq-A anti-
mined concentrations in a 100 ml of staining ware (Zeiss). bodies (BioLegend; see table S1). Hashed sam-
buffer [phosphate-buffered saline (PBS) con- Organoids were imaged using Cell Insight ples were pooled at a 1:1.5:1 ratio of control:
taining 5% fetal bovine serum (FBS) and 1% CX7-LZR (Thermo Scientific), and images were D3:D5 before library preparation (Chromium
HEPES] per 107 cells. Stained cells were re- processed with HCS Studio Cell Analysis soft- Single Cell 3′ Library, 10X Genomics) and
suspended in 4′,6-diamidino-2-phenylindole ware (Thermo Scientific). Maximal projection sequenced on an Illumina HiSeq 4000 as
(DAPI) in FACS buffer before purification or z-stacks are presented, and co-localization was 150-bp paired-end reads. Sequencing results
analysis. For intracellular staining, fixable Live/ interpreted only in single z-stacks. z-stacks were demultiplexed and converted to FASTQ
Dead dye was added before surface antibody were projected using ImageJ. Images were format using Illumina bcl2fastq software. The
incubation, which was performed according to assembled in Adobe Illustrator CC2015.3. Cell Ranger Single-Cell Software Suite was used
the manufacturer’s instructions for the Foxp3/ to perform sample demultiplexing, barcode
Transcription Factor Staining Buffer Set (BD Animal drug treatments processing, and single-cell 3′ gene counting.
Biosciences). Flow cytometry data were ac- EdU pulse The cDNA insert was aligned to the mm10/
quired on LSRII Analyzers (BD Biosciences) Mice were injected intraperitoneally with EdU GRCm38 reference genome. Only confidently
and then analyzed with FlowJo program. FACS (50 mg/g) (Thermo Scientific) either 3 hours mapped, non–polymerase chain reaction (PCR)
was performed using FACSAria Cell Sorters (epithelial cells) or 24 hours (lymphocytes) duplicates with valid barcodes and unique
(BD Biosciences). before analysis. molecular identifiers (UMIs) were used to
generate the gene-barcode matrix. Cell Ranger
Immunofluorescence and image analysis FTY720 output was further analyzed in R using the
Tissue was fixed, processed, and stained as Mice were intraperitoneally injected with Seurat package (32). Surface epitope oligo
previously described (9). EdU reactions were 2 mg/kg of FTY720 or vehicle control for sequences were merged with cell transcriptome
performed according to the manufacturer’s 3 consecutive days before wounding and data by matching the cell barcode IDs.
directions (Life Technologies). Migrating then throughout the wound repair response.
tongues were determined by measuring the Ten microliters of blood were collected through Sample demultiplexing and quality controls
length of K14+ integrin a5+ cells. The number the jugular vein to test FTY70 treatment effi- Further analysis including quality filtering,
of wound-edge proliferating epithelial cells was cacy before wound analysis. the identification of highly variable genes,
determined by counting EdU+ K14+ cells from dimensionality reduction, standard unsuper-
the first hair follicle adjacent to the wound until Tamoxifen vised clustering algorithms, and the discovery
the wound’s leading edge. To detect PIM, slides To activate Cre recombinase, 7- to 8-week-old mice of differentially expressed genes was per-
were pretreated with a mouse-on-mouse im- were treated with 7.5 mg of 4-hydroxytamoxifen formed using the Seurat R package. Samples
munodetection kit (Vector Laboratories) and (Sigma-Aldrich) dissolved in 250 ml of corn oil were demultiplexed to filter out multiplets
then stained with biotin–anti-PIM adduct every other day by oral gavage three times be- (cells mapping to multiple hashtags) and
antibodies per the manufacturer’s guidelines fore wounding and continuously during wound negative cells (cells missing hashtags) with a
(Hydroxyprobe). ImageJ was used to quantify repair. positive quantile threshold of 0.99 between
PIM and glucose transporter 1 integrated den- samples. Individual samples were further
sity, a well-established method of measuring PIM HCl processed to remove cells with >20% mito-
fluorescence intensity that accounts for dif- PIM HCl (hypoxyprobe) was intraperitoneally chondrial gene expression. To exclude low-
ferences in the area of the signal in K14+ and injected into mice at the dose of 60 mg/kg quality cells and remaining multiplets or cells
integrin a5+ migrating tongue. Images are 2 hours before sacrifice. that were extreme outliers, the distribution of

total genes per cell was calculated. Then, con- ually aligned with Adobe Photoshop CS6 to rank sum test and a non-batch-transformed
trol parameters were applied to filter out cells identify the array of spots beneath the tissue. spatial assay was used for this test. The results
with <200 detected genes and >3800 detected were filtered to only include genes with ad-
genes. After removing unwanted cells from Sequence alignment and annotation justed P values <0.1 and natural log (fold-
the dataset, we normalized the data by the Sequencing output and the histology images change) >0.25 for up-regulated genes. The
total expression, multiplied by a scale factor of were processed using space ranger software pathway analysis was performed using the
10,000, and log-transformed the result. (10X Genomics). The space ranger mkfastq cluster Profiler package in R (35).
function was used for sample demultiplex-
Cell clustering and annotation for ing and to converting spatial barcodes and Multimodal integration analysis
CITE-seq data reads into FASTQ format. The space ranger The significance of the overlap between ST
Clustering was performed using the Seurat count function was used to align reads from genes and cell type marker genes from CITE-
R package. PCA was performed to generate FATSQ files to the mouse genome (mm10/ seq and published scRNA-seq data [Gene
graph-based cell clusters, and the top seven GRCm38) and then align the microscopic Expression Omnibus (GEO) accession number
PCs with a resolution of 0.6 were used. This slide image and transcriptome to generate GSE142471] were queried using the hyper-
yielded 16 unique cell clusters, which were barcode/UMI counts, feature spot matrices, geometric cumulative distribution, with all
further plotted on the UMAP. Next, contam- cluster data, and perform gene-expression genes as the background to compute the
inating nonlymphoid clusters were removed: analysis. P value. All genes were calculated as the num-
macrophages (Cd74, Fcgr1a, and Mrc1), fibro- ber of common genes detected between ST
blasts (Col1a2, Pdgfra, and Thy1), and den- Clustering analysis of ST data and CITE-seq in raw data.
dritic cells (Cd74, Cd207, and Itgax). Dendritic The data were then further analyzed using
epidermal T cells (Cd3ehi, Tcrdhi), which are the Seurat R package. To account for var- Low-input bulk-RNA sequencing of
unique to mice and are not found in human iance in sequencing depth across tissue spots wound-edge epithelium
skin, were also excluded. PCA was performed (biological variance driven by heterogeneity in A ring width of 0.25 mm skin around the
again, and the top seven PCs with a resolution cell density and transcriptional activity) while wound edge from control and Rorc-deficient
of 0.5 were used to generate a UMAP with controlling for technical artifacts, scTransform GFP-KI (two or three mice pooled per group
12 unique clusters. Each of these clusters was was used for normalization of gene expression per sequencing run) was excised and digested
then defined and labeled for cell types using by spots (34). Dimensionality reduction and as described in the skin tissue digest section
marker genes obtained by performing differ- clustering of sequencing data were run using above. Would-edge epithelium was FACS pu-
ential gene expression analysis between the the same workflow as CITE-seq. Before clus- rified from live Epcam+ cells on the basis of
clusters using Wilcoxon rank sum test with tering analysis, scab regions were filtered out integrin a6 and integrin a5 expression while
an adjusted P value < 0.1 and natural log (fold- from each ST sample using subset function in excluding CD45+ immune cells, CD31+ endo-
change) >0.25 as cutoff and parallel analysis of the Seurat R package. thelial cells, and CD117+ melanocytes. Total
surface epitope enrichment. Violin plots of For the WT control (unwounded) skin sam- RNA was extracted from FACS-purified integ-
marker genes and surface epitopes in fig. S1 ple, PCA was performed and the top 10 PCs rin a6+ integrin a5+ (migrating tongue) and
generated using the scanpy and seaborn pack- with a resolution of 1 were used to identify integrin a6+ integrin a5– (wound edge) sam-
ages in Python (33). spatial clusters. For the WT-wounded D3 skin ples using the RNeasy Plus Micro Kit (Life
sample, PCA was performed and the top 20 Technologies). Libraries were generated with
Spatial transcriptomics PCs with a resolution of 1 were used to identify the SMART-Seq v4 Ultra Low Input RNA Kit
Sample processing spatial clusters. UMAP dimensionality reduc- for Sequencing and the SMARTer ThruPLEX
Tissues were embedded in optimal cutting tion was performed (Fig. 2B) and then overlaid DNA-Seq Kit (Takara Bio). Libraries were se-
temperature medium and frozen in liquid on a histology image for spatial interpretation quenced on Nova 6000 system with 50-bp
nitrogen–chilled isopentane within 15 min of (Fig. 2A). Clusters were annotated on the basis paired-end reads.
harvesting. Ten-micrometer cryosections were of top differentially expressed genes and lo- Sequencing results were demultiplexed and
mounted onto the ST arrays (10X Genomics calization in tissue as revealed by histology. To converted to FASTQ format using Illumina
Visium) and stored at −80°C until use. Tissue combine and batch align multiple ST samples, bcl2fastq software. The reads were trimmed
sections were fixed in methanol at –20°C and the scTranform pipeline from the Seurat vi- for adapter sequences using trimmomatic
then stained with hematoxylin and eosin. Bright- gnette was used. After batch alignment, PCA (version 0.36) in paired end mode with the
field images were taken on a Leica SCN400F was performed and the top 18 PCs with reso- minimal read length option set to 35 base
slide scanner at 20× resolution. Slides were lution of 0.9 were used to identify spatial pairs, trailing set to 5, and the sliding window
permeabilized with permeabilization enzyme clusters, followed by UMAP dimensionality set to 4:15. The sequencing reads were aligned
for 5 min, as determined by the tissue optimi- reduction (Fig. 3, A and B). The data were to the mouse genome (mm10/GRCm38) using
zation protocol. Polyadenylated RNAs captured normalized using scTransform after batch the splice-aware STAR aligner (36). The feature
on the underlying arrays were resuspended in alignment to regress for replicates, and the Counts program was used to generate counts
1.2 ml of 0.1 N HCl for 5 min and reverse number of genes per spot and the spatial fea- for each gene on the basis of how many aligned
transcribed at 53°C for 45 min, followed by ture expression visualizations were performed reads overlap its exons (37). The raw counts
second-strand synthesis at 65°C for 5 min. After using the normalized data. were further converted to TPM counts to filter
library preparation, samples were sequenced for genes with a minimal TPM value ≥10 be-
on a Novaseq 600 (Illumina) Differential expression analysis for ST data fore differential expression analysis. The raw
The markers genes were determined using gene counts were then subset using these TPM-
Spot selection and image alignment FindAllMarkers function in Seurat, and the filtered genes (n = 9412) and used to test for
After probe cleavage, fluorescent images were Wilcoxon rank-sum test was used for differen- differential expression using negative binomial
taken on a Hamamatsu NanoZoomer whole- tial testing. The differential expression analysis generalized linear models implemented by
slide fluorescence scanner. Bright-field images between the wounded GFP-KI and wounded the DESeq2 R package (38). Volcano plots of
of the tissue and fluorescent images were man- WT group was also performed using Wilcoxon differentially expressed genes (log2-fold change

≥0.25, adjusted P value <0.1) were generated b-mercaptoethanol supplemented with the Halt monolayer with a 1-ml pipette tip. Cells were
using the Enhanced Volcano package in R. Protease and Phosphatase Inhibitor Cocktail treated with rhIL-17A (100 ng/ml, Peprotech).
Pathways analysis of differentially expressed (Thermo Fisher)] and dissociated by passing The scratch area was imaged at 0 and 24 hours
and up-regulated genes (log2-fold change ≥0.25, through a 30-gauge needle. Organoids cultured using EVOS (Thermo Fisher) and quantified
adjusted P value <0.1) was performed using in a hypoxia chamber were lysed directly in using ImageJ.
the cluster Profiler package in R with the Gene the wells without releasing with organoid-
Ontology (GO) (39) and Kyoto Encyclopedia of harvesting solution. After incubation at 100°C Statistics
Genes and Genomes (KEGG) (40) databases. for 10 min, protein concentration was measured Data are presented as mean ± SEM. Group
with the DC Protein Assay Kit (Bio-Rad). XT sizes were determined on the basis of the re-
In vivo oxygen measurements sample buffer was added to the protein lysate sults of preliminary experiments. Mice were
Tissue oxygen levels were measured using (Bio-Rad) and incubated in 95°C for 5 min assigned at random to groups. Experiments
OxyLite probes (Oxford Optronix). The OxyLite before being resolved in SDS–polyacrylamide were not performed in a blinded fashion. Sta-
NXpO2 Barer Fiber Sensor with a ∼250-mm tip gel electrophoresis and analyzed with immu- tistical significance for each experiment was
was directly inserted into wounded or healthy noblot. Antibodies used for immunoblots are determined as shown in the figure legends,
skin of an anesthetized mouse to continuously listed in table S2. Figures present represent- where n = the number of independent biologi-
monitor oxygen pressure (pO2) for 15 to 20 min. ative immunoblots quantified with ImageJ; cal replicates (animals, unless noted as cells)
pO2 levels were recorded and analyzed with a integrated density for each band was normal- per group, and N = the number of indepen-
data-acquisition system (LabChart Reader, Chart ized with internal b-actin. dent experimental replicates. Statistical signif-
version 8 for Windows; AD Instruments) icance of bulk RNA-seq data was calculated
Seahorse extracellular acidification using an adjusted P value cutoff <0.1 in the
Organoid culture rate analysis DESeq2 R package. Statistical analyses were
Epithelial organoids were generated as pre- Extracellular acidification rate (ECAR) mea- performed in Prism (GraphPad), DESeq2, or
viously described from the dorsal skin of surements were performed using an XFe96 R. No data points were excluded.
C56BL/6 mice (ages postnatal days 46 to 50) analyzer (Agilent Technologies). Seahorse
REFERENCES AND NOTES
(41). One day after culturing in normoxia XFe96 plates (Agilent Technologies) were
1. R. Chovatiya, R. Medzhitov, Stress, inflammation, and defense
(21% O2) or hypoxia (2% O2), rmIL-17A (500 ng/ml, coated with a 10× dilution of BME in Dulbecco’s of homeostasis. Mol. Cell 54, 281–288 (2014). doi: 10.1016/
Peprotech), TNFa (100 ng/ml, Peprotech) (42), modified Eagle’s medium (DMEM; Sigma- j.molcel.2014.03.030; pmid: 24766892
IL-1b (100 ng/ml, Peprotech) (43), IL-6 (100 ng/ Aldrich). IL-17A–treated or control organoids 2. S. A. Eming, P. Martin, M. Tomic-Canic, Wound repair and
regeneration: Mechanisms, signaling, and translation.
ml, Peprotech) (44), epidermal growth factor were collected with organoid harvest solution, Sci. Transl. Med. 6, 265sr6 (2014). doi: 10.1126/
(500 ng/ml, Peprotech) (41), or PBS was added washed once in DMEM basic medium, resus- scitranslmed.3009337; pmid: 25473038
to the organoid culture medium and replen- pended in DMEM basic medium with 2 mM 3. Y. Belkaid, O. J. Harrison, Homeostatic immunity and the
microbiota. Immunity 46, 562–576 (2017). doi: 10.1016/
ished on D3. Organoids were harvested on D6 glutamine, and seeded onto the coated plates j.immuni.2017.04.008; pmid: 28423337
using a nonenzymatic cultrex organoid harvest- at 175 ml per well. Respiratory rates were mea- 4. O. J. Harrison et al., Commensal-specific T cell plasticity
ing solution (R&D Systems) for downstream sured in response to sequential injections of promotes rapid tissue adaptation to injury. Science 363,
eaat6280 (2019). doi: 10.1126/science.aat6280;
analyses. For inhibitor studies, organoids were glucose (10 mM), oligomycin (2 mM), and 2-DG pmid: 30523076
treated 12 hours before harvest with 60 nM (50 mM) (all Cayman Chemical). Values were 5. S. Naik, S. B. Larsen, C. J. Cowley, E. Fuchs, Two to Tango:
rapamycin (LC Laboratories) (45) and vehicle normalized to a total number of cells in or- Dialog between Immunity and Stem Cells in Health and
Disease. Cell 175, 908–920 (2018). doi: 10.1016/
control or 3 hours with U0126-ERK1/2 inhibitor ganoid culture well using the CyQUANT Cell
j.cell.2018.08.071; pmid: 30388451
(10 mM, MedChemExpress) (46) or MK-2206- Proliferation Assay (Thermo Fisher). 6. M. Aragona et al., Defining stem cell dynamics and migration
AKT inhibitor (5 mM, MedChemExpress) (47) during wound healing in mouse skin epidermis. Nat. Commun. 8,
and vehicle control. Inhibitor concentrations Lactate secretion assay 14684 (2017). doi: 10.1038/ncomms14684; pmid: 28248284
7. S. Bukhari, A. F. Mertz, S. Naik, Eavesdropping on the
and time course were chosen on the basis of Organoid culture supernatant was harvested conversation between immune cells and the skin epithelium.
published studies, efficacy of inhibition, and and centrifuged at 1000g to pellet debris. Int. Immunol. 31, 415–422 (2019). doi: 10.1093/intimm/
conditions that minimized organoid cell death Lactate was measured using the colorimetric dxy088; pmid: 30721971
8. R. Moncada et al., Integrating microarray-based spatial
in our cultures. For protein stability studies, L-Lactate Assay Kit (Abcam). Values were nor- transcriptomics and single-cell RNA-seq reveals tissue
cycloheximide (50 mg/ml, Cayman Chemical) malized to a total number of cells in each architecture in pancreatic ductal adenocarcinomas.
was added to the culture for 5, 10, and 15 min. organoid culture well using the CyQUANT Cell Nat. Biotechnol. 38, 333–342 (2020). doi: 10.1038/
s41587-019-0392-8; pmid: 31932730
For mRNA stability studies, actinomycin D Proliferation Assay (Thermo Fisher). 9. S. Naik et al., Inflammatory memory sensitizes skin epithelial
(10 mg/ml, Cayman Chemical) was added to stem cells to tissue damage. Nature 550, 475–480 (2017).
the culture for 2, 5, and 8 hours. Human primary keratinocyte scratch assay doi: 10.1038/nature24271; pmid: 29045388
Primary human keratinocytes were isolated 10. G. L. Semenza, Hypoxia-inducible factors in physiology and
Organoid assays from the foreskin and then cultured in serum-
medicine. Cell 148, 399–408 (2012). doi: 10.1016/
j.cell.2012.01.021; pmid: 22304911
Quantitative PCR free Keratinocytes Growth Medium KGM- 11. M. Stoeckius et al., Simultaneous epitope and transcriptome
Total RNA was extracted from organoids using Gold (Lonza, Basel, Switzerland) supplemented measurement in single cells. Nat. Methods 14, 865–868
(2017). doi: 10.1038/nmeth.4380; pmid: 28759029
RNeasy Plus Micro Kit, and equal amounts of with bovine pituitary extract, human endothe-
12. J. J. O’Shea, W. E. Paul, Mechanisms underlying lineage
RNA were reverse-transcribed using the super- lial growth factor, bovine insulin, hydrocortisone, commitment and plasticity of helper CD4+ T cells. Science
script VILO cDNA synthesis kit (Invitrogen). gentamicin-amphotericin B (GA-1000), epineph- 327, 1098–1102 (2010). doi: 10.1126/science.1178334;
Analytes were normalized using the house- rine, and transferrin. pmid: 20185720
13. G. Eberl, RORgt, a multitask nuclear receptor at mucosal
keeping gene Actb. For the complete list of For scratch migration assays, keratinocytes surfaces. Mucosal Immunol. 10, 27–34 (2017). doi: 10.1038/
quantitative PCR primers, refer to table S3. were allowed to reach confluency. Cells were mi.2016.86; pmid: 27706126
pretreated with mitomycin C (8 mg/ml) 2 hours 14. M. J. McGeachy, D. J. Cua, S. L. Gaffen, The IL-17 family of
Immunoblot before and with 2-DG (40 mM) and BAY872243 cytokines in health and disease. Immunity 50, 892–906
(2019). doi: 10.1016/j.immuni.2019.03.021; pmid: 30995505
Organoids were lysed with RIPA buffer [62.5 mM (10 mM) 1 hour before scratch assay. Scratches 15. G. Eberl et al., An essential function for the nuclear receptor
Tris-HCl, pH 6.8, 2% SDS, 25% glycerol, 5% were created by manual scraping of the cell RORgamma(t) in the generation of fetal lymphoid tissue

inducer cells. Nat. Immunol. 5, 64–73 (2004). doi: 10.1038/ prostate and breast cancer cells. Oncotarget 6, 130–143 47. Y. Li et al., Induction of expansion and folding in human
ni1022; pmid: 14691482 (2015). doi: 10.18632/oncotarget.2766; pmid: 25426557 cerebral organoids. Cell Stem Cell 20, 385–396.e3 (2017).
16. J. R. Huh et al., Digoxin and its derivatives suppress TH17 cell 32. A. Butler, P. Hoffman, P. Smibert, E. Papalexi, R. Satija, doi: 10.1016/j.stem.2016.11.017; pmid: 28041895
differentiation by antagonizing RORgt activity. Nature 472, Integrating single-cell transcriptomic data across different
486–490 (2011). doi: 10.1038/nature09978; pmid: 21441909 conditions, technologies, and species. Nat. Biotechnol. 36, AC KNOWL ED GME NTS
17. I. Pastar et al., Epithelialization in wound healing: A 411–420 (2018). doi: 10.1038/nbt.4096; pmid: 29608179 We thank D. R. Littman, D. Rosenblum, L. Guenin-Mace, B. G. Neel,
comprehensive review. Adv. Wound Care 3, 445–464 (2014). 33. F. A. Wolf, P. Angerer, F. J. Theis, SCANPY: Large-scale single- and M. R. Phillips for helpful discussions, advice, and/or critical
doi: 10.1089/wound.2013.0473; pmid: 25032064 cell gene expression data analysis. Genome Biol. 19, 15 (2018). reading of this manuscript. The following core facilities enabled our
18. A. Baeyens, V. Fang, C. Chen, S. R. Schwab, Exit strategies: doi: 10.1186/s13059-017-1382-0; pmid: 29409532 study: NYULMC High-Performance Computing, Flow Cytometry,
S1P signaling and T cell migration. Trends Immunol. 36, 34. C. Hafemeister, R. Satija, Normalization and variance Genome Technology Center, Histopathology Core, the Microscopy
778–787 (2015). doi: 10.1016/j.it.2015.10.005; stabilization of single-cell RNA-seq data using regularized Laboratory, and Applied Bioinformatics Laboratory. Funding:
pmid: 26596799 negative binomial regression. Genome Biol. 20, 296 (2019). This work was supported by Cancer Center Support Grant
19. D. Haensel et al., Defining epidermal basal cell states during doi: 10.1186/s13059-019-1874-1; pmid: 31870423 P30CA016087 from the Laura and Isaac Perlmutter Cancer Center
skin homeostasis and wound healing using single-cell 35. G. Yu, L. G. Wang, Y. Han, Q. Y. He, clusterProfiler: An (core facility subsidies), the National Psoriasis Foundation (Early
transcriptomics. Cell Rep. 30, 3932–3947.e6 (2020). R package for comparing biological themes among gene Career Research Grant to P.K.), a Bernard Levine Postdoctoral
doi: 10.1016/j.celrep.2020.02.091; pmid: 32187560 clusters. OMICS 16, 284–287 (2012). doi: 10.1089/ Fellowship in Immunology to P.K., the American Association
20. M. G. Constantinides et al., MAIT cells are imprinted by the omi.2011.0118; pmid: 22455463 of Immunologists (Intersect Fellowship to Y.X.), the Howard
microbiota in early life and promote tissue repair. Science 366, 36. A. Dobin et al., STAR: Ultrafast universal RNA-seq aligner. Hughes Medical Institute (HHMI Gilliam Fellowship GT13641 to
eaax6624 (2019). doi: 10.1126/science.aax6624; Bioinformatics 29, 15–21 (2013). doi: 10.1093/bioinformatics/ K.P.M), a Pew Stewert Scholar Award (00034119 to S.N.), and
pmid: 31649166 bts635; pmid: 23104886 the National Institutes of Health (NIH grants 1K22AI135099-01,
21. S. Majumder et al., IL-17 metabolically reprograms activated 37. Y. Liao, G. K. Smyth, W. Shi, featureCounts: An efficient general 1DP2AR079173-01, and R01-AI168462 to S.N.). S.N. is a NYSCF
fibroblastic reticular cells for proliferation and survival. purpose program for assigning sequence reads to genomic Robertson Stem Cell Investigator and a Packard Fellow. Author
Nat. Immunol. 20, 534–545 (2019). doi: 10.1038/ features. Bioinformatics 30, 923–930 (2014). doi: 10.1093/ contributions: S.N, P.K., and Y.X. conceptualized the study,
s41590-019-0367-4; pmid: 30962593 bioinformatics/btt656; pmid: 24227677 designed experiments, and wrote the manuscript. P.K., Y.X. and
22. X. Chen et al., IL-17R-EGFR axis links wound healing to 38. M. I. Love, W. Huber, S. Anders, Moderated estimation of fold I.S. performed animal, flow cytometry, microscopy, organoid, and
tumorigenesis in Lrig1+ stem cells. J. Exp. Med. 216, 195–214 change and dispersion for RNA-seq data with DESeq2. Genome genomic experiments. I.S. and A.T. performed computational
(2019). doi: 10.1084/jem.20171849; pmid: 30578323 Biol. 15, 550 (2014). doi: 10.1186/s13059-014-0550-8; analysis for CITE-seq, ST, and low-input RNA-seq. P.K. K.P.M, C.L.,
23. A. J. Majmundar, W. J. Wong, M. C. Simon, Hypoxia-inducible pmid: 25516281 and A.H. performed ST. B.H. and K.P.M aided in animal studies and
factors and the response to hypoxic stress. Mol. Cell 40, 39. M. Ashburner et al., Gene ontology: Tool for the unification of colony maintenance. D.E.B. assisted with the seahorse glycolysis
294–309 (2010). doi: 10.1016/j.molcel.2010.09.022; biology. Nat. Genet. 25, 25–29 (2000). doi: 10.1038/75556; stress test. M.C. aided in fluorescent image acquisition and
pmid: 20965423 pmid: 10802651 analysis. S.B.L. assisted with rapamycin studies. A.N.T. assisted
24. M. Laplante, D. M. Sabatini, mTOR signaling in growth control 40. M. Kanehisa, M. Furumichi, Y. Sato, M. Ishiguro-Watanabe, with low-input RNA-seq. D.L and N.X.L provided human acute
and disease. Cell 149, 274–293 (2012). doi: 10.1016/ M. Tanabe, KEGG: Integrating viruses and cellular organisms. wound samples. Competing interests: S.N. is on the SAB of Seed
j.cell.2012.03.017; pmid: 22500797 Nucleic Acids Res. 49 (D1), D545–D551 (2021). doi: 10.1093/ Inc., is a consultant for BiomX, and receives funding from Takeda
25. H. Harada et al., The Akt/mTOR pathway assures the synthesis nar/gkaa970; pmid: 33125081 Pharmaceuticals. A.T. is a scientific advisor to Intelligencia.ai.
of HIF-1alpha protein in a glucose- and reoxygenation- 41. K. E. Boonekamp et al., Long-term expansion and The remaining authors declare no competing interests. Data and
dependent manner in irradiated tumors. J. Biol. Chem. 284, differentiation of adult murine epidermal stem cells in materials availability: All data to support the conclusions in this
5332–5342 (2009). doi: 10.1074/jbc.M806653200; 3D organoid cultures. Proc. Natl. Acad. Sci. U.S.A. 116, manuscript can be found in the main text or the supplemental
pmid: 19098000 14630–14638 (2019). doi: 10.1073/pnas.1715272116; materials. Genomic data are publicly available at the GEO
26. J. Zhu, C. B. Thompson, Metabolic regulation of cell growth pmid: 31253707 (accession number GSE166950). Raw data for in vivo measurements
and proliferation. Nat. Rev. Mol. Cell Biol. 20, 436–450 (2019). 42. T. Grabinger et al., Ex vivo culture of intestinal crypt organoids of wound repair and image quantifications from our study may be
doi: 10.1038/s41580-019-0123-5; pmid: 30976106 as a model system for assessing cell death induction in requested from the corresponding authors. License information:
27. C. T. Taylor, J. C. McElwain, Ancient atmospheres and the intestinal epithelial cells and enteropathy. Cell Death Dis. 5, Copyright © 2022 the authors, some rights reserved; exclusive
evolution of oxygen sensing via the hypoxia-inducible factor in e1228 (2014). doi: 10.1038/cddis.2014.183; pmid: 24832600 licensee American Association for the Advancement of Science. No
metazoans. Physiology 25, 272–279 (2010). doi: 10.1152/ 43. H. F. Farin et al., Paneth cell extrusion and release of claim to original US government works. https://www.science.org/
physiol.00029.2010; pmid: 20940432 antimicrobial products is directly controlled by immune about/science-licenses-journal-article-reuse
28. H. R. Rezvani et al., Loss of epidermal hypoxia-inducible factor- cell-derived IFN-g. J. Exp. Med. 211, 1393–1405 (2014).
1a accelerates epidermal aging and affects re-epithelialization doi: 10.1084/jem.20130753; pmid: 24980747 SUPPLEMENTARY MATERIALS
in human and mouse. J. Cell Sci. 124, 4172–4183 (2011). 44. M. Soutto et al., Activation of STAT3 signaling is mediated by
science.org/doi/10.1126/science.abg9302
doi: 10.1242/jcs.082370; pmid: 22193962 TFF1 silencing in gastric neoplasia. Nat. Commun. 10, 3039
Figs. S1 to S14
29. Y. Wang et al., Long-term culture captures injury-repair cycles (2019). doi: 10.1038/s41467-019-11011-4; pmid: 31292446
Tables S1 to S4
of colonic stem cells. Cell 179, 1144–1159.e15 (2019). 45. K. B. Jung et al., Interleukin-2 induces the in vitro maturation
MDAR Reproducibility Checklist
doi: 10.1016/j.cell.2019.10.015; pmid: 31708126 of human pluripotent stem cell-derived intestinal organoids. Nat.
30. G. L. Semenza, Targeting HIF-1 for cancer therapy. Nat. Rev. Commun. 9, 3039 (2018). doi: 10.1038/s41467-018-05450-8; View/request a protocol for this paper from Bio-protocol.
Cancer 3, 721–732 (2003). doi: 10.1038/nrc1187; pmid: 30072687
pmid: 13130303 46. Y. H. Tsai et al., In vitro patterning of pluripotent stem Submitted 4 February 2021; resubmitted 19 October 2021
31. T. Shiraishi et al., Glycolysis is the primary bioenergetic cell-derived intestine recapitulates in vivo human development. Accepted 31 May 2022
pathway for cell motility and cytoskeletal remodeling in human Development 144, 1045–1055 (2017). pmid: 27927684 10.1126/science.abg9302

RES EARCH
◥ aptic connectivity. Therefore, the analysis of

RESEARCH ARTICLE SUMMARY nonepileptic surgery tissue seemed of particu-
lar importance. We also included data from
NEUROSCIENCE one macaque individual, who was not known
to have any brain-related pathology.
Connectomic comparison of mouse and human cortex
RESULTS: We acquired three-dimensional elec-
Sahil Loomba†, Jakob Straehle†, Vijayan Gangadharan†, Natalie Heike†, Abdelrahman Khalifa†, tron microscopy data from temporal and frontal
Alessandro Motta†, Niansheng Ju, Meike Sievers, Jens Gempt, Hanno S. Meyer, Moritz Helmstaedter* cortex of human and temporal and parietal cor-
tex of macaque. From these, we obtained con-
nectomic reconstructions and compared these
INTRODUCTION: The analysis of the human is on the circuit configurations found in the with five connectomes from mouse cortex. On
brain is a central goal of neuroscience, but for human cerebral cortex is not known. This is, the basis of these data, we were able to deter-
methodological reasons, research has focused however, of particular interest also to the study mine the effect of the about 2.5-fold expansion
on model organisms, the mouse in particular. of neuropsychiatric disorders because in these, of the interneuron pool in macaque and human
Because substantial homology was found at the the alteration of inhibitory-to-excitatory synap- cortex compared with that of mouse. Contrary
level of ion channels, transcriptional programs, tic balance has been identified as one possible to expectation, the inhibitory-to-excitatory syn-
and basic neuronal types, a strong similarity of mechanistic underpinning. aptic balance on pyramidal neurons in macaque
neuronal circuits across species has also been and human cortex was not substantially altered.
assumed. However, a rigorous test of the con- RATIONALE: We used recent methodological Rather, the interneuron pool was selectively
figuration of local neuronal circuitry in mouse improvements in connectomics to acquire data expanded for bipolar-type interneurons, which
versus human—in particular, in the gray matter from one macaque and two human individuals, prefer the innervation of other interneurons,
of the cerebral cortex—is missing. using biopsies of the temporal, parietal, and and which further increased their preference for
The about 1000-fold increase in number of frontal cortex. Human tissue was obtained from interneuron innervation from mouse to human.
neurons is the most obvious evolutionary change neurosurgical interventions related to tumor re- These changes were each multifold, yielding in
of neuronal network properties from mouse moval, in which access path tissue was harvested effect an about 10-fold expanded interneuron-
to human. Whether the structure of the local that was not primarily affected by the under- to-interneuron network in the human cortex
cortical circuitry has changed as well is, how- lying disease. A key concern in the analysis of that is only sparsely present in mouse. The
ever, unclear. Recent data from transcriptomic human patient tissue has been the relation total amount of synaptic input to pyramidal
analyses has indicated an increase in the pro- to epilepsy surgery, when the underlying dis- neurons, however, did not change according to
portion of inhibitory interneurons from mouse ease has required often year-long treatment the threefold thickening of the cortex; rather,
to human. But what the effect of such a change with pharmaceuticals, plausibly altering syn- a modest increase from about 12,000 synaptic
inputs in mouse to about 15,000 in human
was found.
A B
Human CONCLUSION: The principal cells of the cerebral
cortex, pyramidal neurons, maintain almost
Human
constant inhibitory-to-excitatory input balance
and total synaptic input across 100 million years
of evolutionary divergence, which is particularly
noteworthy with the concomitant 1000-fold ex-
pansion of the neuronal network size and the
2.5-fold increase of inhibitory interneurons from
mouse to human. Rather, the key network change
from mouse to human is an expansion of almost
~10-fold an order of magnitude of an interneuron-to-
expansion interneuron network that is virtually absent
Macaque of the
inte
interneuron-interneuron
in mouse but constitutes a substantial part
20 M years of the human cortical network. Whether this
network
Mouse
Mous new network is primarily created through
the expansion of existing neuronal types, or
100 M years
yea is related to the creation of new interneuron
subtypes, requires further study. The discov-
ery of this network component in human
cortex encourages detailed analysis of its
Mouse
function in health and disease.
▪
Connectomic screening across mammalian species: Comparison of five mouse, two macaque, and The list of author affiliations is available in the full article online.
two human connectomic datasets from the cerebral cortex. (A) Automated reconstructions of all *Corresponding author. Email: mh@brain.mpg.de
neurons with their cell bodies in the volume shown, using random colors. The analyzed connectomes
Cite this article as S. Loomba et al., Science 377, eabo0924
comprised a total of ~1.6 million synapses. Arrows indicate evolutionary divergence: the last common (2022). DOI: 10.1126/science.abo0924
ancestor between human and mouse, approximately 100 million years ago, and the last common ancestor
between human and macaque, about 20 million years ago. (B) Illustration of the about 10-fold expansion of READ THE FULL ARTICLE AT
the interneuron-to-interneuron network from mouse to human. https://doi.org/10.1126/science.abo0924
Loomba et al., Science 377, 171 (2022) 8 July 2022 1 of 1

RES EARCH
◥ additional data from macaque in our study,

RESEARCH ARTICLE we furthermore could ensure that none of the
reported effects was exclusively attributable to
NEUROSCIENCE a diseased state of the human brain. We deter-
mined the circuit effects of an expanded inhib-
Connectomic comparison of mouse and human cortex itory neuron population in the human cortex
and obtained a quantitative picture of the hu-
Sahil Loomba1,2†, Jakob Straehle1†‡§, Vijayan Gangadharan1†¶, Natalie Heike1†, man pyramidal cell synaptic input architecture.
Abdelrahman Khalifa1†, Alessandro Motta1†, Niansheng Ju1, Meike Sievers1,2, Jens Gempt3,
Hanno S. Meyer3, Moritz Helmstaedter1* Results
We obtained tissue samples (Fig. 1) from the
The human cerebral cortex houses 1000 times more neurons than that of the cerebral cortex of a cerebral cortex of two human individuals (a
mouse, but the possible differences in synaptic circuits between these species are still poorly 69-year-old female and an 80-year-old male)
understood. We used three-dimensional electron microscopy of mouse, macaque, and human cortical who underwent neurosurgical operations
samples to study their cell type composition and synaptic circuit architecture. The 2.5-fold increase in (access tissue that had to be removed during
interneurons in humans compared with mice was compensated by a change in axonal connection surgery and would have been discarded other-
probabilities and therefore did not yield a commensurate increase in inhibitory-versus-excitatory wise) (Materials and methods) and of one
synaptic input balance on human pyramidal cells. Rather, increased inhibition created an expanded rhesus macaque (Macaca mulatta, 15.6-year-
interneuron-to-interneuron network, driven by an expansion of interneuron-targeting interneuron old male). Individual ages corresponded to
types and an increase in their synaptic selectivity for interneuron innervation. These constitute key 83, 102, and 52% of statistical life expectancy,
neuronal network alterations in the human cortex. respectively.
From these tissue samples, we acquired 3D
T
EM image datasets in cortical L2/3 using either
he human brain, with its 86 billion nerve 25%, (3, 4, 31–37), thus potentially on par with serial blockface EM (SBEM) (50) [macaque
cells (1) forming a network of unparal- data from primates (range 15 to 37%) (38–42). primary somatosensory cortex (S1) dataset
leled complexity, is of special interest for Recent transcriptomic data substantiates an at sized 176 by 224 by 100 mm; superior temporal
neuroscience. Yet over the past 50 years, least twofold increase (4, 9), suggesting the gyrus (STG) sized 175 by 225 by 108 mm; human
rodents (in particular, the mouse) have balance between inhibitory and excitatory (H5) STG sized 166 by 216 by 112 mm, and human
emerged as key research subjects, offering synapses to be substantially shifted toward (H6) inferior frontal gyrus (IFG) sized 170 by
methodological opportunities not available inhibition. Furthermore, on the basis of the 215 by 79 mm, all at a voxel size of 11.24 by 11.24
for the study of the human brain. Because at about threefold larger extent of the dendritic by (28 to 30) nm] or automated tape-collecting
the molecular level evolutionary homology is trees of human pyramidal cells, it has been ultramicrotome multiple scanning EM (ATUM-
substantial for ion channels, synaptic recep- assumed that human cortical neurons receive multiSEM) (51–53) [dataset human (H5) STG
tors, and other key molecular constituents of substantially more synapses than those of ro- spanning all cortical layers, sized 1.7 mm by
the brain, similar homology has been assumed dents (for example, 10,000 in rodents versus 2.1 mm by 28.3 mm and dataset from the same
for neuronal circuits, especially in the cerebral 30,000 in human) (29, 43–46). sample focused on L2/3 sized 1.1 mm by 1.1 mm
cortex. However, comparative synaptic-resolution However, a circuit-level analysis of the by 82 mm (total 0.41 PB of data) both at a voxel
connectomic studies of mouse and human human cortex that addresses the potential size of 4 by 4 by (35 to 40) nm]. For comparison
cortex are required to determine the degree to effect of multifold increased inhibitory cir- with mouse cortex, we densely reconstructed
which circuit structure may have evolved be- cuit elements is still missing. Would, as a (49, 54) previously published 3D EM datasets
tween species. result, the inhibitory-to-excitatory synaptic from L2/3 of S1, secondary visual (V2), posterior
Detailed studies of the human cerebral cor- balance be increased in the human cortex? parietal (PPC), and anterior cingulate (ACC)
tex have provided data on cellular composi- Because inhibitory-to-excitatory synaptic bal- cortical areas (55) and acquired an additional
tion of the neuropil (1–14), synaptic properties ance has emerged as a key set point that might dataset from mouse secondary auditory cor-
(15–25), and neuronal morphology (14, 26–28), be altered in neuropsychiatric diseases [studied tex (A2) sized 115 by 175 by 109 mm, approx-
yielding a comparative description with refer- primarily in mouse models (47, 48)], this imately corresponding to the location of the
ence to rodents that indicates a larger number of question is relevant for the healthy and diseased STG datasets from macaque and human. In the
glial cells (2), larger synapses (20, 29, 30), and human brain. following, we report effects that were multi-
more inhibitory interneurons (INs) (4, 9) to form We used three-dimensional (3D) electron fold between mouse (n = 5 individuals, n = 5
the neuronal network of the human cortex. microscopy (EM) followed by sparse and cortical regions) and macaque or human (n = 3
This latter comparison had been confounded dense circuit reconstruction (49) to map the individuals, n = 4 cortical regions).
by rodent data reporting an IN fraction of 8 to synaptic and cellular composition of layer 2/3
(L2/3) in mouse, macaque, and human cortex. Inhibitory versus excitatory neuronal composition
1
Department of Connectomics, Max Planck Institute for Brain Sampling from multiple individuals and corti- The fraction of nerve cells that were INs in-
Research, Frankfurt, Germany. 2Faculty of Science, Radboud cal regions, we screened for key connectomic creased 2.5-fold between mouse and macaque
University, Nijmegen, Netherlands. 3Department of
Neurosurgery, Klinikum rechts der Isar, School of Medicine,
alterations between mouse and human cortex, or human [n = 88 of 734 versus 278 of 914
Technical University of Munich, Germany. which a single dataset from a given human neurons from n = 5 mouse and n = 4 macaque
*Corresponding author. Email: mh@brain.mpg.de individual and disease condition would not and human datasets pooled, P < 0.001 (Mate-
have allowed (10). Although our human data rials and methods)] (Fig. 2, A and B), which is
‡Present address: Department of Neurosurgery, University Freiburg,
Freiburg, Germany. was from diseased individuals who underwent consistent with recent reports from transcrip-
§Present address: Berta-Ottenstein Clinician Scientist Program, neurosurgical interventions, the patients had tomic cell type analyses (4, 9). This change
Faculty of Medicine, University Freiburg, Freiburg, Germany. not undergone yearlong treatment for epi- in the neuronal composition of cortical tissue
¶Present address: Institute for Vascular and Islet Cell Biology, German
Diabetes Center, Leibniz Center for Diabetes Research at Heinrich leptic disorders that could have substantially could have profound effects on the synaptic in-
Heine University Düsseldorf, Düsseldorf, Germany. altered the synaptic network. By including put to pyramidal (excitatory) neurons (ExNs)
Loomba et al., Science 377, eabo0924 (2022) 8 July 2022 1 of 13

(mouse) versus 0.86 ± 0.36 (macaque) and

0.67 ± 0.25 (human); mean ± SD; n = 46, 36,
and 49, respectively; P < 10−18, KS test mouse
versus macaque and human] (Fig. 2, D and E).
As a result, the fraction of synapses made onto
Human
216 µm
215 µm
the dendritic shafts of pyramidal cells (out of
all input synapses to these dendrites) was
Chim-
panzee increased 2.7-fold (n = 1111 synapses, mouse
~5-6 M and n = 1638 synapses macaque and human,
years
166 µm 112 µm P < 0.001, KS test) (Fig. 2, F and H). Synaptic
Superior Inferior frontal 170
1 µmµm inputs to spine necks, stubs, or double inner-
temporal gyrus (IFG)
gyrus (STG) 79 µm
vation of dendritic spines were rare but slightly
increased in macaque and human (Fig. 2G;
statistics are provided in the caption).
In mouse, synapses onto dendritic spines
have been reported to be predominantly ex-
citatory (49, 56) and synapses onto dendritic
shafts predominantly inhibitory (55, 57, 58).
219 µm
Macaque
218 µm
~20 M Therefore, the finding of a 2.7-fold increased

years
fraction of shaft synapses could indicate a
substantial shift in the set point for the i/e
balance in human [as recently claimed for
4 cm
1 cm 167 µm nonhuman primate (59)]. For this to be cor-
Superior 167 µm 100 µm rect, however, the properties of excitatory and
temporal 106 µm S1
gyrus (STG)
142 µm inhibitory axons—in particular, their prefer-
88 µm
91 µm
ence for establishing synapses onto dendritic
213 µm
shafts and spines—would have to remain unal-
56 µm
tered between mouse and macaque and human.

Rat Mouse We investigated this next (Figs. 3 and 4).
S1 PPC 71 µm
~30 M
years
152 µm Synaptic properties of excitatory and
inhibitory neurons
87 µm
134 µm
We first reconstructed axons of pyramidal cells

108 µm and INs in mouse and human (Fig. 3, A to C)
to determine their synaptic preferences. The
V2 68 µm
96 µm
68 µm rate of spine innervation showed the well-
~100 M years
164 µm
ACC documented dichotomy in mouse (Fig. 3, A and

B) (49, 55, 60): Pyramidal cell axons targeted
A2 dendritic spines with 61.47 ± 9.26% (mean ± SD)
(temporal
105 µm cortex) of their output synapses (n = 11 axons, n = 367
output synapses), whereas INs only very rarely
Fig. 1. Comparative connectomic analysis of mouse, macaque, and human cortex. Dense connectomic made an output synapses as the only synapse
reconstructions from L2/3 of five cortical areas of mouse (bottom, n = 5 individuals) and from four cortical onto a dendritic spine (n = 1 of 263 IN output
areas of macaque and human (n = 3 individuals). There are matched cortical areas (A2 and STG) across synapses in mouse), with no overlap between
all three species and paired samples from S1 (mouse and macaque). A total of 202,954 axons and 1,618,129 the populations. In human, however, axons re-
synapses were analyzed (Materials and methods). The raw 3D EM data of mouse datasets S1, V2, PPC, and constructed from the cell bodies of pyramidal
ACC were previously published (55), but their dense reconstruction have not. (Left) Simplified phylogenetic cells made only 30.37 ± 16.16% (mean ± SD)
tree [based on (100)] indicating time to last common ancestor between human (Homo sapiens), rhesus of their output synapses onto dendritic spines
macaque (Macaca mulatta), and mouse (Mus musculus). Scale bars apply to the brain sketches. S1, primary (n = 15 axons, n = 1126 output synapses),
somatosensory cortex; A2, secondary auditory cortex; V2, secondary visual cortex; PPC, posterior parietal whereas INs maintained their almost com-
cortex; ACC, anterior cingulate cortex; STG, superior temporal gyrus; IFG, inferior frontal gyrus. plete exclusion of single spine innervation (Fig.
3, B and C) [n = 12 axons, n = 345 output
synapses, of which n = 1 were single spine
(Fig. 2C): If all other circuit properties were pyramidal cells remained largely constant synapses; IN axons did, however, innervate
unchanged between mouse and human, the from mouse to macaque and only slightly dendritic spines as double innervations, in
inhibitory-to-excitatory balance (i/e balance) increased in human [0.16 ± 0.13 per microm- all species, as reported for subtypes of INs (fig.
would also be 2.5-fold shifted toward inhibi- eter of dendritic shaft length versus 0.17 ± 0.11 S3) (61–63); only the single spine innervation
tion compared with that in mouse. and 0.21 ± 0.08; n = 46, 36, and 49; P = 0.6 and rate was used for the following analyses].
P = 0.003 for mouse to macaque and mouse Thus, the distributions of synaptic target prop-
Synaptic inputs to pyramidal cell dendrites to human, respectively; Kolmogorov–Smirnov erties between pyramidal and IN axons recon-
We therefore first analyzed the synaptic input (KS) test; mean ± SD] (Fig. 2, D and E), structed from their cell body of origin were
to pyramidal cell dendrites in mouse, macaque, synapses onto dendritic spines were 2.7-fold less separated in human (Fig. 3B) and indi-
and human (Fig. 2, D to H). Although the sparser in macaque and human [1.99 ± 0.65 cated a change in axonal synaptic preference
density of synapses onto the dendritic shaft of per micrometer of dendritic shaft length from mouse to human.

Fig. 2. Multifold changes in A B

cellular composition and synap-
Fraction of interneurons (of all neurons) (%)

40
tic input of mouse versus
macaque and human cortex.
87
(A and B) Reconstruction of ExNs µm 30
(magenta) and inhibitory INs
(white) shows 2.54-fold expansion 152 µm
68 µm 20
of the IN pool from mouse to
216
macaque and human, which µm
(C) would predict a similar-fold Pyramidal (excitatory) 10

increase in the inhibitory input neurons, ExNs Interneurons
onto ExNs, which would substan- (INs)
0
tially alter the set point for the 112 µm
166 µm Mouse Macaque/
inhibitory versus excitatory syn- Mouse (V2) Human (STG) Human
aptic input balance in human
compared with mouse (illustrated
C D Synapse onto E
Input synapses along pyramidal cell dendrites

for threefold increase). (D to Pyramidal
dendritic spine
neurons 4
F) Mapping of (D) the synaptic 3-fold
interneuron Synapse onto
Synapse onto
input to excitatory neuron den- fraction dendritic shaft
(per µm dendrite length)

dendritic spine
3
drites showed (E) an unaltered
rate of shaft input synapses but a Inter- Synapse
neuron 2 onto
2.65-fold decrease of spine input dendritic
Pyramidal
synapses, resulting in (F) a IN axon
neuron
shaft
(ExN)
2.72-fold increase in the fraction ExN
dendrite
1
of input synapses made onto Spine Pyramidal 20 µm

dendritic shafts. (G) Other Shaft
synapse
2 µm neuron
(ExN)
0
S1 V2 ACC S1 STG
A2 PPC STG IFG
dendritic inputs were rare but Mouse Macaque
Human
slightly increased from mouse to Mouse Human Mouse (A2) Human (STG)
macaque and human: doubly

innervated spines, 4.49 ± 0.01 F Synapse
G
onto spine
versus 4.68 ± 0.01 and 6.91 ± Onto spines
Fraction of synapses along ExN dendrite (%)
*
100
0.01%; stub synapses, 1.44 ± *
*
0.00 versus 2.17 ± 0.01 and *
80
*
Double spine innervation Synapse onto dendritic Synapse onto spine neck
4.71 ± 0.01%; spine neck innerva- stub
tions, 0.26 ± 0.00 versus 60
H
Fraction of synapses onto
Fraction of synapses onto

0.84 ± 0.00 and 0.67 ± 0.00%; 8
ExN dendrite (%)
Synapse Double 20
ExN dendrite (%)

n = 1111, 598, and 1040 total onto shaft 40 Onto shafts
synapses, respectively. [(F) and 4
Stub
(G)] (Insets) EM images of exam- 20

Neck
10 Shaft
ple synapses from (F) human STG * synapses
0 0
multiSEM-imaged and (G) (left S1 A2 V2 PPC ACC S1 STG STG IFG
S1 V2 ACC
A2 PPC
S1
STG
STG
IFG 0
2 µm
to right) macaque STG, human Mouse Macaque Human Mouse Macaque Mouse Macaque/
Human Human
IFG, and macaque S1, respectively.
(H) Concomitant increase of shaft synapse input could support the altered i/e balance model in (C), if axonal properties remain unchanged from mouse to macaque
and human (analyzed in Figs. 3 and 4). Scale bars, 1 mm unless indicated otherwise. Data in (A) are from automated reconstructions, and data in (B) and (D) to (H)
are from expert reconstructions.
In the dense cortical neuropil, only a frac- not reproduce the soma-based axonal proper- bias axonal reconstructions toward local (not
tion of axons originates from neurons whose ties but showed a broader distribution of spine distally located) neurons and to the proximal
cell body is located in close proximity, whereas innervation from 0 to 80% spine preference parts of these neurons’ axons. When analyzing
most have more distal origins. To study the (here and in the following, spine preference the distance-dependence of axonal targeting
target properties of all axons in the neuropil, implies fraction of an axon’s synapses estab- properties for pyramidal axons from human
also those without their cell body of origin lished as single spine innervations, not con- (Fig. 3, G to I), we found evidence for a sub-
in the dataset, we next analyzed the dense sidering doubly innervated spines). stantial change from proximal axonal preference
population of axons in mouse versus macaque What could yield this difference in axonal of shaft innervation (64–66) to distal preference
and human that we obtained from the dense properties between soma-based and dense for spine innervation along the axon’s path [see
automated reconstructions (n = 202,954 axons reconstructions in macaque and human? In (67) for a first report of such a path length–
total with n = 1,618,129 output synapses) (Fig. particular, we needed to understand the origin dependent synaptic sorting phenomenon along
3, D to F). Mouse axons predominantly in- of axons with >40% spine targeting in ma- axons in the mammalian cerebral cortex, and
nervated spines (Fig. 3E), as expected from the caque and human (Fig. 3E) that we had rarely (68, 69) for earlier reports on the bird brain].
soma-based reconstructions (Fig. 3B) and the observed in the soma-based reconstructions We had to take this effect (Fig. 3, B, E, and I)
small fraction of INs (Fig. 2B). Densely sam- (Fig. 3B). Soma-based reconstructions in data- into account when comparing the axonal
pled macaque and human axons, however, did sets smaller than the complete axonal arbor properties in these species (Fig. 4).

Fig. 3. Synaptic target proper- A Mouse B 100 Mouse Human C Human
Fraction of output synapses onto

ties of excitatory and inhibitory
neurons in mouse and human. Interneuron
80 ExN axons
spines (per axon, %)

(A to C) Axonal target properties
from identified pyramidal cells and 60 Pyramidal
INs (n = 50 neurons, n = 2101 neuron
output synapses, cross-validated Pyramidal Interneuron 40

neuron
expert reconstructions). All
reconstructions are available at 20
100 µm
https://wklink.org/7881 and 100 µm
https://wklink.org/3103 (mouse), 0
IN axons
and https://wklink.org/9448
Axon cont.
and https://wklink.org/2204
Mouse Macaque Human
(human). (B) Absence of single D E S1 A2 V2 PPC ACC S1 STG STG IFG
F
spine innervation for INs in mouse Mouse (A2) 100 Human (STG)
Fraction of axonal synapses

and human, but shift in spine
made onto spine heads (%)

innervation of excitatory axons
from mouse to human, yields a
less separated synaptic prefer-
ence of excitatory versus inhibi-
tory axons. (D to F) Axonal target
properties from dense axonal
reconstructions in mouse,
macaque, and human. Examples 0 0 0.02 0 0.02 0 0.02 0 0.02 0 0.02 0 0.01 0 0.01 0 01 . 0 0.02
of dense axon populations from Axons (from dense automated reconstructions),
(D) mouse A2 and (F) human STG, Pyramidal
probability density
all axons traversing boxes of 10 G cell body H Axon I
by 10 by 10 mm size shown Pyramidal
cell body
Human
(n = 120 and n = 89, respectively).
Quantifications in (E) are based on 100
all axons in the respective data-
Fraction of axonal synapses

sets (n = 202,954 axons, with 5 to Onto dendritic shaft of IN
Output synapse onto.. Axon
Onto dendritic spine of ExN
Human
onto target (%)

25 output synapses each; ..dendritic shaft of IN
..dendritic shaft
n = 1,618,129 synapses). Gray of ExN
indicates distribution of spine
target fraction for all axons in the
datasets, uncorrected. Black lines ..dendritic spine
of ExN
indicate average likelihood
100 µm Onto dendritic shaft of ExN
function of true spine target frac- 0
0 500 1000 0 500 1000
tion under consideration of error Axon path length from soma (µm) Axon path length from soma (µm)
rates (Materials and methods).
There is a broadening of axon target property distribution toward lower spine targeting in macaque and human compared with mouse that is, however, less pronounced than
in (B) soma-based axon reconstructions. (G to I) Path length–dependent axonal synapse sorting (PLASS) as a possible origin of broader axonal target property
distributions in human. (G) Example of pyramidal cell axon from human STG [path length, 3.74 mm, n = 132 output synapses; same neuron as shown in (C)]. Colors
indicate synaptic target of axonal output synapses. [(H) and (I)] Distribution of axonal output synapses along pyramidal cell axons from human STG (n = 15 axons,
1126 output synapses, expert reconstructions, n = 12 reconstructions from mSEM dataset shown). All axons from mSEM and SBEM datasets are available at
https://wklink.org/9448 and https://wklink.org/2204. Synapse symbols are as in (G). (I) Substantial increase in targeting of ExN spines over axonal path length. Data in
(A) to (C) and (G) to (I) are from expert reconstructions, and data from (D) to (F) are from dense reconstructions.
Excitatory versus inhibitory synaptic input balance and human (Fig. 4A; compared with Fig. 3, B has to be composed by the intermixing of the
We first identified synapses onto a dendritic and E). Because this data did not allow the spine preference of the various axon types
shaft of distal pyramidal cell dendrites and simple threshold-based classification of axons present in the neuropil and similarly for other
reconstructed the presynaptic axon from these into inhibitory versus excitatory that can be types of synapses. Because dense 3D EM
synapses (“shaft-seeded” axons). Then we de- used in mouse, we needed a more rigorous data allowed us to concomitantly measure the
termined these axons’ synaptic target proper- approach for axon type determination (Fig. properties of extended stretches of axons and
ties on the basis of their other output synapses 4, C to E). dendrites together with the dense volumetric
(Fig. 4A). Whereas in mouse, as expected, axons We used the notion that the various types synaptic composition of the neuropil, we could
were clearly identifiable as excitatory versus of axons and dendrites and their synaptic determine the occurrence and target proper-
inhibitory on the basis of their dichotomy in output and input properties in dense cortical ties of excitatory and inhibitory axons without
spine targeting preference (Fig. 4A), the dis- neuropil are ultimately constrained by the syn- having to make prior assumptions about their
tributions of axonal properties were much less aptic composition of this neuropil volume relative prevalence (Materials and methods).
distinct in human, recapitulating the proper- (Fig. 4B). The volumetric density of synapses We built a model in which the synaptic
ties of densely reconstructed axons in macaque onto dendritic spines in the neuropil, for example, properties of excitatory and inhibitory axons

Fraction of
ExN dendrite
A Axon seeded at .. B 1 ..in volume
input data
Fraction of
ExN dendrite .. Shaft spine synapses .. ..along ExN 1 0 1
2 dendrites 100
Fraction of spine
..along
synapses (%)
2 µm 3 spine-seeded
axons
.. Spine
Mouse Human 4 ..along ExN shaft-
seeded axons
5 IN dendrite
Fraction of spine synapses (%)

100
..along IN-shaft-seeded axons
Fraction of output synapses
onto spine along axon (%)
5 4 3 0
80 100 100
60
Axons
seeded
40 at spine
Axons
20 seeded 0 0 S1A2V2ACC S1 STG
at shaft S1A2V2ACC S1 STG S1A2V2ACC S1 STG S1A2V2ACC S1 STG
PPC STG IFG
PPC STG IFG PPC STG IFG PPC STG IFG 2
Mouse MQ Mouse MQ Mouse MQ Mouse MQ
0 Human
Human Human Human
C Mouse D Human E
Synaptic preference of axons (model fit)
ExN axons IN axons Axons seeded ExN axons IN axons Axons seeded
Inhibitory input balance i/(i+e), (%)

40
synapses synapses at... synapses synapses at...
onto.. onto.. onto.. onto..
from IN fraction increase

100 ..Spine 100 ..Spines ..Spine
10 10
Fraction of synapses (%)
30
Fraction of synapses (%)
..Spines ..ExN All synapse types
Prediction
shafts
Shaft & Spine only
0 0
..ExN shaft 20
Spine synapses
Spine synapses ..ExN shaft

10 ..IN shafts 10
.. IN
p(inhibitory|axon)
1
shafts ..ExN 10 Max
.. ExN 0 0
.. IN ..IN shaft shafts ..IN shaft
shafts 10 .. ExN 10
shafts
shafts
0 0 0
..IN shafts
S1 A2 V2 PPC ACC S1 STG STG IFG
..Spines 0 ..Spines 0
0 0 10
Shaft synapses
0 0 10 Mouse Macaque
0 PDF 15 0 PDF 4 Shaft synapses Human
Fig. 4. Detailed analysis of i/e balance onto ExN dendrites. (A) Target and black line indicates IN shaft targeting by ExN axons. Shown is a broadening
properties of axons seeded at dendritic spines and shafts of ExN dendrites in of ExN axons’ spine targeting fraction and shift of excitatory and inhibitory
mouse and human. (Top) Example reconstruction (macaque STG). (Bottom) shaft targeting from mouse to human. MQ, macaque. (E) Resulting estimates of
Data from expert reconstructions in mouse S1 and human STG (n = 53 axons, inhibitory input fraction [i/(i+e) balance] onto ExN dendrites in mouse, macaque,
n = 626 output synapses) showing shift of axonal targeting analogous to densely and human. Violin plots indicate expected inhibitory input synapse fraction along
reconstructed axons in mouse versus human (compared with Fig. 3E). (B to ExN dendrites (distribution across n = 1000 bootstrap samples per dataset).
D) Model incorporating dense volumetric synapse densities together with Open shading indicates only shaft and single spine inputs considered; gray
dendritic and axonal targeting properties, but not using any assumption about shading indicates that multiply innervated spines and other inputs are included
excitatory versus inhibitory synapses or axons. (B) Input data from expert (Fig. 2G). Synaptic input balance does not approach the inhibitory bias predicted
annotations (n = 754 axons, n = 11,308 synapses total) that [(C) and (D)] fully by the increased fraction of INs in macaque and human. Blue shading indicates
constrain the model (Results and Materials and methods) (model validation is prediction from mouse to macaque and human (24.9 ± 3.2%, mean ± SD;
provided in fig. S2). Shaded magenta curve indicates distribution of spine P < 0.001, by bootstrap sampling) (Materials and methods). All data are
targeting by ExN axons; magenta line indicates ExN shaft targeting by ExN axons; from expert reconstructions.
and their relative prevalence were determined spiny (ExN) dendrites, (iii) fraction of output STG datasets): point estimates for the fraction
(Fig. 4, C and D): Inhibitory axons were synapses onto spines along axons seeded from of synapses made onto ExN spines, ExN shafts,
modeled with a multinomial distribution ExN dendritic spines, (iv) fraction of output and IN shafts for inhibitory axons (Fig. 4, C
based on point estimates for their synaptic synapses onto spines along axons seeded from and D) and distributions of these target frac-
target preferences, whereas for excitatory ExN dendritic shafts, and (v) fraction of output tions for excitatory axons. Together with the
axons, a distribution of target preference synapses onto spines along axons seeded from modeled relative prevalence of inhibitory ver-
was modeled with a Dirichlet-multinomial IN (smooth) dendritic shafts. These input data sus excitatory synapses, for any axonal stretch
to account for the effects of broadened synap- did not require a pre hoc classification of the with any combination of x synapses onto shaft
tic preference shown in Fig. 3 (Materials and synapses or axons as inhibitory versus excit- and y synapses onto spine, we obtained a
methods). The model was fully constrained by atory and did not contain the axons recon- probability of this axon to be excitatory versus
the following input data, which we obtained structed from identified cell bodies (Fig. 3, A to inhibitory (Fig. 4, C and D, right). For valida-
from carefully curated expert reconstructions C). With this input data, we obtained fits of tion of the model, we used the soma-based
in all datasets (n = 11,308 synapses annotated) the synaptic target properties of excitatory axon reconstructions (Fig. 3, A to C) for which
(Fig. 4B): (i) volumetric fraction of spine synapses, and inhibitory axons in each dataset (Fig. 4, we had certainty about their excitatory versus
(ii) fraction of input synapses onto spines along C and D, shown for mouse S1 and human inhibitory property, sampled local stretches

from these ground truth axons, and asked the axons increased from 56% (Fig. 5, F and G) (70 sity in the human cortex (Fig. 2E), we won-
model to predict their excitatory versus inhib- IN targets out of 126 synapses, n = 4 axons) to dered what the quantitative input structure of
itory character (E→I and I→E misclassification 72% (Fig. 5G) (225 IN targets out of 314 syn- human pyramidal cells would be (Fig. 6, A and
rates: 0 and 0% for mouse, 6.9 and 8.4% for apses, n = 9 axons, P = 0.018) (Materials and B, and figs. S1 and S4).
primates, respectively) (fig. S2 and supplemen- methods) (63). The changed IN pool compo- We used our large-scale 3D EM datasets ob-
tary materials, materials and methods). sition together with the changed distribution tained in one human individual that spanned
We then applied the model to determine of IN targeting properties would predict a the entire depth of cortical gray matter to re-
the expected inhibitory versus excitatory syn- substantial increase of IN-preferring inhibitory construct a substantial portion of L2/3 pyrami-
aptic properties of ExN dendrites in all data- axons in the dense neuropil of human versus dal cells. On the basis of these reconstructions,
sets (bootstrapped to account for sample size in mouse (Fig. 5F). When analyzing the target we extrapolated to the full extent of the dendritic
input data and controlled for initial conditions; preferences of inhibitory axons for smooth (IN) tree. The large dataset extent in-plane allowed
Materials and methods) (Fig. 4E). We found that dendrites in mouse versus macaque and human us to map some pyramidal cell dendrites in their
the inhibitory input balance increased only from the dense automated reconstructions (Fig. full extent from soma to dendritic tip and to
moderately from 9.3 ± 0.8% (mean ± SD) in 5H), we found a substantial shift toward IN-to- use the properties of these completely mapped
mouse to 13.8 ± 1.4% (P = 0.001, bootstrap IN connectivity that can account for the in- dendrites for estimating the total path length
sampling) in macaque and human, revoking hibitory synapses contributed by the expanded of pyramidal cell dendrites in human (Materials
a setting in which the inhibitory-to-excitatory IN pool in macaque and human (average and methods). The resulting estimates of total
input balance were to change in proportion to smooth dendrite targeting probability of dendritic path length (about 9 to 20 mm path
the 2.5-fold increase in IN fraction from mouse inhibitory axons, 8.0 ± 15.0% in mouse versus length) (Fig. 6B and fig. S4) were consistent
to macaque and human (Fig. 4E; compared 21.4 ± 29.0% in human; n = 6565 versus with light-microscopic reconstructions (43, 45).
with Fig. 3C) (P = 0.003; when also consider- n = 2048 axons; P = 2.2 × 10−104, one-sided KS Because we measured synaptic input density
ing all other types of input synapses, P < 0.001; test). Together with the increased IN fraction in parallel, we could exclude a compensation
bootstrap sampling; Materials and methods). (Fig. 2B), this would estimate the IN-to-IN of lower dendritic path length by higher spine-
Rather, the increased fraction of shaft input network to expand 6.7-fold from mouse to targeting synapse density for individual pyram-
synapses coincided with a change in axonal human. On the basis of the model (Fig. 4), we idal cells. Together, total synaptic input to
targeting properties: Excitatory axons made estimated the expansion to be 8.6-fold (from pyramidal cells was 12,000 to 17,000 in human
0.7 ± 0.3% of their synapses onto ExN shafts in 1.0 ± 0.2% IN-IN connectivity in mouse to L2/3, which is far below an increase correspond-
mouse, but 12.0 ± 1.1% in macaque and human 8.6 ± 1.4% in macaque and human) (fig. S2F ing to the about threefold thicker cerebral cor-
(P < 0.001, bootstrap), and their IN shaft tar- and supplementary materials, materials and tex yielding about threefold larger pyramidal
geting changed from 4.4 ± 0.7% to 17.3 ± 1.7% methods). cell dendritic trees compared with those of
(P < 0.001, bootstrap), which is consistent with To better understand the contributions of mouse (Fig. 6B).
the automated axon reconstructions. types of INs to this enhanced network, we
This lack of enhanced inhibition onto ExN then analyzed the synaptic input balance onto Discussion
dendrites was also found for the proximal in- IN dendrites in MP and BP INs in mouse, The comparative analysis of mouse, macaque,
put domains of pyramidal cells [axon initial macaque, and human (Fig. 5, I to K). To our and human cortical neuronal networks revealed
segment (AIS), soma, and proximal dendrites] surprise, already in mouse, MP INs and BP INs that the most substantial changes in neuronal
(fig. S1), excluding the possibility that inhibi- have different inhibitory input balance (Fig. 5, architecture—the increase of pyramidal cell size
tory synapses had been redistributed toward I and K): Whereas MP INs receive 7.6 ± 2.5% and the numeric expansion of the IN pool—have
the perisomatic domains. inhibitory input (bootstrapped mean ± SD; not resulted in the most immediate possible
These data yielded the question of where n = 105 axons from n = 5 mouse datasets, circuit changes: neither an increase of total
the expanded inhibitory population in human n = 977 synapses total), BP INs receive 26.2 ± 5.0% synaptic input on pyramidal cells nor an over-
is establishing its synapses. (n = 88 axons, n = 694 synapses, P = 0.001). In all shift of their synaptic input balance toward
macaque and human, MP INs receive inhib- inhibition. Rather, we found an at least six- to
Properties of the expanded inhibitory neuronal itory input that is commensurate to the frac- eightfold expanded IN-to-IN network in the
network in human tion of INs in the cortex (24.2 ± 6.4%, n = 69 macaque and human cortex (Fig. 6C and fig.
To determine the inhibitory network proper- axons from n = 4 datasets, n = 514 synapses; S2F). These circuit alterations point toward IN-
ties in mouse versus human, we first applied a P = 0.6 for scaling by IN fraction increase), to-IN connectivity as a key evolutionary change
simple IN classification based on the config- whereas BP INs receive further enhanced from mouse to primates, including human.
uration of IN dendrites as multipolar (MP) ver- inhibition (44.3 ± 7.8%, n = 71 axons, n = 506
sus non-MP [which included bipolar, bitufted, synapses, P = 0.026). These data indicate dif- Dependence of spine densities on age
and vertically oriented dendritic trees (Fig. 5, A ferential inhibitory targeting across species and Our finding of an almost constant total syn-
to D), labeled as “bipolar” (BP) for simplicity]. may imply separate IN-to-IN circuits to be sub- aptic input to human pyramidal cells when
The IN pool changed from dominated by MP stantially enhanced in macaque and human. compared with those of mouse may be af-
INs in mouse (about 70 to 82% MP versus 18 fected by a reported age-dependent decline of
to 30% BP, n = 52 versus 17) (Fig. 5D) to a Excitatory synaptic network dendritic spines in cortex (70–72), which could
majority of BP INs in macaque and human Last, we wanted to determine the magnitude of amount to a reduction in spine density of 45 to
(53% BP; n = 122 MP versus n = 135 BP, P < 0.01; the synaptic input to pyramidal cells in human 48% during puberty (72). The human samples
a test is described in Materials and meth- cortex. With the expanded size of pyramidal were from individuals in the upper quarter of
ods) (Fig. 5D). When mapping the synaptic cell dendritic trees, it has been assumed that expected life length. However, we found low
target properties of these classes of INs in human pyramidal cells receive a larger num- spine densities in macaque as well, at about
mouse versus human (Fig. 5, E and F), we ber of input synapses than that of mouse py- 50% life expectancy. Similarly, rodents of 50
found a fraction of BP INs with almost ex- ramidal cells (from about 10,000 to about to 80% life expectancy show spine densities
clusive innervation of other INs in human that 30,000) (15, 29, 43, 44, 46). Because we found a comparable with those reported here for youn-
was not found in mouse. IN targeting by BP IN strongly reduced spine-targeting synapse den- ger mice (55, 67). It is therefore unlikely that

A Mouse B C Human (STG) D
Fraction of Bipolar Interneurons

Bipolar 80
Inter-
(of all interneurons, %)

neurons
S1 60
40
20
A2
Multipolar 0
S1 V2 ACC S1 STG
Interneurons A2 PPC STG IFG
Mouse Macaque
Human
e in
inin e e e
ee e
in
in
e
in
in
in in
e
in e
e
ee ee Inhibitory axons
E F Macaque/ G H
ee eein
in
inin
in in
in inein
in in inin
in
in
in
in
in inin inin
in
in
in ee e e
in
ee ein
e eee
ee
e
eee ee e e e
e
in
e
e
e
Macaque/ (dense, measured)
IN-to-IN synapses (% per inhibitory axon)

in in e in
e e
ein
in
Mouse 100
Human Mouse Human
in e eee
ein
e ininin in ee
in
in
e ine
eeee
e
e
e in
eeee ee e
e
ee BP
ininin
inin
inin in
e
e e
in
e
e
e
in
ee
e
ee e
ee
in
100 Macaque/
in e einee e e
in ee ein ee e e
in
in e
eee ineee e
e
Human
(% per IN, bootstrapped)

in e ee e
ee e e ein
BP
in ee
inininin
MQ inin
ininin
in in
e eeeeein
in
e
ee
in
eeee
e
e
e in
e e in e
ee
80
(% per interneuron)
in e ee eee in
in
IN e e e e
80
IN-to-IN synapses
ee e
inin e
S1 e
e in
e
eee
e
e
in
e
IN-IN synapses
ine e
e
in e in
eeeeeee e ee
e
ExN e e
e
e
eee ee
ee
50 µm e
60
ine
e
in in
in
e
e
e
in
60
in e in in
e
BP MP
in
Synaptic targets ≥20

e
e
in e
40
Inhibitory axons
e
in
in e e
e
e
Mouse 40
in e
in e
in e
e
e
ee e
20 MP
(%)
Hum in
in
inin
in e
e
in
e e
ee
e
e
e
STG
in
in
in
in e e
ee
ee
e
e e
ee
20
in
in in
in
in
e
in
e
0
in
in
in
in
e
in
in
e
e
0 0.6 0 0.6 0 0.4 0 0.2 0 0.6 65 81 74 78 71 29 29 48 49 0
e e in
0
Inhibitory axons
e
in
INs
in e
S1 A2 V2 ACC STG IFG
in e
e e
Interneurons PPC S1 STG

(dense, predicted)
Mouse Macaque/
I J 100 K Human
80
Inhibitory input fraction of IN dendrites
BP IN
Inhibitory input fraction i/(i+e), (%)

MP IN 80
Mouse 20 µm 2 µm 60
(PPC) BP IN
60
i/(i+e) (%)
40
40
Human
(STG) 20 20
p(exc|syn)
1 Excitatory
Synapse
0.5
0 Inhibitory
0 MP IN
S1 A2 V2 PPC ACC S1 STG STG IFG 0
Mouse Macaque &
Mouse Macaque Human
Human
Fig. 5. Selective expansion of IN types and its effect on the inhibitory- of IN dendrites (n = 94,391 synapses, n = 11,384 axons) by inference of
to-inhibitory network. (A to C) Dense reconstruction of all INs in (A) the most likely smooth dendrite targeting probability under consideration of
mouse and (C) human with (B) labeling of MP (purple) versus BP (cyan) INs the error rates of automated synapse detection. (F) Expansion of IN-to-IN
showing sparsity of BP INs in (A) mouse compared with (C) human. connectivity as predicted from soma-based reconstructions. (I to K) Detailed
(D) Proportion of BP INs 2.3-fold expanded from mouse to macaque and analysis of inhibitory input balance to MP versus BP IN subtypes across
human. (E) Synaptic targets of BP versus MP INs in macaque and human species. (I) Example reconstructions of input synapses onto IN dendrites.
show selectivity for (left) inhibitory (93% of synaptic targets) versus (right) (J) Inhibitory input determined from the model in Fig. 4 for IN
excitatory (87% of synaptic targets) postsynaptic partners. (F) Broader dendrites. The 2.35-fold increase from mouse to human (18.9 ± 2.5%
distribution of IN innervation selectivity in macaque and human versus to 44.5 ± 7.3%, mean ± SD) is consistent with the prediction from IN
mouse from soma-based axon reconstructions. This predicts expansion of expansion for IN dendrites (P = 0.662). [(J) and (K)] Separate analysis of
inhibitory-to-inhibitory connectivity in macaque and human versus mouse inhibitory input to MP versus BP INs reveals a difference in inhibitory
(right). (G) Within the BP INs, IN targeting is further enhanced (from input already in mouse (7.6 versus 26.2%; P = 0.001) that is further
56.63 ± 4.46% to 71.67 ± 2.48%; P < 0.05, bootstrapped from 126 and enhanced in macaque and human (24.2 versus 44.3%; P = 0.026). Data
314 synapses, respectively, for mouse versus macaque and human). in (H) are from dense reconstructions; all other data are from expert
(H) Analysis of densely reconstructed inhibitory axons and their targeting reconstructions.

Fig. 6. Scaling of pyramidal cell A Pia Pia B

synaptic input and network Layer 1
Mouse Human
Layer 1
properties from mouse to Layer 2
Brain
human. (A) Large-scale 3D EM Complete L2/3 volume
pyramidal cell 3
(mm )
dataset spanning all cortical layers Layer 3 Layer 2
10 Basal 106
from human cortex (STG) for Dendritic dendrites
reconstruction of dendritic arbors path Apical tuft

Layer 4
length Apical 105
of L2/3 pyramidal neurons. (mm) 1 obliques
Layer 3
(B) With about threefold reduced Layer 5 Pyramidal
cell
spine density from mouse to Human 10 4
total input
human, the about threefold cortex 500 µm Human STG Synapse 1

Spine
Layer 6 ~1.7 x 2.1 x .03 mm density synapses Number
increase in pyramidal cell den- (per µm) of neurons
103 entire brain
dritic path length yields only Excitatory IN-to-IN Shaft (million)
synapses
moderately increased number of C neuron (ExN) connections 0.1
IN-to-ExN
input synapses for human pyram- connections 102
Total input Pyramidal
idal cells compared with that in 10,000 Spine
cell
total dendritic
synapses
mouse. Dashed lines indicate Input 10
length (mm)
Shaft
macaque. (C) Expanded IN-to-IN synapses synapses Cortical
thickness
network, illustrated according to (mm)
1,000 1
model fit (Fig. 4; fig. S2F; and Mouse Human Mouse Human
supplementary materials, Interneuron
(IN)
materials and methods). There is
an 8.6-fold expansion of IN-to-IN Mouse Macaque & Human
connectivity from mouse to
macaque and human. Data in (A) and (B) are from expert reconstructions. In (B), right, lower limits of pyramidal cell total dendritic length and total input are based on
(45); upper limits are based on data in this study. Brain volume is based on (101), and the number of neurons in the entire brain is based on (1).
age-related effects have dominated the finding of working memory through enhanced IN- All patients had given their written informed
of reduced spine densities in human com- to-IN connectivity and the ensuing network consent.
pared with mouse. Also, although temperature- dynamics (83). Our data indicate that a de- The macaque brain tissue sample was col-
dependent changes in spine densities have tailed investigation of such phenomena is re- lected at German Primate Center (DPZ) GmbH,
been reported (73), these are unlikely the cause of quired for an understanding of the human Göttingen.
our measured spine densities (Fig. 2E and sup- cortex. In particular, alterations in IN-to-IN
plementary materials) (29, 44, 46, 55, 66, 74–77). connectivity should become a focus of study in Tissue extraction and preparation
the context of possible pathological alterations Mouse tissue was processed and imaged as
Synaptic strength versus synaptic number of human cortex. described previously (49, 55, 84). These con-
Our data indicate the maintenance of the rel- sisted of four 3D-EM datasets from layers 2/3
ative number of inhibitory versus excitatory Materials and methods of mouse S1, V2, PPC and ACC sized between
input synapses on the dendrites of pyramidal Animal experiments and human tissue samples 72 by 93 by 141 mm and 88 by 56 by 213 mm
cells in human versus mouse. The effective bal- All animal-related experimental procedures were (Fig. 1) acquired at a voxel size of 11.24 to 12 ×
ance between inhibition and excitation could performed according to the law of animal ex- 11.24 to 12 by 28 to 30 nm the present study
be altered by different strengths of synapses. perimentation issued by the German Federal densely reconstructed and reanalyzed these
Evidence for larger unitary excitatory post- Government under the supervision of local published datasets from mouse, together with
synaptic potentials and for larger synapses and ethics committees and according to the guidelines a newly acquired dataset from mouse A2 cor-
presynaptic vesicle pools has been found in ex- of the Max Planck Society. Experimental proce- tex. The published mouse dataset “PPC-2” (55)
periments on human cortical slices (16, 30, 78, 79). dures were approved by Regierungspräsidium covering layers 1 to 5 was also used for expert
At the same time, the impact of single inhibitory Darmstadt, AZ: F 126/1002 (mouse) and reconstructions.
synapses is enhanced (19, 20). Whether synaptic Regierungspräsidium Marburg AZ: V54 -19c The human and macaque samples were
number is a proper indicator of overall synaptic 20 15 hours 01 MR 13/1 Nr. 78/2012 (macaque). processed as follows. All tissue specimen were
impact is a matter of investigation, with evi- The human brain tissue samples were col- fixed by either immersion (human) or transcar-
dence in ferret (80) pointing toward a dominant lected during neurosurgical procedures that dial perfusion (macaque) using an EM fixative
effect of synapse number, as we quantified in were indicated for medical reasons and inde- composed of 2.5% paraformaldehyde (Sigma),
this work. pendently from this research project at the 1.25% glutaraldehyde (Serva) and 2 mM calcium
Department of Neurosurgery at the Klinikum chloride (Sigma) in 80 mM cacodylate buffer
Increased complexity of inhibitory networks rechts der Isar of the Technical University of adjusted to pH 7.4 with an osmolarity ranging
The IN-to-IN network expansion found in ma- Munich. They were obtained from access tissue from 700 to 800 mOsmol/kg (85).
caque and human could have a multitude of (presumably healthy brain parenchyma that The human “H5” tissue was obtained from
dynamic effects. Although the most immedi- had to be removed as part of the procedure the right superior temporal gyrus of an 80 year
ate consequence could be a more evolved dis- and would have been discarded otherwise) old male patient during resection of a tempo-
inhibitory network capability—for example, before removal of the respective target lesions, ral mass lesion (final diagnosis: glioblastoma
for the gating of otherwise inhibited excitatory as approved by the Ethics Committee of the multiforme). After removal the sample was im-
activity (81, 82)—theoretical studies have also Technical University of Munich School of Med- mersed in cold (13°C) EM fixative and trans-
indicated possible effects on the maintenance icine (Ethikvotum 184/16S and 273/21 S-EB). ported to a nearby laboratory (transport time

about 6 min). Partially submerged in cold fix- coronal sections. The sections were subsequently Image alignment for the two multiSEM
ative the sample was manually trimmed along immersed in cold EM fixative, transported to the datasets from sample H5 were performed sim-
the pia-WM axis and mounted on a vibratome research laboratory and kept at 4°C over night. ilar to (55), following the alignment routines in
stage. Then, submerged in cold EM fixative, the Within 48 hours the tissue was further dissected (89) and https://github.com/billkarsh/Alignment_
sample was cut into 500 mm thick slices with a (with storage in 4°C EM fixative between pro- Projects, with modifications.
vibratome (Leica VT 1200S). The slices were cessing steps). A ~5 mm wide tissue block cen-
then transported in 8 to 12°C cooled fixative tered on the anatomically defined arm/finger Cell type classification
and stored over night at 4°C. The next day, region of the left postcentral gyrus (Area 3a/b) For analyses in Fig. 2, A and B, cell bodies were
samples spanning the entire cortical depth medially adjacent to the rostral end of the manually annotated by an expert annotator in
and about 1.5 to 1.7 mm in width were cut out intraparietal sulcus was dissected. The block webKnossos (87). All cell bodies were identi-
and prepared for electron microscopy as de- was placed in 0.15 M cacodylate buffer and fied and classified into pyramidal cell, IN and
scribed in (55) with the modification that the cut along the medio-lateral axis into 600 mm glia. Pyramidal cells were identified based on
sample was embedded in Epon Hard for sec- thick slices using a vibratome (Microm HM650V, presence of an apical dendrite directed toward
tioning in the ATUM. For this, samples were Thermo Scientific). Using a medical biopsy the pia, an AIS directed toward the white matter,
infiltrated through a graded series (3:1 for punch (KAI medicals, USA), a 1.5-mm-wide and spiny dendrites. INs were identified based
4 hours, 1:1 12h/overnight, 1:3 for 4 hours) of sample spanning almost the entire thickness on their large somata, which contained large
aceton and Epon resin (Epon hard mixture: of the cortex was cut and subsequently pre- numbers of mitochondria, an axon often exit-
5.9 g Epoxy, 2.25 g DDSA, 3.7 g NMA, 205 ml pared according to the methods described in ing from a dendrite, lack of a clear apical den-
DMP; Sigma-Aldrich). Samples were then in- (55) using an automated tissue processor (Leica drite or, if presence of an apical-like dendrite,
cubated in pure resin for 4 hours at room tem- EM AMW). lack of basal dendrites and WM-directed AIS,
perature, 12 hours/overnight at 4°, and another and smooth dendrites. Non-neuronal cells were
4 to 5 hours at room temperature. Samples were 3D EM imaging and image alignment distinguished primarily by their smaller cell
directly embedded in pure resin on aluminum SBEM datasets of the human (H5, H6), macaque bodies and different nuclear shapes.
pins and kept in a pre-warmed oven (60°) for and mouse (A2) samples were acquired using a For distinction of MP versus BP INs (Fig. 5,
2 to 3 days. Fresh resin was prepared for each custom-built SBEM microtome [(50) courtesy A to D), the dendrites of all identified INs were
incubation step. of W. Denk] mounted inside the chamber of a reconstructed and inspected in the coronal
The cured sample was trimmed into a hex- scanning electron microscope (FEI Verios, and tangential plane (90–98). INs with short
agonal shape (size 3.1 by 1.8 mm) with a dia- Thermo Fisher Scientific, USA). The image dendritic lengths were excluded from this
mond milling head using an EM trimmer (Leica acquisition and SBEM microtome were con- classification.
EM TRIM2, Leica Microsystems, Wetzlar, trolled using custom written software (55).
Germany). Next, the sample was cut into 35- Focus and stigmation were adjusted using cus- Dendrite reconstructions
to 40-nm-thick slices at 0.3 mm s−1 cutting tom written auto-correction routines. Imaging Pyramidal and IN dendrites (Fig. 2, D to G)
speed using a 4 mm ultra35° knife (DiATOME, parameters were as follows: 4 by 6 (macaque, were reconstructed by an expert annotator by
Nidau, Switzerland). The ultrathin sections were H5, H6) or 3 by 4 (mouse A2) mosaic tiles of following their trajectory throughout the dataset
collected on plasma-treated, carbon coated images sized 4096 by 3536 voxels with an in- volume and placing nodes as described previ-
Kapton tape (custom-processed) with a cus- plane voxel size of (11.24 nm)2 and 30 nm nom- ously in (55, 84). From these dendrites for which
tomized ATUM-tome (RMC Boeckeler, Tucson, inal cutting thickness at 400 ns (macaque) or the identity of the originating cell body had
USA) (51, 52). The tape was then mounted on 700 ns (H5, H6, A2) dwell time with a nominal been determined, a distal stretch of 3 to 49 mm
silicon wafers using double-sided adhesive car- beam current of 800 pA (macaque) or 400 pA length (mouse) and 7 to 44 mm length (macaque
bon tape (P77819-25, Science Services GmbH, (H5, H6, A2). and human) was used for annotation of all input
Munich, Germany). In total, 7009 slices (corre- Two multi-SEM datasets of sample H5 were synapses.
sponding to an extent of 270.25 mm) were cut. acquired as follows. In one experiment, 767
The human “H6” sample was obtained from slices (476 at 35 nm, 291 at 40 nm) were im- Annotation of input synapses on dendrite,
inferior frontal gyrus from a 69 year old female aged with a 61-beam multiSEM (multiSEM soma, AISs
patient during surgical removal of frontal mass 505, Carl Zeiss Microscopy GmbH, Oberkochen, Analyses reported in Fig. 2, D to G, and fig. S1
lesion (final diagnosis: glioblastoma multi- Germany) at a landing energy of 1.5 kV, a pixel were conducted as follows. For a given post-
forme). Following surgical removal, tissue was size of 4 nm and a pixel dwell time of 50 ns with synaptic target class (dendrites, somata, or AIS),
directly collected in fix solution kept at 4°C. an FOV per slice of 1.7 by 2.1 mm. In a second all input synapses were identified based on
The tissue was immediately sliced into 500 mm experiment (H5_ext), 1342 additional slices the presence of a presynaptic vesicle cloud
thin slices in cold fixative using vibratome. (thickness 35 to 40 nm, corresponding to about and postsynaptic density [as described in
Slices were kept at 4°C overnight. Samples were 54 mm extent) were imaged with a smaller FOV (55, 67, 84); see following section]. These
then collected using 1mm circular medical per slice (1.1 by 1.1 mm) and aligned together synapses were labeled as single spine when
biopsy punch (covering L2/3) and prepared with the previous experiment resulting in a only one presynaptic bouton was found for a
for SBEM as described in (55). total of 82 mm depth. dendritic spine; double spine when two input
The macaque sample was acquired from Image alignment for SBEM datasets was synapses were found for which a clear dis-
a 15.6 year old right handed male animal. performed using global 3D relaxation of shift tinction into primary (excitatory) and secondary
Transcardial perfusion was performed under vectors calculated by SURF feature detection (inhibitory) was not possible; primary spine and
anesthesia (Ketamin, Xylazin) after an addi- as in (86) followed by subimage alignment as in secondary spine when two input synapses were
tional lethal dose of Barbituate (90 to 120 mg/kg (49). The aligned image volume was then saved found for which this distinction was possible;
intravenously). After flushing with ~2 l of 0.15 M in the webKnossos (87) three-dimensional image neck for spine neck innervations, stubby spine
cacodylate buffer (pH 7.4, duration 15 min) the format. Human H5 (STG), H6 (IFG), macaque synapses when a short dendritic protrusion of
perfusion solution was changed to ~2 l of EM (STG), and mouse (A2) SBEM datasets were larger diameter than a spine neck and without
fixative. The brain was removed and the anterior aligned by scalable minds, Postdam (supplemen- clear diameter change at the end (no clear spine
half of the brain was sectioned into 5-mm-thick tary materials, materials and methods) (88). head) was synaptically innervated, and shaft

synapses when the synapse was clearly placed an expert annotator; for unclear cases, these The estimations for the lengths of the above
on the main dendritic process without noticeable were re-annotated for expert consensus be- compartments were done as follows. The length
protrusion. Distances of synapses from the soma tween two or three experts. for apical dendrite trunk compartment was
were measured using minimum spanning tree estimated by averaging the lengths over all
on the annotated nodes of the dendrites (see Volumetric model of synapse and axon types, the samples from their soma exit until the
“skeleton” class in code repository). inference of synapse and axon types, automated main bifurcation. For each of the remaining
reconstruction, and error analysis compartments (i.e., apical tufts, oblique and
Soma based axon reconstructions These methods are reported in the supplemen- basal dendrites), the path lengths (PLs) be-
Analyses reported in Figs. 3, A to C and G to tary materials. tween consecutive branch points (BPs) were
I, and 5, E and F, were conducted as follows. measured until a true end was reached. The
The axons of identified pyramidal cell and INs Estimates of total dendritic path length of order of a branch point (“order”) was defined
were reconstructed by first identifying the exit human pyramidal cells as the number of edges along the unique path
at the AIS. Then the trajectory of the axon was For the estimation of the complete synaptic between it and the cell body node (for exam-
followed throughout the dataset and com- input onto a L2/3 pyramidal cell (Fig. 6, A and ple, order is 0 at the cell body node, 1 at first
ments were added at the outgoing synapses. B, and fig. S4), all dendrites of 10 L2/3 py- branch point, 2 at second branch point and so
The post-target of each synapse was further ramidal cells were reconstructed until their on). Then for the entire compartment (apical
classified into excitatory or inhibitory class end in dataset H5, which was either the actual tuft, oblique or a basal dendrite), the total length
based on their spine-targeting synapse den- ending of the dendrite in the neuropil, or the was calculated by summing the path lengths
sity (when target was dendritic) or cell body end of the dendrite at the dataset boundary of each segment times 2 to the power X order of
BP
type (when target was a soma or an AIS). (Fig. 6A). In addition, nine pyramidal cells in the corresponding branch point:
order¼0
Axons with fewer than 10 synapses were ex- the extended dataset H5_ext were analyzed PLsego rder 2 order
where order = order of a
cluded to allow higher signal-to-noise ratio with a larger fraction of in-dataset dendrite branch point, BP = mean number of branch
per axon. endings. Results from both datasets yielded points for each compartment and PLseg_order =
similar ranges for total dendritic path length path length of the succeeding segment at the
Synapse-seeded axon reconstructions estimates, as detailed in the following. corresponding branch point order. Then, the
Analyses reported in Fig. 4, A and B, were For the following calculations, only dendrites resulting total length of each compartment
conducted as follows. For a given postsynaptic with actual in-dataset endings were used (one was multiplied with the average number of
target with identified input synapses, a skele- exception were apical tuft dendrites in which basal, oblique or apical tuft exits to estimate
ton node was placed in the presynaptic axon’s some dendrites without in-dataset endings had lengths for all basal, oblique and apical tufts
vesicle cloud and commented as “seed” synapse. more branch points and were therefore in- respectively. The model of the branching as
The presynaptic axon was then reconstructed cluded in the estimate). This approach could binary to the final branch point order would
throughout the entire dataset volume, and all correspond to an underestimation of dendrite likely overestimate total dendritic path length,
of the axons’ other output synapses and their length; therefore, in addition to the length thus not affecting the conclusion about low
corresponding postsynaptic targets were iden- measurements described here, we also used total input synapse numbers in human.
tified as described in the following section. The length reports from light-microscopically im- To estimate the number of input synapses
“seed“ synapse was excluded when quantifying aged human L3 pyramidal neurons, which for each of the dendritic compartments, we
axonal spine target properties in Fig. 4A. provided similar path length measurements proceeded as follows. For each compartment
(ranges in Fig. 6B) (45). (apical trunk, apical tuft, oblique or basal den-
Synapse identification for reconstructed axons For each pyramidal cell, the following dendrite), dendritic segments of path length 10 to
For analyses reported in Figs. 3, A to C and G dritic compartments were distinguished: (i) 50 mm were sampled and all input synapses
to I; 4, A and B; and 5, E and F, the following apical dendrite trunk (measured from the exit were annotated. For computing the spine-
synapse identification was applied. For each at the cell body toward pia along the cortical targeting synapse density per dendritic shaft
reconstructed axon, synapses were identified axis, until the main bifurcation), (ii) apical tuft path length, both single spine head and pri-
manually when following the trajectory of axon. (measured from the main bifurcation point of mary spine head synapses were included. For
First, vesicle clouds in the axon were identified the apical dendrite), (iii) oblique dendrites shaft synapse density, only the shaft synapses
as accumulations of vesicles. Subsequently, the (measured from the exit at the apical dendritic were included. For total synapse density, ad-
most likely postsynaptic target was identified by trunk), and (iv) basal dendrites (all dendrites ditionally secondary spine head, spine neck
the following criteria: direct apposition with exiting at the cell body except the apical den- and dendrite “stub” synapses were included
vesicle cloud; presence of a darkening and slight drite) (55). (total fraction of these additional synapses:
broadening of the synaptic membrane; vesicles For the basal, oblique and apical tuft com- less than 10%). The average spine, shaft and
at close proximity to the plasma membrane at partment n = 226, 211, and 167 dendrites were total synapse densities were then multiplied
the site of potential synaptic contact. Synapses reconstructed of which n = 25, 28, and 32 with the total path lengths of the corresponding
were marked as uncertain whenever the signs dendrites with in-dataset endings were found compartments. The apical dendrite trunk was
of darkened postsynaptic density could not be (n = 21 cells). additionally distinguished into a proximal seg-
clearly identified. All analysis in this study were For the estimation of the average number of ment (up to 50 mm from cell body) and a distal
conducted only on synapses that had been branch points for the apical tuft compartment segment (50 mm from soma until the main
classified as certain. For each axon, at every dendrites without in-dataset endings were in- bifurcation). Similarly, the basal dendrites were
subsequent synapse location, a node was placed cluded to avoid an underestimation of den- additionally distinguished into a proximal seg-
in the vesicle cloud and commented with the drite length as some of them had more branch ment which showed high shaft synapse density
corresponding postsynaptic target’s identity points than the dendrites with in-dataset end- (up to 30 mm from cell body for mouse, 80 mm
(for example, soma, AIS, dendritic shaft, den- ings. For this, all apical tuft dendrites beyond a for human).
dritic single spine, dendritic double-innervated 300-mm threshold were included (n = 64 den- For the basal compartment we measured
spine, spine neck, stub, somatic spines or drites, with n = 32 in-dataset endings, n = 21 spine-targeting and shaft synapse densities
filopodium). All synapses were annotated by cells). of 0.66 ± 0.23 per micrometer and 0.23 ±

0.09 per micrometer (n = 13), for the oblique human brain: A review of 150 years of cell counting. J. Comp. somatostatin- and calretinin-immunoreactive axon terminals.
compartment 0.67 ± 0.23 per micrometer and Neurol. 524, 3865–3895 (2016). doi: 10.1002/cne.24040; Synapse 62, 456–465 (2008). doi: 10.1002/syn.20514;
pmid: 27187682 pmid: 18361442
0.19 ± 0.06 per micrometer (n = 14), for the apical 3. R. D. Hodge et al., Conserved cell types with divergent 24. N. Cano-Astorga, J. DeFelipe, L. Alonso-Nanclares, Three-
tuft compartment 0.73 ± 0.34 per micrometer features in human versus mouse cortex. Nature 573, 61–68 dimensional synaptic organization of layer III of the human
(n = 12) and 0.25 ± 0.07 per micrometer (n = (2019). doi: 10.1038/s41586-019-1506-7; pmid: 31435019 temporal neocortex, three-dimensional synaptic organization
4. F. M. Krienen et al., Innovations present in the primate of layer III of the human temporal neocortex. Cereb. Cortex
12), respectively. interneuron repertoire. Nature 586, 262–269 (2020). 31, 4742–4764 (2021). doi: 10.1093/cercor/bhab120;
For the proximal segment of the apical trunk doi: 10.1038/s41586-020-2781-z; pmid: 32999462 pmid: 33999122
we measured spine-targeting and shaft synapse 5. K. Letinic, R. Zoncu, P. Rakic, Origin of GABAergic neurons in 25. B. G. Cragg, Ultrastructural features of human cerebral
the human neocortex. Nature 417, 645–649 (2002). cortex. J. Anat. 121, 331–362 (1976). pmid: 179969
densities of 0.06 ± 0.14 per micrometer and doi: 10.1038/nature00779; pmid: 12050665 26. I. B. Yáñez et al., Double bouquet cell in the human cerebral
0.44 ± 0.16 per micrometer (n = 9), for the 6. S. R. y. Cajal, Estudios sobre la corteza cerebral humana. II. cortex and a comparison with other mammals. J. Comp.
distal segment until the main bifurcation Estructura de la corteza motriz del hombre y mamíferos Neurol. 486, 344–360 (2005). doi: 10.1002/cne.20533;
superiores. Revista Trimestral Microgáfica 4, 117–200 (1899). pmid: 15846784
0.45 ± 0.39 per micrometer and 0.20 ± 0.08
7. R. N. Kooijmans, W. Sierhuis, M. W. Self, P. R. Roelfsema, A 27. E. A. Nimchinsky et al., A neuronal morphologic type unique
per micrometer (n = 7), respectively. quantitative comparison of inhibitory interneuron size and to humans and great apes. Proc. Natl. Acad. Sci. U.S.A. 96,
distribution between mouse and macaque V1, using calcium- 5268–5273 (1999). doi: 10.1073/pnas.96.9.5268;
Statistical tests binding proteins. Cereb. Cortex Commun. 1, tgaa068 (2020). pmid: 10220455
doi: 10.1093/texcom/tgaa068; pmid: 34296129 28. M. R. del Río, J. DeFelipe, Double bouquet cell axons in the
All statistical tests were KS for fractions and 8. J. Berg et al., Human neocortical expansion involves human temporal neocortex: Relationship to bundles of
ratios, and Wilcoxon ranksum tests, otherwise. glutamatergic neuron diversification. Nature 598, 151–158 myelinated axons and colocalization of calretinin and
Data are shown as box or violin plots with (2021). doi: 10.1038/s41586-021-03813-8; pmid: 34616067 calbindin D-28k immunoreactivities. J. Chem. Neuroanat. 13,
9. T. E. Bakken et al., Comparative cellular analysis of motor 243–251 (1997). doi: 10.1016/S0891-0618(97)00050-1;
median, 25th and 75th percentile and outliers cortex in human, marmoset and mouse. Nature 598, 111–119 pmid: 9412906
(using MATLAB, Mathworks). All tests are doc- (2021). doi: 10.1038/s41586-021-03465-8; pmid: 34616062 29. R. Benavides-Piccione, I. Ballesteros-Yáñez, J. DeFelipe,
umented in the code repository (next section). 10. A. Shapson-Coe et al., A connectomic study of a petascale R. Yuste, Cortical area and species differences in dendritic
fragment of human cerebral cortex. bioRxiv, 446289 spine morphology. J. Neurocytol. 31, 337–346 (2002).
The test of significance for increase in IN [Preprint] (2021); .doi: 10.1101/2021.05.29.446289 doi: 10.1023/A:1024134312173; pmid: 12815251
fraction from mouse to macaque/human was 11. M. Field et al., Tonic GABAA receptor mediated currents of 30. V. Szegedi et al., Plasticity in Single Axon Glutamatergic
done by pooling all identified excitatory neu- human cortical GABAergic interneurons vary amongst cell Connection to GABAergic Interneurons Regulates Complex
types. J. Neurosci. 41, 9702–9719 (2021). doi: 10.1523/ Events in the Human Neocortex. PLOS Biol. 14, e2000237
rons and INs counts per species group and JNEUROSCI.0175-21.2021; pmid: 34667071 (2016). doi: 10.1371/journal.pbio.2000237; pmid: 27828957
generating 1000 bootstrap samples (99) with 12. V. Szegedi et al., Robust perisomatic GABAergic self- 31. K. D. Micheva, C. Beaulieu, Postnatal development of GABA
replacement. IN fraction (INfrac) was computed innervation inhibits basket cells in the human and mouse neurons in the rat somatosensory barrel cortex:
for each bootstrap sample. The P-value was supragranular neocortex. eLife 9, e51691 (2020). A quantitative study. Eur. J. Neurosci. 7, 419–430 (1995).
doi: 10.7554/eLife.51691; pmid: 31916939 doi: 10.1111/j.1460-9568.1995.tb00338.x; pmid: 7773439
calculated as the fraction of bootstrap samples 13. A. M. M. Sousa et al., Molecular and cellular reorganization of 32. H. S. Meyer et al., Inhibitory interneurons in a cortical column
in mouse for which INfrac was greater than or neural circuits in the human lineage. Science 358, 1027–1032 form hot zones of inhibition in layers 2 and 5A. Proc. Natl.
equal to the lowest INfrac in macaque/human (2017). doi: 10.1126/science.aan3456; pmid: 29170230 Acad. Sci. U.S.A. 108, 16807–16812 (2011). doi: 10.1073/
14. E. Boldog et al., Transcriptomic and morphophysiological pnas.1113648108; pmid: 21949377
bootstrap group (Fig. 2B, error bars 10th to evidence for a specialized human cortical GABAergic cell 33. D. Džaja, A. Hladnik, I. Bičanić, M. Baković, Z. Petanjek,
90th percentiles). The same bootstrap approach type. Nat. Neurosci. 21, 1185–1195 (2018). doi: 10.1038/ Neocortical calretinin neurons in primates: Increase in
was used as a test of significance for increase in s41593-018-0205-2; pmid: 30150662 proportion and microcircuitry structure. Front. Neuroanat. 8,
15. J. DeFelipe, L. Alonso-Nanclares, J. I. Arellano, Microstructure 103–103 (2014). pmid: 25309344
non- MP IN fraction from mouse to macaque/ of the neocortex: Comparative aspects. J. Neurocytol. 31, 34. B. Tasic et al., Shared and distinct transcriptomic cell types
human by pooling all MP and non- MP INs in 299–316 (2002). doi: 10.1023/A:1024130211265; across neocortical areas. Nature 563, 72–78 (2018).
volume per species group (Fig. 5D, error bars pmid: 12815249 doi: 10.1038/s41586-018-0654-5; pmid: 30382198
16. A. Rollenhagen et al., Synaptic organization of the human 35. J. Q. Ren, Y. Aika, C. W. Heizmann, T. Kosaka, Quantitative
10th-90th percentiles) and similarly for testing temporal lobe neocortex as revealed by high-resolution analysis of neurons and glial cells in the rat somatosensory
increase of IN-targeting fraction of non-MP IN transmission, focused ion beam scanning, and electron cortex, with special reference to GABAergic neurons and
axons from mouse to macaque/human by pool- microscopic tomography. Int. J. Mol. Sci. 21, 5558 (2020). parvalbumin-containing neurons. Exp. Brain Res. 92, 1–14
doi: 10.3390/ijms21155558; pmid: 32756507 (1992). doi: 10.1007/BF00230378; pmid: 1486945
ing all output synapses of non-MP IN axons per
17. M. Domínguez-Álvaro, M. Montero-Crespo, 36. C. Beaulieu, G. Campistron, C. Crevier, Quantitative aspects
species group (Fig. 5, F and G). L. Blazquez-Llorca, J. DeFelipe, L. Alonso-Nanclares, 3D of the GABA circuitry in the primary visual cortex of the adult
electron microscopy study of synaptic organization of the rat. J. Comp. Neurol. 339, 559–572 (1994). doi: 10.1002/
Data availability normal human transentorhinal cortex and its possible cne.903390407; pmid: 8144746
alterations in Alzheimer’s disease. eNeuro 6, ENEURO.0140- 37. S. Lefort, C. Tomm, J. C. Floyd Sarria, C. C. H. Petersen, The
All electron microscopy datasets are publicly 19.2019 (2019). doi: 10.1523/ENEURO.0140-19.2019; pmid: excitatory neuronal network of the C2 barrel column in
available for browsing at webknossos.org: ma- 31217195 mouse primary somatosensory cortex. Neuron 61, 301–316
caque L2/3 (S1), https://wklink.org/1186; ma- 18. P. Marco, J. DeFelipe, Altered synaptic circuitry in the human (2009). doi: 10.1016/j.neuron.2008.12.020; pmid: 19186171
temporal neocortex removed from epileptic patients. 38. C. Beaulieu, Z. Kisvarday, P. Somogyi, M. Cynader, A. Cowey,
caque L2/3 (STG), https://wklink.org/1319; human Exp. Brain Res. 114, 1–10 (1997). doi: 10.1007/PL00005608; Quantitative distribution of GABA-immunopositive and
(H5) L23 (STG), https://wklink.org/7861; human pmid: 9125446 -immunonegative neurons and synapses in the monkey
(H5) L23 (STG) multiSEM, https://wklink.org/ 19. B. Wang et al., A subtype of inhibitory interneuron with striate cortex (area 17). Cereb. Cortex 2, 295–309 (1992).
intrinsic persistent activity in human and monkey doi: 10.1093/cercor/2.4.295; pmid: 1330121
5364; human (H5) L1-6 (STG) multiSEM, https://
neocortex. Cell Rep. 10, 1450–1458 (2015). doi: 10.1016/ 39. P. L. A. Gabbott, S. J. Bacon, Local circuit neurons in the
wklink.org/1742; human (H6) L2/3 (IFG), https:// j.celrep.2015.02.018; pmid: 25753411 medial prefrontal cortex (areas 24a,b,c, 25 and 32) in the
wklink.org/7299; mouse L2/3 (S1), https://wklink. 20. G. Molnár et al., Complex events initiated by individual spikes monkey: I. Cell morphology and morphometrics. J. Comp.
org/9045; mouse L2/3 (PPC), https://wklink.org/ in the human cerebral cortex. PLOS Biol. 6, e222 (2008). Neurol. 364, 567–608 (1996). doi: 10.1002/(SICI)1096-9861
doi: 10.1371/journal.pbio.0060222; pmid: 18767905 (19960122)364:4<567::AID-CNE1>3.0.CO;2-1; pmid: 8821449
2581; mouse L2 (ACC), https://wklink.org/7415;
21. D. S. Melchitzky, S. R. Sesack, D. A. Lewis, Parvalbumin- 40. J.-P. Hornung, N. De Tribolet, Distribution of GABA-containing
mouse L2/3 (V2), https://wklink.org/3592; mouse immunoreactive axon terminals in macaque monkey and neurons in human frontal cortex: A quantitative
L2/3 (A2), https://wklink.org/7193; mouse L1-5 human prefrontal cortex: Laminar, regional, and target immunocytochemical study. Anat. Embryol. 189, 139–145
(PPC-2), https://wklink.org/4814. specificity of type I and type II synapses. J. Comp. Neurol. (1994). doi: 10.1007/BF00185772; pmid: 8010412
408, 11–22 (1999). doi: 10.1002/(SICI)1096-9861(19990524) 41. M. R. del Río, J. DeFelipe, Colocalization of calbindin D-28k,
408:1<11::AID-CNE2>3.0.CO;2-T; pmid: 10331577 calretinin, and GABA immunoreactivities in neurons of the
RE FE RENCES AND N OT ES 22. M. Domínguez-Álvaro, M. Montero-Crespo, L. Blazquez-Llorca, human temporal cortex. J. Comp. Neurol. 369, 472–482
1. S. Herculano-Houzel, The human brain in numbers: A linearly J. DeFelipe, L. Alonso-Nanclares, 3D ultrastructural study of (1996). doi: 10.1002/(SICI)1096-9861(19960603)
scaled-up primate brain. Front. Hum. Neurosci. 3, 31 (2009). synapses in the human entorhinal cortex. Cereb. Cortex 1, 369:3<472::AID-CNE11>3.0.CO;2-K; pmid: 8743426
doi: 10.3389/neuro.09.031.2009; pmid: 19915731 410–435 (2020). 42. E. G. Jones, G. W. Huntley, D. L. Benson, Alpha calcium/
2. C. S. von Bartheld, J. Bahney, S. Herculano-Houzel, The 23. D. S. Melchitzky, D. A. Lewis, Dendritic-targeting GABA calmodulin-dependent protein kinase II selectively expressed
search for true numbers of neurons and glial cells in the neurons in monkey prefrontal cortex: Comparison of in a subpopulation of excitatory neurons in monkey sensory-

motor cortex: Comparison with GAD-67 expression. J. 64. C. Xu et al., Structure and plasticity of silent synapses in Nat. Commun. 6, 7923 (2015). doi: 10.1038/ncomms8923;
Neurosci. 14, 611–629 (1994). doi: 10.1523/JNEUROSCI.14- developing hippocampal neurons visualized by super-resolution pmid: 26235643
02-00611.1994; pmid: 8301355 imaging. Cell Discov. 6, 8 (2020). doi: 10.1038/s41421-019- 86. F. Drawitsch, A. Karimi, K. M. Boergens, M. Helmstaedter,
43. G. Eyal et al., Human cortical pyramidal neurons: From spines 0139-1; pmid: 32133151 FluoEM, virtual labeling of axons in three-dimensional electron
to spikes via models. Front. Cell. Neurosci. 12, 181 (2018). 65. G. M. Durand, Y. Kovalchuk, A. Konnerth, Long-term microscopy data for long-range connectomics. eLife 7, e38976
doi: 10.3389/fncel.2018.00181; pmid: 30008663 potentiation and functional synapse induction in developing (2018). doi: 10.7554/eLife.38976; pmid: 30106377
44. R. Benavides-Piccione, I. Fernaud-Espinosa, V. Robles, hippocampus. Nature 381, 71–75 (1996). doi: 10.1038/ 87. K. M. Boergens et al., webKnossos: Efficient online 3D data
R. Yuste, J. DeFelipe, Age-based comparison of human 381071a0; pmid: 8609991 annotation for connectomics. Nat. Methods 14, 691–694
dendritic spine structure using complete three-dimensional 66. M. Lenz et al., All-trans retinoic acid induces synaptic (2017). doi: 10.1038/nmeth.4331; pmid: 28604722
reconstructions. Cereb. Cortex 23, 1798–1810 (2013). plasticity in human cortical neurons. eLife 10, e63026 (2021). 88. M. A. Fischler, R. C. Bolles, Random sample consensus:
doi: 10.1093/cercor/bhs154; pmid: 22710613 doi: 10.7554/eLife.63026; pmid: 33781382 A paradigm for model fitting with applications to image
45. H. Mohan et al., Dendritic and axonal architecture of 67. H. Schmidt et al., Axonal synapse sorting in medial entorhinal analysis and automated cartography. Commun. ACM 24,
individual pyramidal neurons across layers of adult human cortex. Nature 549, 469–475 (2017). doi: 10.1038/ 381–395 (1981). doi: 10.1145/358669.358692
neocortex. Cereb. Cortex 25, 4839–4853 (2015). nature24005; pmid: 28959971 89. L. K. Scheffer, B. Karsh, S. Vitaladevun, Automated alignment
doi: 10.1093/cercor/bhv188; pmid: 26318661 68. J. Kornfeld et al., EM connectomics reveals axonal target of imperfect EM images for neural reconstruction.
46. G. N. Elston, R. Benavides-Piccione, J. DeFelipe, The variation in a sequence-generating network. eLife 6, e24364 arXiv:1304.6034v1 [q-bio.QM] (2013).
pyramidal cell in cognition: A comparative study in human (2017). doi: 10.7554/eLife.24364; pmid: 28346140 90. A. Peters, Cellular components of the cerebral cortex, in
and monkey. J. Neurosci. 21, RC163–RC163 (2001). 69. C. E. Carr, M. Konishi, A circuit for detection of interaural Cerebral Cortex (Univ. Chicago Press, 1984), vol. 1, pp. 107–200.
doi: 10.1523/JNEUROSCI.21-17-j0002.2001; pmid: 11511694 time differences in the brain stem of the barn owl. 91. P. Somogyi, A. Cowey, Combined Golgi and electron
47. M. B. Ramocki, H. Y. Zoghbi, Failure of neuronal homeostasis J. Neurosci. 10, 3227–3246 (1990). doi: 10.1523/ microscopic study on the synapses formed by double
results in common neuropsychiatric phenotypes. Nature 455, JNEUROSCI.10-10-03227.1990; pmid: 2213141 bouquet cells in the visual cortex of the cat and monkey. J.
912–918 (2008). doi: 10.1038/nature07457; pmid: 18923513 70. K. D. Micheva, C. Beaulieu, Quantitative aspects of Comp. Neurol. 195, 547–566 (1981). doi: 10.1002/
48. J. L. R. Rubenstein, M. M. Merzenich, Model of autism: synaptogenesis in the rat barrel field cortex with special cne.901950402; pmid: 7462443
Increased ratio of excitation/inhibition in key neural systems. reference to GABA circuitry. J. Comp. Neurol. 373, 340–354 92. A. D. de Lima, J. H. Morrison, Ultrastructural analysis of
Genes Brain Behav. 2, 255–267 (2003). doi: 10.1034/j.1601- (1996). doi: 10.1002/(SICI)1096-9861(19960923)373:3<340:: somatostatin-immunoreactive neurons and synapses in the
183X.2003.00037.x; pmid: 14606691 AID-CNE3>3.0.CO;2-2; pmid: 8889932 temporal and occipital cortex of the macaque monkey.
49. A. Motta et al., Dense connectomic reconstruction in layer 4 71. C. Calì et al., The effects of aging on neuropil structure in J. Comp. Neurol. 283, 212–227 (1989). doi: 10.1002/
of the somatosensory cortex. Science 366, eaay3134 (2019). mouse somatosensory cortex-A 3D electron microscopy cne.902830205; pmid: 2567743
doi: 10.1126/science.aay3134; pmid: 31649140 analysis of layer 1. PLOS ONE 13, e0198131 (2018). 93. J. DeFelipe, S. H. Hendry, E. G. Jones, Synapses of double
50. W. Denk, H. Horstmann, Serial block-face scanning electron doi: 10.1371/journal.pone.0198131; pmid: 29966021 bouquet cells in monkey cerebral cortex visualized by
microscopy to reconstruct three-dimensional tissue 72. J. P. Bourgeois, P. Rakic, Changes of synaptic density in the calbindin immunoreactivity. Brain Res. 503, 49–54 (1989).
nanostructure. PLOS Biol. 2, e329 (2004). doi: 10.1371/ primary visual cortex of the macaque monkey from fetal to doi: 10.1016/0006-8993(89)91702-2; pmid: 2611658
journal.pbio.0020329; pmid: 15514700 adult stage. J. Neurosci. 13, 2801–2820 (1993). doi: 10.1523/ 94. J. DeFelipe, S. H. Hendry, T. Hashikawa, M. Molinari,
51. K. J. Hayworth et al., Imaging ATUM ultrathin section libraries JNEUROSCI.13-07-02801.1993; pmid: 8331373 E. G. Jones, A microcolumnar structure of monkey cerebral
with WaferMapper: A multi-scale approach to EM 73. S. A. Kirov, L. J. Petrak, J. C. Fiala, K. M. Harris, Dendritic cortex revealed by immunocytochemical studies of double
reconstruction of neural circuits. Front. Neural Circuits 8, 68 spines disappear with chilling but proliferate excessively bouquet cell axons. Neuroscience 37, 655–673 (1990).
(2014). doi: 10.3389/fncir.2014.00068; pmid: 25018701 upon rewarming of mature hippocampus. Neuroscience 127, doi: 10.1016/0306-4522(90)90097-N; pmid: 1701039
52. K. J. Hayworth, N. Kasthuri, R. Schalek, J. W. Lichtman, 69–80 (2004). doi: 10.1016/j.neuroscience.2004.04.053; 95. P. Somogyi, T. F. Freund, A. Cowey, The axo-axonic
Automating the collection of ultrathin serial sections for large pmid: 15219670 interneuron in the cerebral cortex of the rat, cat and monkey.
volume TEM reconstructions. Microsc. Microanal. 12 (S02), 74. L. A. Glantz, D. A. Lewis, Decreased dendritic spine density Neuroscience 7, 2577–2607 (1982). doi: 10.1016/0306-4522
86–87 (2006). doi: 10.1017/S1431927606066268 on prefrontal cortical pyramidal neurons in schizophrenia. (82)90086-0; pmid: 7155343
53. A. L. Eberle et al., High-resolution, high-throughput imaging Arch. Gen. Psychiatry 57, 65–73 (2000). doi: 10.1001/ 96. B. Cauli et al., Molecular and physiological diversity of cortical
with a multibeam scanning electron microscope. J. Microsc. archpsyc.57.1.65; pmid: 10632234 nonpyramidal cells. J. Neurosci. 17, 3894–3906 (1997).
259, 114–120 (2015). doi: 10.1111/jmi.12224; pmid: 25627873 75. M. Medalla, J. P. Gilman, J.-Y. Wang, J. I. Luebke, Strength and doi: 10.1523/JNEUROSCI.17-10-03894.1997; pmid: 9133407
54. K. Lee, J. Zung, P. Li, V. Jain, H. S. Seung, Superhuman diversity of inhibitory signaling differentiates primate anterior 97. Y. Kawaguchi, Y. Kubota, Physiological and morphological
accuracy on the SNEMI3D connectomics challenge. cingulate from lateral prefrontal cortex. J. Neurosci. 37, identification of somatostatin- or vasoactive intestinal
arXiv:1706.00120v1 [cs.CV] (2017). 4717–4734 (2017). doi: 10.1523/JNEUROSCI.3757-16.2017; polypeptide-containing cells among GABAergic cell subtypes
55. A. Karimi, J. Odenthal, F. Drawitsch, K. M. Boergens, pmid: 28381592 in rat frontal cortex. J. Neurosci. 16, 2701–2715 (1996).
M. Helmstaedter, Cell-type specific innervation of cortical 76. A. Hsu, J. I. Luebke, M. Medalla, Comparative ultrastructural doi: 10.1523/JNEUROSCI.16-08-02701.1996; pmid: 8786446
pyramidal cells at their apical dendrites. eLife 9, e46876 features of excitatory synapses in the visual and frontal 98. C. K. Pfeffer, M. Xue, M. He, Z. J. Huang, M. Scanziani,
(2020). doi: 10.7554/eLife.46876; pmid: 32108571 cortices of the adult mouse and monkey. J. Comp. Neurol. Inhibition of inhibition in visual cortex: The logic of connections
56. V. Braitenberg, A. Schüz, Cortex: Statistics and Geometry of 525, 2175–2191 (2017). doi: 10.1002/cne.24196; between molecularly distinct interneurons. Nat. Neurosci. 16,
Neuronal Connectivity (Springer Science & Business Media, pmid: 28256708 1068–1076 (2013). doi: 10.1038/nn.3446; pmid: 23817549
1998). 77. J. P. Gilman, M. Medalla, J. I. Luebke, Area-specific features 99. B. Efron, Bootstrap methods: Another look at the jackknife.
of pyramidal neurons-a comparative study in mouse and Ann. Stat. 7, 1–26 (1979). doi: 10.1214/aos/1176344552
57. T. Kwon et al., Ultrastructural, molecular and functional
rhesus monkey. Cereb. Cortex 27, 2078–2094 (2017).
mapping of GABAergic synapses on dendritic spines 100. S. B. Hedges, S. Kumar, Genomics. Vertebrate genomes
pmid: 26965903
and shafts of neocortical pyramidal neurons. Cereb. Cortex 29, compared. Science 297, 1283–1285 (2002). doi: 10.1126/
2771–2781 (2019). doi: 10.1093/cercor/bhy143; 78. A. Gidon et al., Dendritic action potentials and computation in science.1076231; pmid: 12193771
pmid: 30113619 human layer 2/3 cortical neurons. Science 367, 83–87 101. M. A. Hofman, Evolution of the human brain: When bigger is
58. Y. Kubota, F. Karube, M. Nomura, Y. Kawaguchi, The diversity of (2020). doi: 10.1126/science.aax6239; pmid: 31896716 better. Front. Neuroanat. 8, 15 (2014). doi: 10.3389/
cortical inhibitory synapses. Front. Neural Circuits 10, 27–27 79. G. Eyal et al., Unique membrane properties and enhanced fnana.2014.00015; pmid: 24723857
(2016). doi: 10.3389/fncir.2016.00027; pmid: 27199670 signal processing in human neocortical neurons. eLife 5,
59. G. A. Wildenberg et al., Primate neuronal connections are e16553 (2016). doi: 10.7554/eLife.16553; pmid: 27710767 AC KNOWL ED GME NTS
sparse in cortex as compared to mouse. Cell Rep. 36, 109709 80. B. Scholl, C. I. Thomas, M. A. Ryan, N. Kamasawa,
We thank K. Mätz-Rensing at the German Primate Center and
(2021). doi: 10.1016/j.celrep.2021.109709; pmid: 34525373 D. Fitzpatrick, Cortical response selectivity derives from
F. Bremmer and A. Kaminiarz at the University of Marburg for
60. A. T. Kuan et al., Dense neuronal reconstruction through strength in numbers of synapses. Nature 590, 111–114
providing tissue of macaque brain; anonymous individuals H5 and
x-ray holographic nano-tomography. Nat. Neurosci. 23, (2021). doi: 10.1038/s41586-020-03044-3; pmid: 33328635
H6 for consent to provide tissue from neurosurgery interventions;
1637–1643 (2020). doi: 10.1038/s41593-020-0704-9; 81. J. J. Letzkus et al., A disinhibitory microcircuit for associative
H. Wissler for support with visualizations and for management of
pmid: 32929244 fear learning in the auditory cortex. Nature 480, 331–335
the neuron reconstruction team; I. Wolf and J. Kim for staining
61. Y. Kubota, S. Hatada, S. Kondo, F. Karube, Y. Kawaguchi, (2011). doi: 10.1038/nature10674; pmid: 22158104
support; A. Karimi, F. Drawitsch, and K. M. Boergens for providing
Neocortical inhibitory terminals innervate dendritic spines targeted 82. J. J. Letzkus, S. B. E. Wolff, A. Lüthi, Disinhibition, a circuit custom written image acquisition and alignment routines; S. Babl,
by thalamocortical afferents. J. Neurosci. 27, 1139–1150 (2007). mechanism for associative learning and memory. Neuron 88, L. Bezzenberger, R. Jakoby, R. Kneisl, and M. Kronawitter for
doi: 10.1523/JNEUROSCI.3846-06.2007; pmid: 17267569 264–276 (2015). doi: 10.1016/j.neuron.2015.09.024; annotator training and task management; D. Werner, V. Pinkau,
62. M. R. del Río, J. DeFelipe, A light and electron microscopic study pmid: 26494276 G. Wiese, F. Meinel, J. Striebel, P. Otto, T. Herold, and N. Rzepka at
of calbindin D-28k immunoreactive double bouquet cells in the 83. R. Kim, T. J. Sejnowski, Strong inhibitory signaling underlies scalable minds, Potsdam, Germany for collaboration on automated
human temporal cortex. Brain Res. 690, 133–140 (1995). stable temporal dynamics and working memory in spiking data analysis; T. Garbowski and D. Lechner (Zeiss) for help with
doi: 10.1016/0006-8993(95)00641-3; pmid: 7496800 neural networks. Nat. Neurosci. 24, 129–139 (2021). setting up multiSEM; M. Fahland and N. Prager (Fraunhofer FEP,
63. I. P. Lukacs et al., Differential effects of group III doi: 10.1038/s41593-020-00753-w; pmid: 33288909 Dresden) for collaboration on Kapton tape; R. Schalek and J. Lichtman
metabotropic glutamate receptors on spontaneous inhibitory 84. A. Gour et al., Postnatal connectomic development of for initial advice on setting up ATUM; G. Tushev, A. Kotowicz,
synaptic currents in spine-innervating double bouquet and inhibition in mouse barrel cortex. Science 371, eabb4534 T. Zeller, and C. Guggenberger for IT support; S. Soworka, I. Wolf,
parvalbumin-expressing dendrite-targeting GABAergic (2021). doi: 10.1126/science.abb4534; pmid: 33273061 S. Horn and L Dadashev for excellent technical support; and three
interneurons in human neocortex. bioRxiv, 483105 [Preprint] 85. Y. Hua, P. Laserstein, M. Helmstaedter, Large-volume en-bloc anonymous reviewers for helpful advice. Additional methodological
(2022). doi: 10.1101/2022.03.05.483105 staining for electron microscopy-based connectomics. acknowledgments are provided in the supplementary materials.

Funding: Funding was provided by the Max-Planck Society. Data and materials availability: All image data and reconstructions SUPPLEMENTARY MATERIALS
S.L. was partly funded by Neuronex2 grant (DFG, German Research are publicly available through https://www.webknossos.org under science.org/doi/10.1126/science.abo0924
Foundation–HE 7321/1-1). Author contributions: M.H. conceived, the links at the end of the Materials and methods section. All Materials and Methods
initiated and supervised the study. J.S., N.H., A.M.K., V.G., and software used for analysis is available at the following address Supplementary Text
N.J. carried out experiments. H.S.M. and J.G. provided human tissue. and have been made publicly available under the MIT license: Figs. S1 to S4
M.S. provided ATUM-multiSEM methodology. A.M. developed and https://gitlab.mpcdf.mpg.de/connectomics/human_primate. MDAR Reproducibility Checklist
applied analysis methodology. S.L., J.S., N.H., A.M.K., V.G., and License information: Copyright © 2022 the authors, some rights
M.H. provided reconstructions and analyzed data. S.L. and M.H. wrote reserved; exclusive licensee American Association for the Submitted 13 January 2022; accepted 30 May 2022
the manuscript, with contributions from all authors. Competing Advancement of Science. No claim to original US government works. Published online 23 June 2022
interests: The authors declare no competing financial interests. https://www.science.org/about/science-licenses-journal-article-reuse 10.1126/science.abo0924

RES EARCH
◥ not have been possible with earlier versions

RESEARCH ARTICLE of the ChemPU, which could not use cDL
(14–16). It also signifies a massive step up in
ORGANIC CHEMISTRY the number of validated cDL procedures com-
pared with the seminal paper on cDL (15) and
Digitization and validation of a chemical synthesis is testimony to the increased reliability of
the hardware employed in this paper. We
literature database in the ChemPU designed and built a cDL database called
Chemify (34) for our current 103 entries and
Simon Rohrbach†, Mindaugas Šiaučiulis†, Greig Chisholm†, Petrisor-Alin Pirvan, Michael Saleeb, anticipate this will rapidly expand; the data-
S. Hessam M. Mehr, Ekaterina Trushina, Artem I. Leonov, Graham Keenan, Aamir Khan, base will be available for anyone to run and
Alexander Hammer, Leroy Cronin* validate on suitable hardware. Not only could
these cDL entries be implemented on other
Despite huge potential, automation of synthetic chemistry has only made incremental progress over the automated synthesis platforms and material
past few decades. We present an automatically executable chemical reaction database of 100 molecules generated on demand, but statistics could be
representative of the range of reactions found in contemporary organic synthesis. These reactions include gathered and new versions suggested if re-
transition metal–catalyzed coupling reactions, heterocycle formations, functional group interconversions, and quired. In addition to directly repeating the
multicomponent reactions. The chemical reaction codes or cDLs for the reactions have been stored in a validated procedures, the substrate scope for
database for version control, validation, collaboration, and data mining. Of these syntheses, more than each cDL can be gradually expanded by
50 entries from the database have been downloaded and robotically run in seven modular ChemPU’s changing the substrates and adjusting key
with yields and purities comparable to those achieved by an expert chemist. We also demonstrate the parameters—such as temperature or time—
automatic purification of a range of compounds using a chromatography module seamlessly coupled to of the reaction while keeping the rest of the
the platform and programmed with the same language. process unchanged. Because we have selected
reactions based on popularity, the resulting
T
set of validated cDLs covers a substantial
o replicate a known chemical reaction flow, but even these have bespoke instruction range of common reactions and constitutes an
the protocol must be obtained from sets with no simple semantic link among entry point to automation of the entire organic
either the literature or a database so them or to the literature. To fully exploit the synthesis toolbox. Further, through perform-
that it can be run manually in the lab- potential of automation in chemical synthesis ing 53 procedures of highly diverse chemistry,
oratory (1). However, not all literature and ensure reproducibility of procedures, prog- the hardware and software of the ChemPU
or database entries can be easily reproduced ress is needed on two fronts (31, 32). First, a has been pushed to the limit and a path to full
(2). This is a barrier not only to the synthesis truly universal automation platform is required universality demonstrated. To do this, key ad-
of new molecules but also to accumulation that can perform all unit operations (14, 16); vances have been made by incorporating a
of high-quality data for machine learning (3–6) second, a standardized and precise syntax to cDL-enabled flash column chromatography
and is exacerbated by the fact that there is describe these chemical processes is essential system in the hardware library. This means
also no open standard for coding the pro- to reliably capture all the critical details of a that the ChemPU can perform not only the
cedures or a way to widely report and correct given chemical process (15). Such a code must reaction, work up, and concentration, but also
failed experiments (7, 8). An approach that also be independent of the type of hardware the chromatographic separation of the pro-
unambiguously captures and codes a chemi- employed for automation and thus be com- duct to directly deliver the purified compound
cal synthesis protocol for use by an automated piled to work flawlessly on any compatible on demand. To achieve this, we show that the
system (9–16) with capacity to be versioned hardware system. platform can react in a dynamic manner, re-
similar to software and record failed experi- We present the design, construction, and sponding to detection of the product to collect
ments would transform the field. Organic validation of a workflow that allows us to cap- the appropriate fractions.
synthesis currently requires intensive, highly ture the chemical synthesis literature from The workflow starting from a literature pro-
skilled labor (17) and a typical synthesis can manual operation to a fully described and cedure to a validated entry in our Chemify
require multiple complex unit operations that universal Chemical Description Language (cDL) database is illustrated in Fig. 1. By contrast
are difficult to explicitly encode. This is be- (15, 33) to be run automatically in the Chem- to earlier work on cDL, the focus was not on
cause the tacit knowledge required is often ical Processing Unit or ChemPU. The process an exact translation of the original procedure
context-dependent, resulting in ambiguities of running the cDL on the ChemPU we text to cDL but rather on the implementation
in the published literature that limit reprodu- call chemputation (similar to computation) of a chemical process providing the target
cibility, automation, or data mining (18). and is the reliable conversion of code and molecule. Following this approach allowed us
These limits have been overcome in some reagents into products. We not only show not only to reproduce the literature but also to
specific areas such as oligopeptide (19), oli- that the cDL can be compiled to run on many improve the processes in several instances.
gosaccharide (20), and oligonucleotide (21) different ChemPU configurations but also Chemical reactions can be captured in cDL,
chemistry, and in recent years much progress demonstrate the capacity of the cDL language which represents synthetic steps as sequen-
has been made in automating chemical reac- to encode a wide range of synthetic proce- ces of physical processes such as Add, Dissolve,
tions more broadly (22–30). However, most dures, which are representative of the organic Evaporate, and more. There are currently 44
automated synthetic chemistry platforms re- chemistry toolbox. Overall, 103 different reac- steps within the cDL framework with each
main task-specific or represent islands of tions of highly varied chemistry have been step having a fully customizable set of param-
automation (10) in an otherwise manual work- translated from the literature to reliable cDL eters. All often-used tasks in organic syntheses
codes, and 53 of these programs have been have a boilerplate cDL step to represent them,
School of Chemistry, the University of Glasgow, University validated on the hardware with yields and such as EvacuateAndRefill to establish an
Avenue, Glasgow G12 8QQ, UK.
*Corresponding author. Email: lee.cronin@glasgow.ac.uk purities comparable to that in the literature. inert atmosphere or Separate to perform a
†These authors contributed equally to this work. This increased synthesis throughput would liquid-liquid separation and extraction. The
Rohrbach et al., Science 377, 172–180 (2022) 8 July 2022 1 of 9

Fig. 1. Schematic of cDL

protocol optimization.
Chemical reactions can
be captured with cDL, "The reaction was
which represents synthetic performed under
steps as sequences of physical inert atmosphere. [...]" time
processes such as Add,
Dissolve, and Evaporate. The "Alkyne (1.75 mL) m/z
initially established cDL was added drop- ppm
protocol is then executed wise. [...]"
on a ChemPU and the purity
and yield of the product are "The product was
determined. The cDL protocol extracted with
can be improved until the ethyl acetate. [...]"
process meets the expectation
of product purity and yield.
At that stage the protocol can
be added to the database as Red. amination v5.1 Me
validated, backed up by the full I OH
Me
characterization of the target Sonogashira v3.1 H
product and the process develop- Me
Mitsunobu v1.2 OH
ment history. Me
Suzuki-Miyaura v2.7
cDL steps help enforce precise descriptions of intangible is now captured in the cDL. At this Further, the database also contains cDL en-
the process and eliminate any ambiguity such stage the protocol can be added to the data- tries that have been translated but not yet
as the number of cycles of evacuation and inert base as a validated process, backed up by the executed on a suitable automation platform.
gas refill or process-critical addition speeds. To full characterization of the target product and Users interested in unvalidated cDL files can
achieve this, we used our web-based Chemis- process development history. Inclusion of pro- access these and have the option to validate
try Development Environment (ChemIDE) (33) cess development history is a distinguishing them. The cDL procedures reported here have
which aids the quick generation of cDL pro- feature of the Chemify database; by showing been validated on a ChemPU, a chemistry
cedures by providing a text-to-cDL translation the results of less successful experiments and automation platform that emulates the man-
tool. This works by using a template library of contrasting them with the final successful run, ual operations of a bench chemist. Although
all available cDL steps and an editor in which critical aspects of the process are highlighted operationally simple and intuitive, the rigor-
individual cDL steps are represented as graph- and can be quantified. ous implementation means that the platform
ical elements, which can be edited and ar- The Chemify database persistently stores in- operates as a finite state machine (Fig. 2). It
ranged as needed (33). ChemIDE was used in formation for cDL procedures, experimental can be in one of a finite number of states and
the generation of all cDL procedures detailed results, and relevant analyses. It is a locally transitions from one state to the next based on
in this work. hosted PostgreSQL database server containing well-defined operations. These operations are
Expression of a chemical procedure in cDL all validated cDL scripts as described above, defined by the program—the cDL synthesis
does not immediately solve the problem of which can be accessed through ChemIDE (the protocol—as well as the sensor feedback [e.g.,
missing information or ambiguity present in web-based cDL development environment) or temperature, conductivity, pressure, or ultra-
the original prose instructions but it does pro- by using a Python 3–based API for automated violet (UV) absorbance]. The direct mapping
vide an unambiguous path to close it. To do database querying. Moreover, for end user of the cDL synthesis instructions to state tran-
this, some process development and iteration experience, ChemIDE is equipped to display sitions or “unit operations” highlights the
may still be required to maximize yields and characterization parameters of each experiment rigorous abstraction of synthesis processes
purity. After appropriate analysis [Nuclear such as product scale, yield, status (translated, in cDL. Moreover, the clear definition of state
magnetic resonance spectroscopy (NMR), liq- validated, failed) and process duration. Users transitions as defined in the cDL procedure is
uid chromatography mass spectrometry, or gas can submit, search, download, and reproduce critical to ensuring the reproducibility of the
chromatography mass spectrometry] of the trusted syntheses. The database contains final cDL synthesis, including on different layouts
target compound from the ChemPU execu- validated synthesis scripts as well as previous of the ChemPU and potentially entirely dif-
tion of the cDL code, an assessment of the developmental versions, which may work to a ferent qualified hardware setups.
quality and purity of the product is made. If varying degree, affording the desired products The ChemPU state machine consists of
necessary the cDL is improved to increase the in lower yields, insufficient purities, or leading three logical parts: the physical input or out-
yield and purity and then executed again. The to process failures (for example, causing block- put (I/O), digital I/O, and the processing unit.
key advantage of cDL is that once a successful ages or formation of emulsions during liquid- The processing unit can transition through
process has been encoded, all subsequent users liquid separations) as a result of insufficient or several states based on either the initial con-
who execute the code on compatible hardware incorrect description of the necessary process ditions of the ChemPU or a combination of the
can expect identical results, with no further parameters for automation. Comparing failed physical and digital I/O, that is, the current
requirements for process development. All or lower-yield experiments to successful at- conditions as defined by the sensors, the pro-
critical knowledge needed to execute the pro- tempts of a given specific reaction or reaction cess variables, and the cDL step being executed.
cess on qualified hardware both tangible and class can unveil critical aspects of the process. The execution of the cDL step according to the

Fig. 2. The ChemPU can

be regarded as a chemical
state machine. The ChemPU
processes the physical input
(i.e., reagents, solvents) based
on the digital input (i.e., the
synthesis script). Each unit
operation defines a route for
the system to progress from
the current state to the next. The
state of the system is always well
defined, and every detail of the
synthesis is known and can be
documented. PV, process
variables.
scheduler gives rise to a new state to be acted parts of the process). The liquid-handling back- medicinal chemistry, process chemistry, and
upon in later steps and results in a physical bone consists of a series of syringe pumps and total synthesis (35–38). There are some nota-
change to the physical I/O, e.g., a change of valves. A typical backbone consists of six of ble differences in the distribution of reaction
location of reagents, a change in temperature, each; however, the backbone is readily con- classes used in synthesis, depending on the
the phase boundary in a liquid-liquid separation, tracted or expanded to accommodate the re- primary goal; for example, medicinal chem-
or the peak elution during chromatography. quirements of the desired chemical process. istry researchers may prefer transition metal–
The scheduler resorts to a graph representation The valves have six positions and seven ports catalyzed C-C bond-forming reactions which
of the hardware (abstract layer) to interpret the each. Each valve in the liquid-handling back- allow for convenient generation of large num-
cDL script and orchestrate the hardware for bone is connected to a pump, its nearest neigh- bers of related compounds for biological assays,
concerted tasks (e.g., moving liquid through boring valves, and a waste container, and can whereas modern total synthesis relies more
the liquid-handling backbone). The abstract connect to three to four different reagents, heavily on elaborate ring-forming reactions
layer defines the locations and connections of solvents, or hardware modules. The connec- for assembly of complex molecular scaffolds
the hardware devices as nodes and contains tivity of the modules to the backbone is re- in the fewest number of steps possible (39).
specific information on each node such as the presented in an abstract manner by a graph as Additionally, although protecting-group
IP address and temperature limits of the device described above. Cleaning of the backbone is chemistry is the cornerstone of some syn-
in question. The graph file together with the carried out through an automated cleaning thesis fields such as peptide synthesis (40)
cDL file can be compiled into an execution file routine that can be defined by the user to ac- or carbohydrate chemistry (41), researchers
(executable cDL or xdlexe), which is platform- count for different types of contamination working on total syntheses often prefer more
specific. The strict separation of the description present after different procedures. In addition elegant protecting-group–free approaches (42).
of the chemical process into the cDL file and to the liquid-handling backbone, the ChemPU Despite the minor variations, these categories
the hardware platform description into the systems used to execute the syntheses re- embody the varied toolbox of modern organic
graph file ensures that the cDL file remains ported here incorporated a reaction module chemistry. To represent these categories with
platform-independent. It also allows for flexi- consisting of a standard hotplate controlled the examples from all types of reactions we
bility in how the platform is designed and through an Ethernet-to-serial convertor, a sep- chose to translate the cDLs of these procedures
what its exact physical layout is. This means arator for liquid-liquid extractions equipped and validate them with the ChemPU (Fig. 4).
that each cDL can be versioned and compiled with an overhead stirrer for agitation, as well The carbon-carbon bond-forming reaction class
to run on any suitable platform and that the as a conductivity sensor for phase boundary was further separated into transition metal–
ChemPU system is highly modular, flexible, detection; it also includes a jacketed filter for catalyzed and transition metal–free reactions.
and extensible (Fig. 3). precipitation and recrystallization of products, Furthermore, a separate multicomponent reac-
By mirroring the unit operations of batch a number of reagent flasks, a rotary evapora- tion class was introduced as these reactions
synthetic chemistry, the ChemPU represents a tor, and an optional chromatography system. generally accomplish multiple chemical trans-
universal, programmable hardware platform formations in one synthetic operation. The
for execution of synthetic chemistry as previ- Validation of literature procedures initial reactions were chosen from the most
ously demonstrated (14–16). The platform can on the ChemPU cited papers in the journal Organic Syntheses
be readily expanded as a result of its modular With the abstraction of chemputation, the (43). This journal is notable in the organic
nature, with individual modules being con- cDL language, and the ChemPU platform, chemistry field in that it publishes practical
nected through the liquid-handling backbone, we set out to translate and automate the typ- methods for either synthesis of notable com-
analogous to the bus of a conventional com- ical reactions from the organic chemistry tool- pounds or execution of important synthetic
puter. Connection to the liquid-handling back- box. Organic chemistry encompasses an methods, and the submitted procedures have
bone (consisting of pumps and valves) is immense diversity of transformations. Despite been repeated at least once by expert chemists
through a single piece of flexible tubing, which a large degree of variety, most reactions can be independent of those who submitted the orig-
allows modules to be easily removed for main- classified succinctly with fewer than ten cat- inal synthesis. Although the procedures from
tenance or rearranged to optimize operations egories. Several studies have analyzed the re- this journal generally have a high level of de-
(e.g., by segregating aqueous and water-sensitive action frequencies in different fields, e.g., tail there was still a need for some process

Fig. 3. The physical layout of the ChemPU A Chromatography Reactor

and the available hardware library can be Filter
Overhead stirrer
represented as a graph. (A) The exact state of
the ChemPU for each individual synthesis is Rotary evaporator Hotplate
represented as a graph. The nodes indicate the Conductivity sensor Flask
modules and the edges define the tubing Separating funnel Pump
connections for liquid transfers and physical
connections. (B) A schematic representation of Pneumatic panel Valve
the ChemPU with images of the individual Liquid transfer path
modules. The ChemPU emulates the manual Physical connection
batch chemistry workflow and uses much of the Inert gas path
typical laboratory hardware. The latest key
addition to the ChemPU hardware library, as
described in this paper, is a flash chromatogra- B
phy system which allows for fully automated
purification of the reaction products. The liquid-
handling backbone consists of an array
of pumps and valves which transfer reagents,
solvents, and solutions of intermediates
between the different units of the system.
Reactions are either carried out in a round
bottom flask reactor or a filter. The work up is
performed in a separator that is agitated with an
overhead stirrer. The phase boundary is
detected with a conductivity sensor. Solutions
are concentrated in a rotary evaporator.
Column chromatography is performed on a
flash column chromatography machine.
development, highlighting the difficulty of classical reactions as the Wittig reaction, prehensive list of all validated and translated
capturing all necessary information in an un- Friedel-Crafts alkylation, and the Aldol and reactions, see supplementary materials figs. S140
structured prose text format as opposed to Claisen condensations. Different types of and S143, respectively) covers the standard orga-
cDL. Selecting these highly cited papers from heteroatom alkylations are represented by nic chemistry toolbox. Crucially, automating fur-
Organic Syntheses covered the top reaction palladium-catalyzed Buchwald-Hartwig cou- ther reactions simply requires translation of the
classes but provided an uneven distribution. pling, copper-catalyzed alkylation, SNAr reac- original synthesis procedure to cDL.
Hence, further examples were manually selection of heteroarenes, and reductive amination The average procedure consists of 20 dis-
ted from notable literature sources to achieve reactions. Functional group interconversions crete, high-level instructions such as Add, Sep-
a more balanced representation of the organic include a Mitsunobu reaction, nitrile forma- arate, and Evaporate with some procedures
chemistry toolbox with our dataset. tion, and esterifications, among others. Mani- having up to 40 such instructions (Fig. 4C).
The reactions chosen for each of the cat- pulations of protecting groups include common Unpacking these high-level cDL steps into the
egories include well-established classical reac- boc, benzyl, and tosyl groups. Ring and hetero- corresponding unit operation—e.g., StartStir,
tions and important contemporary reactions, cycle formations include both classical synthe- WaitForTemp, ApplyVacuum—gives an aver-
as well as some more unconventional synthetic ses such as the Fischer indole synthesis and age of 266 operations that have been executed.
transformations (Fig. 4); for a comprehensive a more exotic formation of a trisubstituted The successful executions of all cDL scripts
list of all translated reactions, see supplement- pyrylium salt. Reduction and oxidation reac- took >1000 hours of chemputation across
ary materials fig. S143. The selected transition tions span conventional hydride reduction, seven different systems. This figure only in-
metal–catalyzed carbon-carbon bond form- Jones oxidation, and a palladium-catalyzed cludes the operations from the final iteration of
ing reactions included commonly used Suzuki, hydrogen transfer reaction. Finally, the multi- each cDL protocol and includes the reaction
Heck, and Sonogashira couplings, as well as a component reactions include the well-known time but does not account for asynchronous
stereoselective Carroll rearrangement. Ugi reaction as well as other more unusual steps, i.e., steps in which two processes are
The transition metal–free carbon-carbon cascade reactions and one-pot multistep man- running in parallel on the same ChemPU
bond-forming reaction class encompasses such ifolds. This diverse set of reactions (for a com- hardware, such as a cleaning step for a rotary

A B
Fig. 4. A representative selection of most-used reaction classes have been translated to cDL and validated on a ChemPU. (A) Number of examples per reaction
class. Once a literature procedure has been captured by cDL it is marked as “translated.” When a translated procedure is successfully executed on a ChemPU it is moved to the
“validated” class of cDL scripts. For reference, the average frequency of reactions over the fields of medicinal chemistry, process chemistry, and total syntheses are shown
(35–38). (B) The distribution of validated reactions and a specific example for illustration is shown. (C) Chemical operations, unit operations, and total runtime per procedure.
evaporator running at the same time as a re- boron trifluoride used in a Friedel-Crafts alkyl- zation cascade has been used for rapid gen-
action. The yields of the reactions performed on ation of a steroid estrone to afford derivative eration of a scaffold containing multiple stereo-
the ChemPU were in general comparable to 4, or Eaton’s reagent (10% phosphorus pent- genic centers 7, with potential for further
that of the literature yields after a period of oxide solution in methanesulfonic acid) used derivatization in a library synthesis. Similarly,
process development. This could be required in a Fischer indole synthesis of 6. Additionally, a copper(I)-catalyzed three-component coupling/
to fill the gaps in the original protocol and is reactions requiring inert atmosphere were suc- palladium(0)-catalyzed annulation cascade was
common to all synthetic development whether cessfully executed on the platform including a also successfully applied, affording product 8
manual or automated, or to adapt elements of palladium-catalyzed enantioselective Carroll re- which contains the indenoisoquinoline scaffold.
the protocol not amenable to automation, such arrangement to give 5. Procedures of up to
as unexpected formation of precipitates that 90 mmol scale were efficiently executed on Expanding the substrate scope
lead to blocked lines. A selection of reactions is our ChemPU platform. Conveniently, once a The substrate scope of validated cDL pro-
shown in Fig. 5 to illustrate the performance of cDL script is produced, a particular reaction cedures can be expanded by generating a
the platforms and give specific examples to can be scaled up or down within the constraints compound library with the ChemPU. One
show the breadth of chemistry that has been of the available vessel sizes and the chemical particularly attractive prospect is the use of
performed. process, such as safety considerations or heat validated cDL procedures for the construction
and mass transfer (see supplementary mate- of large libraries of compounds for biological
Automation of a diverse set of reactions on rials section 3, fig. S141, for the full distribution screening. Such libraries could conveniently
the ChemPU of reaction scales). The cDL procedures for the be accessed simply by changing the starting
The system is tolerant of moisture-sensitive or generation of more complex products arising materials without major modifications to the
highly reactive reagents such as potassium bis from multicomponent and cascade reactions synthesis scripts; i.e., once a process has been
(trimethylsilyl)amide (KHMDS) used in a copper- were also successfully executed on the plat- established it can be applied to many differ-
mediated alkynylation of a carbamate to afford 3, form. For example, a Petasi/Diels-Alder cycli- ent substrates as a general procedure by only

Fig. 5. Representative examples of cDL procedures validated on the ChemPU. Transformations from all the main reaction classes afforded products in yields
comparable to those reported in the literature for manual synthesis. Additionally, multicomponent reactions were performed and a small library of compounds was prepared
by varying the starting materials of one of the multicomponent reactions. Finally, the reproducibility of a validated cDL procedure has been examined and a cDL procedure
of a multistep synthetic sequence has been validated. The counts for total steps and total unit operations include all examples in a given category.
varying key parameters such as the substrates, through a multicomponent Ugi reaction. To combinations from two different isocyanide
reaction solvent, and reaction time. To show- do this, we conducted simultaneous execution and two aldehyde starting materials affording
case such an approach, a small library of of multiple or “multithreaded” reactions in four structurally related a-acylamino amide pro-
a-acylamino amides 9a to 9d was synthesized parallel on the ChemPU by using reactant ducts. Further expansion of the set of reactants

used would rapidly expand the number of script was obtained and the reaction protocol or repeating the same reaction with different
products generated and allow for swift gener- was successfully replicated 10 times in 12 at- substrates once the initial protocol has been
ation of larger libraries. tempted runs. The two failures were caused set up.
by incorrect phase boundary determination
Reproducibility of the ChemPU synthesis during liquid-liquid separations; product could Multistep synthesis
To examine the consistency and reliability of have been recovered through manually restart- The versatility of the platform is further dem-
executing the curated cDL procedures, we set ing the system, but that was not done here. onstrated by the ability to execute multistep
out to repeat the same reaction protocol mul- Crucially, execution of the curated cDL pro- synthetic sequences. Atropine 13, an anticho-
tiple times on the ChemPU platform. An cedure reliably afforded the product in con- linergic medication used in treatment of nerve
alkylation of malonate ester (affording 10) sistent yields (avg. 94%, min 89%, std: 2%) agent poisoning, was synthesized in four steps
was chosen as a suitable reaction for the re- and purities (avg. 96%, min 94%, std: 1%). from simple commercially available starting
producibility study, as accurate temperature Together with the ability to generate libraries materials. Synthetic protocols for individual
control and rate of addition are key to the of compounds, the ChemPU can be used to steps from multiple sources—as well as a re-
success of the process. After the initial pro- automate the highly repetitive work of gen- duction protocol that was previously reported
cess development, a validated cDL procedure erating multiple batches of the same material for related substrates but not for the synthesis
Before the run: optimize run parameters

ChemPU
Run preparation
Material transfers
Communication
Sample injection: transfer of crude material
Chromatography module Separation run
Return product: transfer back to ChemPU

Examples
Fig. 6. The chromatography module. This component stands out from the correction over the gradient, and column equilibration) are performed. Next
ChemPU hardware library in terms of the complexity of the information and the sample is injected onto the column. During the gradient run the chromatography
material flow between the module and the controller and the other hardware. The machine sends the detector signals to the ChemPU controller in real time.
chromatography method consists of two crucial parts: the sample injection The controller performs the peak detection and triggers the fraction collection
protocol and the solvent gradient. Once these parameters have been optimized mechanism of the chromatography machine. When the separation run is
the separation run can be initiated. The ChemPU defines the run parameters on complete the product peak is identified and transferred to the next module
the commercial chromatography unit. Then the run preparations (baseline (usually the rotary evaporator).

of 12—were successfully converted to cDL pro- Once the method is developed and coded that potentially many different robotic ap-
cedures. The ability to efficiently execute multi- in cDL it can be executed on the ChemPU or proaches will be able to use identical cDL
step reaction protocols combined with the equivalent automation system as shown in codes to produce identical results. The use of a
reliability offered by reproducible execution Fig. 6. The platform controller starts the chro- cDL Chemify database will not only facilitate
of a well-defined synthesis script reaffirms matography process by defining the run pa- reproduction of published procedures but also
the universality of the platform toward the rameters on the commercial chromatography provide the community with a rich source of
breadth of synthetic organic chemistry. unit (central hub), such as flowrate and de- validated data amenable to state-of-the-art
tector settings. The actual run preparations, machine learning for reaction optimization,
Fully automated purification on the ChemPU such as baseline corrections and the equilibra- route planning, increased safety, and reduced
Chromatographic separation of the product tion of the column, are then executed. Next, environmental impact of synthesis while subs-
compound from a reaction is the go-to method the sample of crude material is dissolved, tran- tantially reducing labor for bench chemists
of purification for small- and medium-scale sferred to the chromatography machine, and repeating well-known procedures.
organic syntheses. Many commercially avail- injected onto the column. The sample injec-
able chromatography systems exist for assist- tion process also includes a rinsing sequence REFERENCES AND NOTES
ing lab-based chemists in chromatographic to minimize loss of material during the sample 1. W. A. Warr, Mol. Inform. 33, 469–476 (2014).
separations. However, these systems still re- dissolution and transfer. Once the sample load- 2. M. Baker, Nature 533, 452–454 (2016).
3. H. Gelernter, J. R. Rose, C. H. Chen, J. Chem. Inf. Comput. Sci.
quire a substantial amount of user interac- ing is complete the gradient run is commenced. 30, 492–504 (1990).
tion. For example, the crude material must be During the gradient run the chromatography 4. M. H. S. Segler, M. Preuss, M. P. Waller, Nature 555, 604–610 (2018).
manually loaded onto the column and the machine continuously reads the detector sig- 5. O. Engkvist et al., Drug Discov. Today 23, 1203–1218 (2018).
6. B. A. Grzybowski, K. J. Bishop, B. Kowalczyk, C. E. Wilmer,
product fractions must be manually identi- nals and sends them to the ChemPU con- Nat. Chem. 1, 31–36 (2009).
fied, washed out of the fraction vials, and troller software in real time. 7. I. W. Davies, Nature 570, 175–181 (2019).
combined. Further, these commercial systems The ChemPU controller then performs the 8. N. Matosin, E. Frank, M. Engel, J. S. Lum, K. A. Newell,
Dis. Model. Mech. 7, 171–173 (2014).
require user interactions at several different peak detection and triggers the fraction col-
9. M. Trobe, M. D. Burke, Angew. Chem. Int. Ed. 57, 4192–4214
stages, thus tying the chemist to the lab even lection mechanism of the chromatography (2018).
if it is only for a trivial task such as loading machine. The controller also keeps track of 10. J. Li et al., Science 347, 1221–1226 (2015).
the sample onto the column. To integrate the fraction vial filling levels and various run pa- 11. C. W. Coley et al., Science 365, eaax1566 (2019).
12. T. Jiang et al., Chem. Sci. 12, 6977–6982 (2021).
Buchi Pure C-815 chromatography system rameters such as back pressure buildup, sol- 13. A. C. Bédard et al., Science 361, 1220–1225 (2018).
with the ChemPU, two auxiliary hardware vent vapor levels, and solvent levels of the 14. S. Steiner et al., Science 363, eaav2211 (2019).
units were built: a column carousel that allows gradient solvents and the solvent waste drum. 15. S. H. M. Mehr, M. Craven, A. I. Leonov, G. Keenan, L. Cronin,
Science 370, 101–108 (2020).
preinstallation of different columns on the If any of these parameters exceed the specified 16. D. Angelone et al., Nat. Chem. 13, 63–69 (2021).
system and an extension to the fraction tray. threshold, an appropriate error-handling rou- 17. P. G. Nantermet, Chem 1, 335–336 (2016).
The latter allows for recovery of the product tine is initiated that pauses the chromato- 18. Z. Wang, W. Zhao, G. F. Hao, B. A. Song, Drug Discov. Today
25, 2006–2011 (2020).
fraction by the ChemPU. The first opera- graphic separation in a controlled way. When 19. R. B. Merrifield, Science 150, 178–185 (1965).
tion that is challenging to automate is the the separation run is complete the product 20. O. J. Plante, E. R. Palmacci, P. H. Seeberger, Science 291,
sample loading onto the column. The labora- peak is identified and transferred to the next 1523–1527 (2001).
21. G. Alvarado-Urbina et al., Science 214, 270–274 (1981).
tory-based chemist usually chooses between module (usually the rotary evaporator). The 22. M. Legrand, P. Bolla, J. Automat. Chem. 7, 31–37 (1985).
dry-loading and liquid injection of the sam- crude material is typically transferred from 23. S. B. Boga et al., React. Chem. Eng. 2, 446–450 (2017).
ple. We aimed to implement the liquid injec- the rotary evaporator to the chromatography 24. B. Burger et al., Nature 583, 237–241 (2020).
25. A. G. Godfrey, T. Masquelin, H. Hemmerle, Drug Discov. Today
tion method which ties in nicely with the module followed by transfer of the purified
18, 795–802 (2013).
liquid-handling backbone of the ChemPU; product back from the chromatography mod- 26. B. P. MacLeod et al., Sci. Adv. 6, eaaz8867 (2020).
further, the liquid injection sample loading ule to the rotary evaporator, and as such the 27. H. Okamoto, K. Deuchi, Lab. Robot. Autom. 12, 2–11
method entailed little process development, rotary evaporator flask needs to be cleaned in (2000).
28. A. Orita, Y. Yasui, J. Otera, Org. Process Res. Dev. 4, 333–336
requiring only the identification of a suitable between. Hence an optional cleaning routine (2000).
solvent mixture and volume to dissolve the for the target vessel of the purified product has 29. Y. Tanaka, S. Fuse, H. Tanaka, T. Doi, T. Takahashi, Org.
crude material. The second challenge to full been implemented and can be performed Process Res. Dev. 13, 1111–1121 (2009).
30. S. Chatterjee, M. Guidi, P. H. Seeberger, K. Gilmore, Nature
automation of normal-phase chromatography during the chromatographic separation. The 579, 379–384 (2020).
is to reliably select the product peak. Usually, integrated chromatographic separation was 31. A. J. S. Hammer, A. I. Leonov, N. L. Bell, L. Cronin,
chemists need to analyze individual fractions used for three reactions. The process of these Chemputation and the Standardization of Chemical
Informatics. JACS Au 1, 1572–1587 (2021).
by thin-layer chromatography, mass spectrom- chromatographic separations has been cap- 32. L. Wilbraham, S. H. M. Mehr, L. Cronin, Acc. Chem. Res. 54,
etry, or NMR after chromatographic separa- tured by cDL, specifying every minute and 253–262 (2021).
tion. For the ChemPU integration of the critical detail in a concise, easy-to-understand 33. M. Craven, G. Keenan, A. Khan, M. Lee, L. Wilbraham,
ChemIDE. https://croningroup.gitlab.io/chemputer/xdlapp/
module several alternative options were conway. Hence, reproduction of the chromato- (2021).
sidered. We found that considering the UV/ graphic separations on another ChemPU or 34. Cronin Group, Chemify Database: https://croningroup.gitlab.
visible response or the signal from the elastic equivalent system or even manually with a io/chemputer/chemify-database/.
light scattering detector of the eluting frac- commercially available chromatography machine 35. S. D. Roughley, A. M. Jordan, J. Med. Chem. 54, 3451–3479
(2011).
tions and choosing the peak with the largest is readily possible. 36. N. I. Vasilevich, R. V. Kombarov, D. V. Genis, M. A. Kirpichenok,
area under the curve for a specified signal trace J. Med. Chem. 55, 7003–7009 (2012).
gave the best trade-off between reliability and Outlook 37. N. Schneider, D. M. Lowe, R. A. Sayle, M. A. Tarselli,
G. A. Landrum, J. Med. Chem. 59, 4385–4402 (2016).
flexibility; for a given well-performing reac- We have shown how the chemical synthesis
38. J. S. Carey, D. Laffan, C. Thomson, M. T. Williams, Org. Biomol. Chem.
tion the product peak can correctly be identi- literature can be easily converted to a univer- 4, 2337–2347 (2006).
fied independent of the exact retention time. sal chemical code that can run on any robot 39. P. S. Baran, J. Am. Chem. Soc. 140, 4751–4755 (2018).
Moreover, this method does not rely on more capable of chemputation; the only require- 40. A. Isidro-Llobet, M. Alvarez, F. Albericio, Chem. Rev. 109,
2455–2504 (2009).
elaborate product identification such as mass ments for this are a batch reactor, a separator, 41. A. G. Volbeda, G. A. Marel, J. D. C. Codée, in Protecting Groups.
spectrometry or NMR. evaporator, and purification system. This means S. Vidal, Ed. (Wiley, 2019), pp. 1–27.

42. R. A. Fernandes, P. Kumar, P. Choudhary, Chem. Commun. 56, (766975 MADONNA), and DARPA (W911NF-18- 2-0036, W911NF- analytical data for all experiments are available at Zenodo (44).
8569–8590 (2020). 17-1-0316, and HR001119S0003). Author contributions: L.C. License information: Copyright © 2022 the authors, some
43. W. R. Roush et al., Organic Syntheses (2021); http://www. conceived the concept, architecture, and programming approach. rights reserved; exclusive licensee American Association for the
orgsyn.org/. S.R., M.Š., A.P., M.S., H.M.M., E.T., A.I.L., and A.H. configured Advancement of Science. No claim to original US government
44. S. Rohrbach synthesis literature database in the Chemputer, the robots, ran the synthetic protocols, and characterized the works. https://www.sciencemag.org/about/science-licenses-
Zenodo (2022); doi:10.5281/zenodo.6534009. products. G.K. and A.K. helped with the development of journal-article-reuse
the database and integration with the ChemIDE. L.C. wrote the
ACKN OW LEDG MEN TS paper together with S.R., M.Š., and G.C. with help from all SUPPLEMENTARY MATERIALS
We thank BUCHI for supplying us with a pure C-815 authors. Competing interests: L.C. is the founder of Chemify Ltd.
science.org/doi/10.1126/science.abo0058
chromatography system and API to interface it with the Chemputer L.C. is listed as an inventor on the UK patent GB 2209476.7., which
Materials and Methods
software package. The authors gratefully acknowledge the describes this system. Data and materials availability:
Figs. S1 to S143
assistance of D. Doran and V. Sandoval in the preparation of and Supplementary materials include full details to reproduce this
Tables S1 to S34
interfacing with the cDL Chemify database. Funding: We gratefully work, including instructions for how to build and run the
References (45–101)
acknowledge financial support from the EPSRC (EP/L023652/1, platform. Additional details of the electronic and mechanical
EP/R020914/1, EP/S030603/1, EP/R01308X/1, EP/S017046/1, components of the platform, videos of the platform working, and Submitted 6 January 2022; accepted 26 May 2022
and EP/S019472/1), the ERC (670467 SMART-POM), the EC the software to produce and run the cDL files, and the raw 10.1126/science.abo0058

RES EARCH
BIOENGINEERING unobtainable via current biofabrication tech-

niques. Basing our study on Sallin’s analytical
Recreating the heart’s helical structure-function model of the heart (3), we used FRJS to man-
ufacture both helically aligned (HA) and cir-
relationship with focused rotary jet spinning cumferentially aligned (CA) tissue-engineered
models of the left ventricle, showing that the
Huibin Chang1†, Qihan Liu1,2†, John F. Zimmerman1†, Keel Yong Lee1, Qianru Jin1, Michael M. Peters1, biomechanics of these single-layer ventricle
Michael Rosnach1, Suji Choi1, Sean L. Kim1, Herdeline Ann M. Ardoña1,3, Luke A. MacQueen1, models is consistent with theoretical predictions
Christophe O. Chantre1, Sarah E. Motta1,4, Elizabeth M. Cordoves1, Kevin Kit Parker1* from the past 50 years (3, 4, 25). Collectively, this
work reveals how a FRJS-based approach to
Helical alignments within the heart’s musculature have been speculated to be important in achieving biofabrication can be used to rapidly create
physiological pumping efficiencies. Testing this possibility is difficult, however, because it is challenging robust model organs, with increased geomet-
to reproduce the fine spatial features and complex structures of the heart’s musculature using ric complexity.
current techniques. Here we report focused rotary jet spinning (FRJS), an additive manufacturing
Fiber manufacture using focused rotary
approach that enables rapid fabrication of micro/nanofiber scaffolds with programmable alignments in
jet spinning
three-dimensional geometries. Seeding these scaffolds with cardiomyocytes enabled the biofabrication
of tissue-engineered ventricles, with helically aligned models displaying more uniform deformations, Methods such as electrospinning (26), melt
greater apical shortening, and increased ejection fractions compared with circumferential alignments. blowing (27), pull spinning (28), and tradi-
The ability of FRJS to control fiber arrangements in three dimensions offers a streamlined approach to tional centrifugal spinning (29) can form
fabricating tissues and organs, with this work demonstrating how helical architectures contribute to micro/nanofibers. When producing single-
cardiac performance. micrometer features, these techniques can
offer orders of magnitude greater through-
T
puts than those achieved by 3D extrusion
he heart’s musculature is organized in a helical alignments are supported by extra- printing (30). However, fiber-spinning meth-
helical fashion, with cardiomyocytes in cellular matrix (ECM) proteins such as collagen ods are often less precise, failing to accurately
the left ventricle smoothly transitioning fibrils (8, 16), the diameters of which are on the recreate complex 3D geometries and align-
transmurally from a left- to right-handed order of a single micrometer. Approaches that ments. This lack of precision is partly be-
helix (1). This helical alignment results in use 3D extrusion printing have demonstrated cause fiber formation and patterning are
a “wringing motion” that was first described in important milestones toward replicating such traditionally interrelated during fiber pro-
1669 by Lower (2). Over the past half-century, it structures, including the production of micro- duction. In FRJS, decoupling the fiber forma-
has been argued that this helical arrangement physiological devices (17, 18), microvasculature tion and patterning processes is realized by
represents a fundamental structural design crit- systems (19), and spontaneously beating heart creating a focused stream of preformed fibers
ical to achieving large ejection fractions (EFs) models (20, 21). However, reproducing the fine (Fig. 1, A to C; fig. S1, and movie S1). Rotary jet
(1, 3–6). Although in vivo studies have proven spatial features required to potentiate muscu- spinning produces fibers by centrifugal force,
to be a powerful tool, both in defining these lar alignment while retaining practical produc- pushing polymer solutions through a small ori-
structural features (7–9) and in correlating mis- tion rates is difficult. This is the result of a fice in the spinneret. Subsequent jet elongation
alignments with cardiac disease and reduced fundamental limitation in 3D extrusion print- results in free-floating single-micrometer fibers
ventricle function (5, 9, 10), they are often char- ing, whereby throughput declines rapidly with (29). This allows for a formation period that is
acterized by concomitant changes in protein respect to feature size. Consider a full-size hu- independent of fiber patterning, generating a
expression and metabolism (11) and are limited man heart. Printing the ECM components at cloud of fibers that surrounds the spinneret.
by their ability to control cardiac alignments current resolutions (~250 mm feature sizes) Next, in a phenomenon known as entrainment
(12, 13). This makes it difficult to distinguish takes hours to days (22), but at native feature (31), fibers are pulled into a jet stream blown
between biomolecular and biomechanical con- sizes (1 mm) could take hundreds of years if from the center of the spinneret. Because air
tributions to cardiac dysfunction. However, current scaling trends are followed (eq. S1). speeds are orders of magnitude slower outside
understanding how helical structures contrib- Because fiber-spinning techniques can repro- of the jet (fig. S2), this entrainment process
ute to cardiac function is important, as some duce these fine spatial features with higher minimally perturbs fiber formation (fig. S3).
cardiomyopathies can exhibit maladaptive tis- throughputs, they offer a potential solution The jet stream then allows for fibers to become
sue remodeling (14, 15), which may result in and have been used previously to engineer aligned and confined in a small region, focus-
more circumferential alignments of the mus- tissue scaffolds, such as heart valves (23) and ing them for patterning.
culature (12, 13, 15). ventricle models (24). However, unlike 3D Focusing the fibers in this manner enables
To study the role of helical muscular align- printing, fiber-spinning approaches often fail them to be conformally deposited, meaning
ment on cardiac function, it is necessary to to recreate complex 3D geometries while main- that fibers will adhere well to convex struc-
recreate the multiscale architectures of the taining controlled alignments. tures or features with curvature radii that
heart, with both controlled alignments and We developed focused rotary jet spinning exceed the focal spot size (fig. S4). To dem-
three-dimensional (3D) geometries. In the heart, (FRJS), an additive manufacturing method onstrate this focusing effect in FRJS, poly-
that uses centrifugal jet spinning to rapidly caprolactone fibers were collected at regular
1
Disease Biophysics Group, John A. Paulson School of form polymeric micro/nanofibers, which are intervals from within the airstream to mea-
Engineering and Applied Science, Harvard University, Boston, then focused and spatially patterned by sure deposition profiles. This showed that, at
MA 02134, USA. 2Department of Mechanical Engineering and means of a controlled airstream. This approach the narrowest point, 95% of fiber deposition
Materials Science, University of Pittsburgh, Pittsburgh, PA
15261, USA. 3Department of Chemical and Biomolecular allows for the rapid manufacturing of fiber occurred within a spot size of 5.0 ± 0.3 cm (2s
Engineering, Samueli School of Engineering, University of constructs with programmable fiber alignments of Gaussian distribution) (Fig. 1B, ii, and fig. S4).
California, Irvine, CA 92697, USA. 4Institute for Regenerative in three dimensions. Because fibers can be used To demonstrate conformal deposition, fibers
Medicine, University of Zurich, Zurich, Switzerland.
*Corresponding author. Email: kkparker@seas.harvard.edu to direct tissue formation, it is possible to re- were then spun onto both a rotating collection
†These authors contributed equally to this work. create complex anatomies that were hitherto mandrel (movie S2) and a 1-cm ventricle model
Chang et al., Science 377, 180–185 (2022) 8 July 2022 1 of 6

Fig. 1. Focused rotary jet A B C

spinning for producing helical
structures. (A) Schematic dia-
gram of the helical alignment of a
human heart. (B and C) FRJS uses
focused air to separate fiber
manufacture into formation (B, i)
and patterning (C) phases,
allowing for controlled alignments
during deposition. (B, ii) Differen-
tial contrast projection of the fiber
stream (maximal projection; D
scale bar, 5 cm). (D) Image of
polycaprolactone spun onto a
mandrel (scale bar, 5 mm), with
corresponding scanning electron
microscopy (SEM) image,
showing aligned fiber formation
(mean fiber diameter, ~900 nm;
scale bar, 5 mm). (E) Schematic
E F
diagram showing that collection
angle (q) dictates fiber deposition
alignment (i), with OOP indicating
the relative average alignment
(error bars, mean ± SD) (ii).
SEM micrographs (iii to v; 0°, 60°,
and 90°, respectively) with
corresponding 2D Fourier trans-
forms inset, indicating that
the degree of alignment is based
on collection angle (scale bars,
5 mm). (F) Schematic diagram showing HA fiber manufacture based on angle (i), a, with a representative fiber-coated rod (ii) and reconstructed
micro–computed tomograph of a HA scaffold (iii; scale bar, 200 mm).
(Fig. 1D). In each case, targets were coated in and randomly aligned (q = 90°, OOP = 0.18 ± hypothesis (3) and the advantages of helically
~5 min, forming standalone structures that 0.03) fiber distributions, respectively. versus circumferentially aligned ventricular
could be removed and manipulated. Addition- This ability to control fiber orientation sug- geometries. Basing our experimental approach
ally, we showed that this approach could be gested that more complex patterning could be Sallin’s analytical model of the heart (3), we
used with a variety of material compositions, achieved by moving the target relative to the then used FRJS to produce both CA (a = 0°)
such as nylon, polyurethane, and gelatin, while stream. To test whether FRJS could recreate and HA (a = 30°) (60° with respect to the ven-
maintaining single-micrometer fiber diame- key geometric features of the heart, such as the tricle’s long axis) single-layer models of the
ters (fig. S5). helical alignments and laminar tissue struc- left ventricle (Fig. 2A). HA fiber orientations
tures identified in a rodent model (fig. S6, A were selected on the basis of energy minimi-
Controlled distribution of fiber alignment and B), fibers were collected on an inclined zation constraints, with angles between 20°
Envisioning that micro/nanofibers could mimic rotating cylinder to generate helical align- and 45° predicted to be the most energy ef-
the structural features of ECM proteins, we ments (Fig. 1F and movie S3) and on an ficient for pumping fluids (4). Fiber scaffolds
examined how FRJS could be used to control incrementally rotating disk to generate mul- were designed to have mechanical proper-
the anisotropic distribution of fiber align- tilayered fiber sheets (fig. S6C and movie S4). ties consistent with those of human heart
ments. We hypothesized that the alignment of We observed that FRJS was able to accurately tissues (fig. S7 and table S2).
fibers in the air stream could enable controlled reproduce these structures, as confirmed by After fiber manufacture, gelatin fiber scaf-
deposition, in which tangential collection (q = x-ray micro–computed tomography (mCT). folds were then seeded with either primary
0°) should minimally perturb airflow, while This suggested that FRJS could be effectively neonatal rat ventricular myocytes (NRVMs)
head-on deposition (q = 90°) would enable employed in hierarchical biofabrication, in or human induced pluripotent stem cell–
divergent patterning. To test this hypothesis, which the focused air stream provides gross derived cardiomyocytes (hiPSC-CMs), result-
the collector angle relative to the fiber stream structural morphology (centimeter scale), ing in confluent tissues with depths between
was modulated during deposition (Fig. 1E), whereas fibers provide fine structural (single- one to three layers of cardiomyocytes (fig. S8).
with the orientation order parameter (OOP) micrometer scale) cues to promote tissue hiPSC-CMs were differentiated in vitro for
used as a metric of subsequent fiber organi- morphogenesis. 15 days total before seeding, forming visible
zation (32). The resulting fibers displayed spontaneous contractions after 7 days and
angle-dependent anisotropy, with tangential, Building helically aligned models of the staining positive for sarcomeric a-actinin
intermediate, and perpendicular collection lead- left ventricle (fig. S9). To ensure that FRJS fiber spinning
ing to highly anisotropic (q = 0°, OOP = 0.60 ± We reasoned that the ability of FRJS to control could be used to direct tissue alignment, we
0.15), intermediate (q = 60°, OOP = 0.50 ± 0.08), fiber alignment could be used to test Sallin’s first examined NRVMs seeded onto laminar

Fig. 2. Tissue scaffolds with A B D

controlled helical alignments.
(A) Bright-field micrograph of a
HA ventricle model (gelatin fibers
seeded with NRVMs; scale bar,
2 mm). (B) SEM micrograph of
fibers from CA [(i), a = 0°] and HA
[(iii), a = 45°] cylinders, with
corresponding immunofluorescent
staining of cardiomyocytes
(ii and iv) [NRVM; blue, DAPI (4′,6-
diamidino-2-phenylindole); green,
f-actin; red, sarcomeres], showing
that fibers help direct tissue
alignment (scale bars, 50 mm). C E
Circ., circumferential. (C) Iso-
chrones (top) with corresponding
still frames (bottom), indicating
calcium transience along an
extended ventricle surface,
showing increased transverse
wave propagation for HA
scaffolds. Tissues were point-
stimulated apically [(i), CA; (ii),
HA] (scale bars, 5 mm).
(D) Schematic diagram of a CA
(left) and HA (right) ventricle,
illustrating differences in wall
displacement during contraction.
CA contracts as concentric rings,
whereas HA follows a wringing motion, resulting in different predicted EFs [k, strain, q, apical shortening, H(a), radial shortening]. (E) Deformation maps generated by
CA (left) and HA (right) ventricle models during contraction (scale bar, 2 mm).
tissue constructs. This resulted in tissues with spect to transverse CVs (TCVs), with LCVs systole, while connecting transversely as a bulk
a highly anisotropic distribution and cell align- and TCVs of 14.9 ± 5.8 cm/s and 9.0 ± 3.7 cm/s, material (37), they generate forces primarily
ment that conformed to the underlying fiber respectively (n = 9 samples) at ratios of ~1.5 to along the length of the fibers. This means that
orientation (Fig. 2B and fig. S10). Next, after 2.0. These values were consistent with previ- longer fibers should result in greater total dis-
seeding 3D scaffolds with either NRVMs ous reports of in vitro tissues composed of im- placements. For CA ventricles, which are made
(movie S5) or hiPSC-CMs (movie S6), spon- mature cardiomyocytes, where factors such as up of concentric rings, fiber length is greatest
taneous contractions were observed after 3 to geometry, calcium handling, and gap junction at the base, where the radius is largest, and
5 days. These data demonstrate that FRJS could expression can influence CV (24, 34, 35). Turn- smallest at the apex (Fig. 2D). This should re-
be used to form 3D contractile cardiac models. ing to 3D models, ventricles with extended sult in nonuniform deformations, with greater
basal regions, allowing for increased travel dis- displacements in the basal region and negligible
Calcium wave propagation tances, were apically stimulated. Isochrones displacements at the apex. For HA ventricles,
To demonstrate syncytium formation, we exam- of the calcium propagation revealed that CA however, fiber length should vary minimally, as
ined the ability of the model ventricles to sustain scaffolds displayed only modest CVs (8.3 cm/s) fibers extend uniformly from the apex to the
uniform calcium wave propagation. In healthy along the ventricle’s long axis, whereas HA base of the ventricle. Overall, this should result
tissues, action potentials propagate faster in the ventricles showed increased CVs (19.1 cm/s) in more homogenous deformations for HA ven-
direction of cell alignment, with myocytes show- with respect to CA ventricles (Fig. 2C and tricles, with increased apical shortening.
ing slower conduction velocities (CVs) in the movie S7). Together, these findings dem- Examining how CA and HA models deformed
transverse direction (33). Given this insight, onstrated that our tissues maintained their during contraction, we visualized displacement
we reasoned that electrical signal propaga- alignment and confluence over the length of by submerging these ventricles into a solution
tion should vary depending on the ventricle’s the ventricle (~1 cm) and further suggested containing nonspecifically adherent fluorescent
fiber alignment and could be used to confirm the importance of cardiomyocyte alignment beads. Using digital image cross-correlation,
long-range tissue formation and directional in regulating the spatiotemporal control of we performed deformation mapping across
alignment in these model systems. To mea- excitation-contraction coupling. the ventricle’s surface during 1-Hz field stim-
sure directional signal conduction in tissue ulation (movie S8). As predicted, CA ventri-
scaffolds, laminar tissues were electrically stim- Ventricular deformation cles displayed greater deformations near the
ulated (corner point stimulation), and the re- It has been hypothesized that HA ventricles base of the ventricle, whereas HA scaffolds
sulting calcium propagation was measured would display increased apical shortening (3, 36) showed more-uniform deformations (Fig. 2E).
with an optical mapping system (fig. S11). We and reduced basal displacement (9, 36), owing to Probing further, we measured changes in the
observed higher CVs in the direction of fiber differences in myofiber lengths. Because cardio- boundary shape using an elliptical fit, indi-
alignment [longitudinal CVs (LCVs)] with re- myocytes contract along their long axis during cating substantial differences in deformation

Fig. 3. Alignment dictates ven- A B

tricular ejection fractions.
(A) (i) Bright-field micrograph of
a tissue-engineered ventricle,
with (ii) a magnified view of the
ventricular basal region. ROI, D
region of interest. (iii) Maximum
intensity projection of fluorescent
beads taken over a single con-
traction cycle, showing particle
displacement. (iv) High-magnification
image of the boxed region in
(iii); the arrow indicates the direc-
C
tion of fluid displacement [scale
bars in (i) to (iii), 2 mm; in (iv),
0.5 mm]. (B) Side view of the
E
ventricle (scale bar, 10 mm).
(C) (i) Schematic diagram of PIV
measurement, with velocity fields
[(ii) and (iii)] taken from the base
of a HA ventricle scaffold during
peak systole [(ii), t = 0.3 s]
and diastole [(iii), t = 0.7 s].
(D) Representative measure-
ments of the instantaneous mass
flux (Nflux) in the region of interest as function of contraction time and ventricle angle. (E) Ensemble measurements of EF for CA and HA ventricle scaffolds
(n = 8 ventricles for each angle; *P < 0.05 by Student’s t test; box plot given in quartiles).
between each alignment. CA samples showed In a healthy heart, kinetic energy is stored lowed for the construction of 2D velocity fields
greater basal and minimal longitudinal con- during contraction in the sarcomeric protein surrounding the basal opening, as shown for a
striction, whereas HA ventricles displayed sig- titin and is subsequently released during re- HA ventricle model (Fig. 3, A and B). Using PIV,
nificant apical shortening with reduced basal laxation (38), giving rise to rotational displace- we then evaluated the instantaneous mass flux
deformation (n ≥ 5 ventricles for each condition) ments, or ventricular twist (1, 38). HA and CA resulting from ventricle contraction (fig. S15 and
(fig. S12). Overall, these results were consistent scaffolds were sutured at the base to a fixed movie S10) and observed cyclic outputs, with
with the projections of previous analytical support, allowing the apex to move freely, and fluid being expelled during systole and refill
and computational models (3, 25, 36) in that were monitored from below during field stim- occurring during diastole (Fig. 3, C and D).
they showed more-homogeneous deformations ulation. Using edge features to detect rotation This was performed for ventricles with both
for HA ventricles. (fig. S13, C to E), we observed that CA scaf- varying fiber alignments and fibroblast com-
These changes in ventricular contraction folds showed minimal twist (1.35 ± 1.1°, n = 7 positions (fig. S16). Summation of the fluid dis-
were also suggestive of further functional dif- ventricles). Conversely, HA scaffolds displayed placement over systole yields the total CO, with
ferences between myofiber alignments. For approximately four times as much twist (5.4 ± values of 11.7 ± 8.9 ml/s and 24.3 ± 13.5 ml/s (n =
instance, apical shortening indicated the 3.6°, n = 7 ventricles) (movie S9), modeling phys- 8 samples each) for CA and HA ventricle scaf-
potential for a wringing motion or ventricle iological ventricular twist in an in vitro system. folds, respectively (fig. S15G). This observation
twist to occur in HA ventricles. Additionally, represented a significant (P < 0.05) CO increase
this suggested the possibility for variable EFs Cardiac output and ejection fractions based purely on ventricular tissue alignment.
on the basis of fiber alignment, as these dif- To determine whether these structural changes Normalizing the CO by diastolic ventricle vol-
ferences in deformation could result in dis- in alignment and deformation lead to altered ume and fluid density, we calculated the result-
tinct volumetric displacements. It has been function, we then used cardiac output (CO) and ing EFs. This yielded average EFs of 1.6 ± 1.1%
predicted that EFs should scale as 1 – k2 for EF as quantitative metrics of cardiac perform- and 3.3 ± 1.7% in the CA and HA case, respec-
CA ventricles (eq. S5) and 1 k2 q H ðaÞ for ance. To measure these values, we first used tively (n = 8 samples each), with a maximum
HA ventricles (eq. S15) (Fig. 2D) (3), with k catheterization to monitor pressure-volume EF of 5.9% observed in the helical case (Fig. 3E).
being the strain along the fiber axis, q being changes in the ventricle scaffolds, observing This indicated that helical alignments confer a
the apical shortening, and H ðaÞ being the the formation of complete oblate loops (fig. relative increase in ventricle output. To compare
radial shortening based on the fiber’s helical S14). However, submerged gelatin fibers are our findings with Sallin’s analytical model of
angle (eqs. S5 to S15). q and H ðaÞ are defined a poor dielectric, making it difficult to ob- ventricle contraction (3), we then normalized
as ≤1, which suggests that HA ventricles could tain consistent results in our synthetic scaf- our results on the basis of relative contractile
produce larger EFs. folds by means of conductance catheterization. strain (eqs. S5 to S15). Sallin’s model predicted
Consequently, particle imaging velocimetry an EF increase of 54 to 64% for HA ventricles,
Ventricular twist (PIV) was used to further quantify cardiac which was consistent with our experimental
To determine whether ventricular twist was performance. observations (within the standard error of the
preserved in our model system, we measured Model ventricles were suspended in a bath mean; Fig. 3E). This indicated that our model
rotational displacement at the apex of sus- containing neutrally buoyant fluorescent beads, system was capable of preserving these fun-
pended ventricle scaffolds (fig. S13, A to B). and bead displacement was tracked. This al- damental scaling laws, signifying that helical

Fig. 4. Multiscale heart models.

(A) Simplified design of a trilay- A B C
ered DCV that mimics the native
ECM alignment of the heart,
highlighting the four-step
manufacturing process.
(B) (i) mCT imaging of the DCV
suspended on the collection y
mandrel, with corresponding cor-
onal cross sections (ii) (scale
bars, 5 mm). (iii to v) High-
magnification mCT images
taken from the intraventricular D
septum (highlighted in purple),
showing trilayer alignments (scale F
bars, 25 mm). RV, right ventricle;
LV, left ventricle. (C) DCV in
culture seeded with NRVMs,
as viewed from the side [(i), E
coronal] and looking into the two
chambers [(ii), transverse]. (D) 3D
extrusion printing scales as a
power law with respect to feature
size, showing that the throughput
of FRJS is ~106 times greater
than that of 3D extrusion printing
for single-micrometer features.
(E) Left ventricle width for different species. (i to iv) Single-layer ventricles of different sizes, which can be rapidly manufactured owing to increased fiber
production rates, while maintaining a single-micrometer feature scale [scale bars (left to right), 5 mm; 15 mm; 4 cm; and 8 cm). (F) Full-scale four-chambered
human heart model composed of single-micrometer fibers (scale bar, 2 cm).
alignments of the ventricular myocardium re- aligned tissues (fig. S18B). Additionally, cal- terning micro/nanofibers onto dissolvable
sult in increased EFs. cium imaging revealed sustained wave propa- collectors in the shape of each of the heart’s
gation across the ventricle surface, primarily four chambers (fig. S20). Individual chambers
Dual-chambered ventricles and full-scale in the direction of fiber alignment, indicating were then connected by chemical annealing
heart models the formation of a continuous cardiac syncytium before the interior supports were dissolved,
Scaling Sallin’s model to physiological relevant (n ≥3 samples for each condition) (fig. S18, C resulting in standalone fiber scaffolds (Fig.
strains (~15 to 20%) (3), we noted that single- and D). Overall, these observations indicated 4F). Although full-size anatomical models
angle ventricles could achieve EFs of only up that fiber scaffolds can also support human stem have previously been produced with thermo-
to ~43% (a = 30°) (see supplementary mate- cell–derived tissues in complex geometries. plastics and hydrogels (22), these systems
rials for details). This EF is comparable to We then spun fibers onto ventricle-shaped typically lack the micrometer-scale features
values experienced during borderline heart fail- targets of varying size, ranging from ~1 to 20 cm needed to direct myocyte alignment. With our
ure and underscores the importance of multi- in diameter (fig. S19 and movie S16). In each fiber-based approach, these local structures
ple helical alignments in maintaining healthy case, targets were coated in <35 min with can be preserved across entire tissue volumes,
cardiac function. To examine whether we could micro/nanofibers, generating conformal coat- allowing for the hierarchical assembly of tis-
mimic these architectures using FRJS, we man- ings on the exterior. Throughout this process, sues. These proof-of-concept 3D organ mod-
ufactured both a dual-chambered ventricle fibers were produced with a total through- els demonstrate that FRJS scaffolds support
(DCV) with multiple helical angles and a full- put rate of 0.1 g/min, or 0.03 ghm (grams per human-derived tissues and allow for rapid
scale human heart model (Fig. 4). DCVs were hole per minute). This throughput is compa- assembly of full-size models of the muscu-
fabricated using a multistage process, creat- rable to those of melt-blowing processes (27) lature. These key features mark fiber-based
ing inner and outer helical layers, with an inter- but is orders of magnitude greater (~106) than manufacturing as a promising approach for
mediate circumferential sheet reminiscent of that of 3D extrusion printing, the speed of achieving whole-organ biofabrication, which
the native myocardium (fig. S17, Fig. 4, A and which declines rapidly at single-micrometer can be used as an alternative to or in conjunc-
B, and movie S11). These competing helical features (as a power law, with a scaling factor tion with emerging biomanufacturing plat-
structures were confirmed with mCT, indicat- of ~2.8) (Fig. 4, D and E, and eq. S1). Collectively, forms such as 3D extrusion printing.
ing three distinct transmural layers along the these observations demonstrated that FRJS
septal wall (Fig. 4B and movies S12 and S13). is amenable to rapidly manufacturing fiber Discussion
DCVs were then seeded with either hiPSC- scaffolds spanning multiple length scales and Biofabrication using FRJS allowed for the rapid
CMs or NRVMs, forming contractile tissue con- can be adapted to different geometries. assembly of functional 3D ventricle models
structs (Fig. 4C, fig. S18A, and movies S14 to To demonstrate the ability of FRJS to enable capable of recapitulating emergent pheno-
S15). Examining excised segments of the left hierarchical biofabrication, a full-size model of mena in vitro, including ventricular twist,
ventricle, cultured with hiPSC-CMs, we ob- the human heart musculature was constructed. strain displacement, and myocardial angle-
served that cells adhered to the fibers, forming This model was assembled by individually pat- dependent EFs. Although the metrics presented

here (e.g., EF) were greater (by a factor of ~3) 5. J. S. Davis et al., Cell 107, 631–641 (2001). 40. SeasDBG, SeasDBG/FRJS: Recreating the Heart’s Helical
than for recently reported in vitro ventricle 6. R. S. Stephenson et al., Clin. Anat. 29, 316–332 (2016). Structure-Function Relationship with Focused Rotary
Jet Spinning- Data Analysis, Zenodo (2022); https://doi.org/
models (24, 39), they were still substantially 7. P. Agger, R. S. Stephenson, J. Cardiovasc. Dev. Dis. 7, 47
10.5281/zenodo.6547775.
(2020).
lower than in vivo values (tables S2 and S3). 8. K. D. Costa, Y. Takayama, A. D. McCulloch, J. W. Covell,
This difference may be the result of limited Am. J. Physiol. 276, H595–H607 (1999).
AC KNOWL ED GME NTS
cardiomyocyte penetration into the fiber scaf- 9. E. D. Carruth, A. D. McCulloch, J. H. Omens, Prog. Biophys.
Mol. Biol. 122, 215–226 (2016). The authors thank A. G. Kleber for discussions regarding cardiac
folds (figs. S7 and S8). Consequently, further physiology, P. Campbell for rodent heart isolation, W. T. Pu
10. S. Puwanant et al., Circulation 121, 259–266 (2010).
work is needed to achieve full-scale de novo 11. J. Zhang, Clin. Exp. Pharmacol. Physiol. 29, 351–359
for providing hiPSC cell lines, and H.-Y. G. Lin for assistance
organ fabrication; this includes improved car- (2002). with micro-CT. Funding: H.A.M.A. thanks the American
Chemical Society for support through the Irving S. Sigal
diac maturation, vascularization, and incorpo- 12. P. Agger et al., J. Cardiovasc. Magn. Reson. 19, 93 (2017).
Postdoctoral Fellowship. This work was sponsored by the John
13. E. D. Carruth et al., J. Cardiovasc. Magn. Reson. 22, 21
rating consistent back-pressures (Frank-Starling A. Paulson School of Engineering and Applied Sciences at
(2020).
mechanism). There is also a need to generate Harvard University, the Wyss Institute for Biologically Inspired
14. M. L. McCain, S. P. Sheehy, A. Grosberg, J. A. Goss, Engineering at Harvard University, the Harvard Materials
the large numbers of cardiomyocytes and other K. K. Parker, Proc. Natl. Acad. Sci. U.S.A. 110, 9770–9775 Research Science and Engineering Center (DMR-1420570 and
diverse cell populations (e.g., neurons, endo- (2013). DMR-2011754), and the National Institutes of Health with the
15. C. L. Hung et al., J. Am. Coll. Cardiol. 56, 1812–1822 Center for Nanoscale Systems (S10OD023519) and National
thelial cells, fibroblasts) required to support (2010). Center for Advancing Translational Sciences (UH3TR000522
organ biofabrication. However, the work pres- 16. J. B. Caulfield, T. K. Borg, Lab. Invest. 40, 364–372 and 1-UG3-HL-141798-01). The content is solely the
ented here provides an initial pathway toward (1979). responsibility of the authors and does not necessarily represent
achieving hierarchical patterning while main- 17. J. U. Lind et al., Nat. Mater. 16, 303–308 (2017). the official views of the National Institutes of Health. Author
18. F. B. Coulter et al., Matter 1, 266–279 (2019). contributions: K.K.P. supervised the research. K.K.P., H.C.,
taining 3D cell alignment. Q.L., J.F.Z., and L.A.M. designed the study. Q.L. and H.C.
19. M. A. Skylar-Scott et al., Sci. Adv. 5, eaaw2459 (2019).
In addition to biofabrication, FRJS may serve 20. A. Lee et al., Science 365, 482–487 (2019). designed and produced the fiber-spinning platform. H.C. and
an important role in other additive manufac- 21. N. Noor et al., Adv. Sci. 6, 1900344 (2019). J.F.Z. manufactured, cultured, and performed the ventricle
turing applications, as it provides production 22. E. Mirdamadi, J. W. Tashman, D. J. Shiwarski, R. N. Palchesko, experiments. K.Y.L. performed optical mapping experiments.
A. W. Feinberg, ACS Biomater. Sci. Eng. 6, 6453–6459 H.C., Q.L., J.F.Z., and K.Y.L. analyzed the data. Q.L. performed
rates comparable to those of current industrial simulations. L.A.M., C.O.C., S.E.M., G.T., and E.M.C. performed
(2020).
processes while enabling micro/nanoscale 23. A. K. Capulli et al., Biomaterials 133, 229–241 (2017).
additional supporting experiments. S.L.K., H.A.M.A., S.C., and
feature sizes and controlled 3D alignments. Q.J. helped with animal protocols and stem cell culture to
24. L. A. MacQueen et al., Nat. Biomed. Eng. 2, 930–941
obtain cardiomyocytes. M.M.P. and H.C. performed SEM
This includes applications in which a ma- (2018).
experiments and analysis. M.R., H.C., J.F.Z., and C.O.C.
terial’s properties are determined by its 25. B. Baillargeon, N. Rebelo, D. D. Fox, R. L. Taylor, E. Kuhl, produced the full-size heart model. All authors discussed the
Eur. J. Mech. A 48, 38–47 (2014).
microstructure and alignment. The high results and contributed to the writing of the final manuscript.
26. D. Li, Y. Xia, Adv. Mater. 16, 1151–1170 (2004). Competing interests: Harvard University filed for intellectual
surface area–to–volume ratio of micro- and 27. Y. Kara, K. Molnár, J. Ind. Text. 10.1177/15280837211019488 property relevant to this manuscript, listing J.F.Z., Q.L., H.C.,
nanofibers makes them ideal candidates (2021). and K.K.P. as inventors (US Patent Application 17/421,047 and
for the controlled absorbance and release 28. L. F. Deravi et al., Macromol. Mater. Eng. 302, 1600404 US Provisional Patent Application 63/234,287). Data and
(2017). materials availability: All data are available in the manuscript,
of chemical species. This suggests that FRJS 29. M. R. Badrossamay, H. A. McIlwee, J. A. Goss, K. K. Parker, the supplementary materials, or Zenodo (40). License
may be an important technique for indus- Nano Lett. 10, 2257–2261 (2010). information: Copyright © 2022 the authors, some rights
trial manufacturing process that use hierar- 30. J. Go, S. N. Schiffres, A. G. Stevens, A. J. Hart, Addit. Manuf. reserved; exclusive licensee American Association for the
16, 1–11 (2017). Advancement of Science. No claim to original US government
chical design principles, for which control over
31. P. Davidson, Turbulence: An Introduction for Scientists and works. https://www.science.org/about/science-licenses-
several orders of spatial magnitude is needed. Engineers (Oxford Univ. Press, 2015). journal-article-reuse
32. F. S. Pasqualini, S. P. Sheehy, A. Agarwal, Y. Aratyn-Schaus,
RE FE RENCES AND N OT ES K. K. Parker, Stem Cell Rep. 4, 340–347 (2015).
33. M. Valderrábano, Prog. Biophys. Mol. Biol. 94, 144–168 SUPPLEMENTARY MATERIALS
1. P. P. Sengupta, A. J. Tajik, K. Chandrasekaran,
B. K. Khandheria, JACC Cardiovasc. Imaging 1, 366–376 (2007). science.org/doi/10.1126/science.abl6395
(2008). 34. S. Rohr, J. P. Kucera, A. G. Kléber, Circ. Res. 83, 781–794 Materials and Methods
2. R. Lower, Tractatus de Corde. Item de Motu & Colore Sanguinis (1998). Figs. S1 to S20
et Chyli in eum Transitu (Allestry, 1669); available at 35. A. G. Kléber, Q. Jin, Biophys. Rev. 2, 031301 (2021). Tables S1 to S3
https://www.woodlibrarymuseum.org/rare-book/lower- 36. N. B. Ingels Jr., Technol. Health Care 5, 45–52 (1997). References (41–54)
r-tractatus-de-corde-item-de-motu-colore-sanguinis-et-chyli-in- 37. R. H. Anderson, P. F. Niederer, D. Sanchez-Quintana, MDAR Reproducibility Checklist
eum-transitu-1669/. R. S. Stephenson, P. Agger, J. Anat. 235, 697–705 (2019). Movies S1 to S16
3. E. A. Sallin, Biophys. J. 9, 954–964 (1969). 38. A. M. S. Omar, S. Vallabhajosyula, P. P. Sengupta, Submitted 26 July 2021; resubmitted 1 April 2022
4. A. Grosberg, M. Gharib, A. Kheradvar, Bull. Math. Biol. 71, Circ. Cardiovasc. Imaging 8, e003029 (2015). Accepted 16 May 2022
1580–1598 (2009). 39. M. E. Kupfer et al., Circ. Res. 127, 207–224 (2020). 10.1126/science.abl6395

RES EARCH
MEMBRANES membrane were known (1) (e.g., isoporous or

track-etched membranes in Fig. 1, A and B).
A two-phase model that unifies and extends More often, however, the pore-scale structure
is known, at best, only statistically (e.g., Fig.
the classical models of membrane transport 1C), instead requiring a Darcy-flow approach,
vs ¼ ∇p=x, wherein a local pressure gradient
Varun H. Hegde, Michael F. Doherty, Todd M. Squires* ∇p drives a local solvent velocity vs, where x is a
local solvent/matrix friction coefficient. Be-
Two models describe solvent transport through swollen, nonporous membranes. The pore-flow cause solvent only occupies a volume fraction
model, based on fluid mechanics, works for porous membranes, whereas the solution-diffusion fs within the membrane, the volumetric sol-
model invokes molecular diffusion to treat nonporous membranes. Both approaches make valid vent flux Js is a factor fs smaller than vs. The
arguments for swollen polymer membranes, but they disagree in their predictions of intramembrane macroscopic solvent flux through a membrane
pressure and concentration profiles. Using a fluid-solid model that treats the solvent and of thickness L is then given by
membrane matrix as separate phases, we show both classical models to be valid, to represent
fs Dp Dp
complementary approaches to the same phenomenon, and to make identical predictions. The fluid- Js ¼ ≡ kPF ð1Þ
solid model clarifies recent reverse osmosis measurements; provides a predictive and mechanistic x L L
basis for empirical high-pressure limiting flux phenomena, in quantitative agreement with classic
measurements; and gives a framework to treat nonporous but mechanically heterogeneous In practice, the friction coefficient x or per-
membrane materials. meability kPF is not determined independently
but rather is typically extracted from macro-
P
scopic permeance measurements. Explicit hydro-
olymer membranes play an essential role persistent diffusive motion relative to the per- dynamic calculations for simplified geometries
in both classic and emerging technolo- meating fluid, consistent with the SD mecha- [see the supplementary materials (SM)] relate x
gies, including water treatment, gas sep- nism (7–9). At the same time, solvent within to pore size Rp and solvent viscosity h, typically
arations, fuel cells, and flow batteries the permeating fluid networks may traverse via x ∼ h=R2p (18, 19).
(1, 2). Membranes generally fall into two such membranes without ever encountering The SD model describes solvent transport
broad classes, on the basis of morphological polymer, and so it would be described by fluid through nonporous membranes, whose uni-
structure and mechanism of action. Porous dynamics, consistent with the PF model (10–12). form structure and composition lack clear,
membranes (e.g., Fig. 1, A to C) are permeated Decades ago, a spirited debate emerged over continuous pores. This model assumes solvent
by fixed pores that permit the passage of some which approach is appropriate for swollen to diffuse down concentration gradients with-
species but not others, typically by means of membranes (7–12), which persists to this day. in the matrix, with flux
size exclusion and/or electrostatics (3). Exam- Variants of both models continue to be dis-
ples include the micro-, nano-, and ultrafiltration cussed and criticized, for example, in treating D Dfs
Js ¼ ð2Þ
of contaminated water (3). By contrast, non- reverse osmosis membranes (13, 14), with on- 1 fs L
porous membranes (e.g., Fig. 1D) are structurally going reframing of the classical models (15, 16).
homogeneous, with permeability and selectivity For example, recent neutron scattering studies where D is the binary diffusivity between solvent
determined by solubility and diffusivity within of water transport in polyamide reverse osmosis and polymer, and Dfs is the drop in solvent
the membrane. Examples include gas separation (PARO) membranes suggested SD to be invalid volume fraction between the feed and per-
membranes (3), organic solvent nanofiltration and motivated a proposed change in the stra- meate sides of the membrane (see the SM).
(4, 5), and reverse osmosis desalination (6). tegic design of future PARO membranes, toward Equation 2 is difficult to apply directly, be-
Two types of models are used to describe reducing tortuosity rather than enhancing selec- cause the solvent volume fraction profile fs is
transport through these two classes of mem- tive solubility and diffusivity (17). generally not known within the membrane.
branes. Porous membranes are treated using Here, we introduce a two-phase fluid-solid Instead, Dfs is related to the transmembrane
pore-flow (PF) models, in which pressure gra- (FS) model for solvent transport through swol- pressure drop Dp. Imposing contact equilib-
dients force fluid to flow within the membrane len, homogeneous membranes that reconciles rium for solvent molecules at the feed and
pores. Nonporous membranes are described the PF and SD mechanisms, revealing them to permeate boundaries of the membrane (de-
using solution-diffusion (SD) models, wherein be entirely consistent with one another, mean- scribed in the SM) gives
solute and solvent dissolve into the membrane ing that they make quantitatively identical pre- !

and diffuse down concentration gradients. dictions. Moreover, the FS model predicts the PF Dp Df0s V s @lnfs Dp
Each approach is built on a distinct transport and SD parameters to be connected through the Js ¼ kSD ≡
L 1 f0s RT @lnas f¼f0 L
mechanism: PF involves the mechanical forc- longitudinal osmotic modulus of the mem-
ing of fluid through porous membranes by brane, which can be independently measured. ð3Þ
means of a momentum balance, whereas SD In fact, FS model predictions are consistent with
describes the thermodynamic forcing down neutron scattering measurements of PARO where V s and as are the molar volume and
concentration gradients, as described by a mass membranes (17), indicating that neither PF activity of the solvent, respectively (2, 8).
balance (3). nor SD is inappropriate for these membranes. Both SD and PF approaches have been used
However, some membranes exhibit features to treat solvent transport through highly swol-
common to both SD and PF mechanisms. Background len membranes, and reasonable arguments
Swollen polymer membranes consist of statis- The PF model treats solvent flux through a can be made to support either. The fundamen-
tically homogeneous polymer meshes, in membrane as a pressure-driven flow through tally diffusive nature of even equilibrium swol-
a porous network with fixed geometry. In prin- len membranes, along with the lack of any
ciple, the Navier-Stokes equations could be clear pores, suggests the stochastic, diffusive
Department of Chemical Engineering, University of California,
Santa Barbara, CA, USA. solved to relate the solvent flux to the pressure approach of SD. However, polymer-free “tubes”
*Corresponding author. Email: tsquires@ucsb.edu gradient, if the particular morphology of the of solvent permeate highly swollen membranes,
Hegde et al., Science 377, 186–191 (2022) 8 July 2022 1 of 6

along which the hydrodynamic pressure must A B

decrease continually as fluid traverses the
membrane—no other element enters the mo-
mentum balance within the tube. From this
standpoint, the pressure-driven PF approach is
warranted.
Both SD and PF models predict flux rela-
tions of the same macroscopic form, Js = kDp/L,
allowing experimentally measured perme-
abilities (k) to be interpreted through either
model to determine Rp and x for PF, and D
and fs for SD. Without independent mea- 1 µm 300 nm
surements of those quantities, macroscopic
permeability measurements cannot differen- C D
tiate between the predictions of the two ap-
proaches. Support for SD came from the classic
Nonporous
experiments of Rosenbaum and Cotton (20),
which measured concentration gradients across
a multilayer stack of swollen gel membranes.
Additionally, Ham and colleagues (21) demon-
strated that the solvent concentration f0s at the Porous
support
feed-side boundary remains constant as pres-
sure increases. Likewise, Gehman (22) mea-
sured fs to decrease on the permeate side, as
500 nm 1 µm
the membrane was compressed against the
support layer. A number of apparent discrep- Fig. 1. SEM images of membrane surfaces. (A) Scanning electron microscopy (SEM) image of a track-
ancies arise between the PF and SD treatments etched polycarbonate membrane with 0.1-mm pores. (B) SEM image of an isoporous PS-P4VP [polystyrene-
of swollen membranes. First, the internal pres- poly(4-vinylpyridine)] membrane where the skin layer is composed of uniformly spaced nanoscale nodules.
sure and concentration profiles predicted by (C) SEM image of a commercial polystyrene-20 ultrafiltration membrane where the skin layer is composed
these two models appear to be at odds with of randomly spaced nanoscale nodules. (D) SEM image of a P1-8 asymmetric gas separation membrane
one another (Fig. 2, C and D). SD predicts the with a dense nonporous skin layer bound to a porous support.
pressure p to be constant within the mem-
brane but a steady gradient in solvent con-
centration fs (Fig. 2D). By contrast, PF predicts
a pressure gradient ∇p within the membrane
but a constant concentration fs (Fig. 2C). Sec- A B Water molecule
ond, the two models describe solvent transport
through fundamentally distinct physical mech-
anisms. The PF approach is mechanical and
deterministic in nature and assumes purely
convective flow, whereas SD is stochastic and
thermodynamic, driven by mutual diffusion
between polymer and solvent. Because the
binary diffusive flux adds to the volume-
averaged reference velocity in Eq. 2, SD trans-
port contains both diffusive and convective
components. However, the Fick’s law basis of
SD underscores the fundamental role of dif-
IMAGES: MOSTAFA NASSER (FIG. 1A); ALEXANDER BRIDGE (FIG. 1, B TO D)
fusion. Finally, the flux relations predicted by

the SD and PF models depend on physically
different, and seemingly unrelated, parameters. C Pore-flow model D Solution-diffusion model
Feed Membrane Permeate Feed Membrane Permeate
The fluid-solid model for gel dynamics
We now describe the FS model, building on µs
µs
the poroelastic theory developed by Biot and
Willis to describe soil compaction (23–25) p Δp p Δp
and extended to gel dynamics by Tanaka and
others (26–29). This approach has been used γscs γscs
to probe gel swelling and deswelling, drying,
crack propagation, and water permeation x = –L x=0 x = –L x=0
(29–31) and the dynamic swelling, buckling,
wrinkling and folding of thin polymer gel Fig. 2. Classical membrane transport models. (A and B) Pictorial representation of the pore-flow and
films (32). The FS model treats the solvent solution-diffusion models, respectively. (C and D) Solvent chemical potential ms, pressure p, and solvent
and polymer matrix individually, as distinct activity as profiles across a pore-flow and a solution-diffusion membrane, respectively.

but interpenetrating phases. The solvent phase that is compressed, rather than the polymer Tanaka and co-workers verified Eq. 8 experimen-
is a Newtonian liquid with local velocity field itself, with solvent flowing in or out as needed tally, using dynamic light scattering to measure
vs(r,t) and pressure p and occupies volume frac- to conserve volume. The final term represents D within swollen gels, macroscopic permeation
tion fs. The polymer phase, which occupies the drag force (per unit volume) exerted on the measurements to determine x, mechanical ex-
volume fraction fp, is treated as a homoge- network by the solvent, with friction coefficient periments to measure elastic moduli (K and G,
neous elastic network with displacement field x. The network exerts an equal and opposite and therefore M), and swelling experiments to
up(r,t) that describes local deformations from force on the fluid phase, which is treated as measure fs (28).
the equilibrium state. Together, the solvent and Darcy flow driven by a hydrodynamic pres- The second key relation connects the me-
matrix phases fill space—meaning fs + fp = 1. sure p. chanical response properties (moduli) of the
Even though the physical solvent and matrix Two important relations emerge from the matrix to the equilibrium thermodynamics of
materials are themselves assumed to be in- FS model that play a key role in the PF–SD swollen membranes (see the SM). Figure 3A
compressible, the matrix network may itself be bridge. The first relates the mutual diffusivity shows a gel initially swollen to thickness L in
compressed or dilated, because solvent flows D to the Darcy friction coefficient x, and the pure solvent, which deswells when an osmo-
into or out of the network to satisfy the space- second relates the thermodynamic proper- lyte (e.g., a polymer excluded from the swollen
filling requirement. The volumetric flux of the ties of a swollen membrane to its mechanical gel) reduces the solvent activity as outside the
FS composite is given by the local volume- properties. membrane (Fig. 3B). The osmotic pressure
averaged velocity, v = fp v p + fs v s , where Solving the polymer momentum equation difference exerts a stress dp on the membrane,
up = @up/@t is the local polymer velocity field (33). (Eq. 5) for vs, inserting into Eq. 4, and relat-
compressing the matrix until it is balanced by

The governing equations of the FS model ing ∇·up to changes in fp gives a diffusion-like the elastic stress −M∇·u. Solving this stress
equation for compressive deformations of the balance gives the longitudinal osmotic modu-
∇ v ¼ ∇ fp up þ 1 fp v s ¼ 0 ð4Þ swollen gel lus M

@fp Mfs 2 RT f0p @lnas
¼ ∇ fp ð7Þ M¼ ð9Þ

G

@t x V s f0s @lnfs T ;f0s
∇ sp ¼ Kþ ∇ð∇ up Þ þ G∇2 up
3
¼ xðv s up Þ

ð5Þ where M = K + 4G/3 is the longitudinal in terms of the thermodynamic relation be-
osmotic modulus, associated with free-draining tween solvent swelling (fs) and solvent ac-
uniaxial compression. Concentration fluctua- tivity (as).
tions of polymer and solvent diffuse relative to Even without explicitly solving the FS equa-
∇ ss ¼ ∇p ¼ xðv s up Þ ð6Þ

one another with mutual diffusivity tions (Eqs. 4 to 6), Eqs. 8 and 9 reveal three
enforce conservation of volume, polymer mo- Mfs distinct expressions for the permeability
mentum, and solvent momentum, respectively. D¼ ≡ MkPF ð8Þ
x !
Polymer and fluid inertia are neglected in both D fs V s @lnfs D
Eq. 5 and Eq. 6, owing to the small length and which is called the collective diffusivity by the kSD ¼ ¼
gel dynamics community (27–29, 35). The 1 fs RT @lnas T ;f0s M
long time scales of membrane permeation.
relationship between the collective diffusivity, f
The first two terms in the polymer momen- ¼ s ¼ kPF ð10Þ
tum balance (Eq. 5) represent the elastic forces which follows from deterministic arguments, x
exerted by a deformed, isotropic elastic solid and the stochastic solvent-polymer mutual
with shear and compressive moduli G and K, diffusivity, is a consequence of the fluctuation- to be quantitatively identical and to all follow
respectively (34). In the gel context, K is called dissipation theorem, analogous to the Stokes- from the FS model. The equality between the
the osmotic modulus, rather than the bulk Einstein relation. Indeed, Paul and Ebra‐Lima first two relates the thermodynamic factor in
modulus, because it is the swollen network suggested a relation of this sort should exist (5). the general solution-diffusive flux (Eq. 3) to
the compressional stiffness of the matrix using
Eq. 9. The second equality relates the binary
A B diffusivity to the mechanical membrane prop-
erties M and x (Eq. 8) to reproduce the PF
permeability (Eq. 1). Although these perme-
ability expressions appear very different, they
simply reflect different ways of expressing the
same, fundamental driving force—in purely
thermodynamic terms (Eq. 3), in purely me-
chanical terms (Eq. 1), or in hybrid terms (Eq. 10).
Figure 4 shows FS model predictions (solu-
tions to Eqs. 4 to 6) for a swollen membrane
of equilibrium thickness L, through which a
transmembrane pressure drop Dp drives a sol-
vent flux Js that is gentle enough to maintain a
linear pressure-flux relationship. The polymer
displacement up is highest [LDp/(2M)] on the
upstream side and vanishes at the rigid support
L δL L – δL
on the permeate side. The corresponding sol-
vent concentration
profile fs decreases linearly
Fig. 3. Hydrogel deswelling. (A) A gel initially swollen to thickness L in pure solvent (B) de-swells by dL by 1 f0s Dp=M over the membrane thick-
when an osmolyte lowers the osmotic pressure by dp. ness L, consistent with SD. Equal and opposite

gradients in fp, not shown, are required by Eq. 8 to eliminate M and x reproduces the both hydraulic permeance Pw ~ 1.6 liters per
space filling. The fluid-phase pressure p de- general SD flux (Eq. 2). Finally, combining m2·hour·bar and reduced Young’s moduli
creases continuously within the membrane, the fluid and solid momentum equations (Er ~ 1.6 GPa) of water-swollen PA membranes
with constant gradient −Dp/L, while the com- relates ∇fp directly to ∇p, without explicit of thickness L ~ 330 ± 75 nm (38). Using these
pressive normal stress s xxp on the polymer reference to J s. measurements as characteristic of PARO films,
phase grows linearly from zero on the (un- the FS model predicts diffusivities D = MPwL ~
compressed) feed side, to a maximum at the Consequences 2 × 10−9 to 4 × 10−9 m2/s. While the relatively
permeate side, with gradient equal and op- In addition to bridging the two distinct models, large experimental uncertainties may be re-
posite to the fluid pressure. the FS model brings qualitative and quantita- sponsible for the aphysically high end of this
tive insight to experimental results that affect range, the diffusivities measured by QENS
Resolving discrepancies the research workflows in next-generation has the magnitude expected for a membrane
The normal stress profiles in Fig. 4 clarify the membranes. Recently, for example, the struc- this stiff, with quantitative values consistent
different pressure profiles predicted by PF and tural and dynamic properties of dry and swol- with FS predictions. Moreover, Shin et al. con-
SD—the key distinction ultimately reflects dif- len PARO membranes were studied using ditioned PARO membranes with a series of
ferences in how the two models define pres- neutron scattering, with the explicit goal of solvents, finding higher permeance Pw with
sure. PF focuses exclusively on the mechanics identifying whether SD or PF correctly treated softer moduli, as expected from Eq. 8, owing
of the solvent phase and uses p to refer to the water permeation (17). Small-angle neutron to the higher volume fraction fs and larger
hydrodynamic pressure within the solvent scattering (SANS) found structural length scales pore sizes (lower x) (38).
phase. SD, by contrast, combines solvent and to not change appreciably when swelling, and The FS model highlights the fundamentally
polymer as one composite phase and uses p quasi-elastic neutron scattering (QENS) sug- interrelated nature of the mechanical (hydro-
to mean the composite stress sT = ss + sp of gested water diffusivity D ~ 2 × 10−9 m2/s, dynamic) and thermodynamic (diffusive) mech-
both fluid and solid components. Given that comparable to its self-diffusivity in bulk. Both anisms of solvent transport in osmotically
no other forces are exerted within the mem- results seemed inconsistent with SD, which active membranes. Collective diffusivity in
brane, mechanical equilibrium requires sT to requires coupled solvent/matrix diffusion driven PARO membranes is high because the swol-
be constant—indeed, adding the polymer and by concentration gradients. Moreover, recent len membrane is very stiff, owing to its high
solvent momentum equations (Eqs. 5 and advances in microscopy revealed nanoscale cross-link diffusivity, low swelling ratio, and
6) gives density variations in PARO membranes (36), therefore large thermodynamic factor. High
which could violate the homogeneity assumed cross-link densities give small mesh spacing
@sT @ss @sp
¼ þ ¼ xvs xvs ¼ 0 ð11Þ by SD. These results motivated an argument (akin to pore size Rp), increasing the friction
@x @x @x
that future PARO membranes should be de- factor x and decreasing permeability k.
Equation 11 holds irrespective of the specific veloped to reduce tortuosity—as appropriate Any case in which the FS connection fails to
form of the constitutive relations ss and sp or for PF—rather than working to enhance se- be made directly signals a breakdown in the
the solvent-polymer friction x. Treating the lective solubility. universality of the mechanism governing sol-
solvent and polymer phases separately, as in Reassessing these results using the FS model, vent transport in the membrane. For example,
FS, gives a nonzero solvent-phase pressure however, reveals no inconsistency. Namely, dense polymer membranes for gas separation
gradient that is balanced by the elastic stress Eq. 8 connects the collective diffusivity D to do not swell appreciably with added gas; rather,
of the compressed network. hydraulic permeability k through the longi- gas molecules dissolve into the free volume of
The FS profiles in Fig. 4 resolve the ap- tudinal osmotic modulus M0, which can be the polymer network. Moreover, the pressure-
parent discrepancies between PF and SD pre- measured independently, for example, using driven flow of bulk gasses looks nothing like the
dictions of concentration and pressure fields. a swelling pressure apparatus (22) or nano- molecular hopping of dissolved gas molecules.
In particular, nonzero gradients arise for both indentation (37, 38). Shin et al. measured The elastic moduli of such membranes reflect
concentration fs and fluid pressure p. Con-
centration gradients are required by SD but
are assumed not to exist in PF. Likewise, the Fig. 4. Steady-state FS profiles
pressure gradient is required by PF but as- across a solvent-swollen non- Feed Membrane Permeate
sumed to be zero in SD. In this regard, SD and porous membrane bound to a per-
PF are both correct in treating their respective meable support at the permeate Δp L
Up(x)
gradient but incorrect in assuming the other boundary. Polymer network
M 2
gradient to be zero. Notably, the PF model displacement, purple (SM, eq. S73);
does not require fs to be constant; it simply normal solvent stress, red (eq. S63); xx
σT (x)
does not rely on a nonzero ∇fs. Similarly, no normal polymer stress, green
pressure gradient is needed to drive a SD flux; (eq. S75); normal composite stress, pf
instead, mechanical equilibrium requires a pink (Eq. 11); and solvent volume Js p(x) Δp
constant normal stress. fraction, blue (eq. S76). A constant
Three small quantities arise in the linear- solvent flux Js compresses the poly- σpxx(x)
pp
response limit of solvent permeation of a swol- mer network. Compression of the
len gel: the solvent flux Js, the solvent pressure polymer network is dictated by the
gradient ∇p, and the solvent (or polymer) con-
φ0s Δp
imposed pressure drop Dp and
(1 – φ0s) M
centration gradient ∇fs. Each quantity is pro- the modulus of uniaxial compression
portional to one of the others while neglecting M. The sum of the solvent and
φs(x)
the third. Solving the solvent equation alone polymer stress, that is, the normal
relates Js to ∇p, while ignoring ∇fs, giving the composite stress, is constant x = –L x=0
PF flux (Eq. 1). Enforcing polymer momentum throughout the membrane, with a
alone relates Js to ∇fs and ignores ∇p; using discontinuity at the permeate boundary.

the compressibility of the polymer itself, with flux develops a nonlinear dependence on Dp, and toughness. Combining these two networks
negligible influence of the gas phase. Likewise, and material properties (e.g., M, x, fp, fs, and provides orthogonal control over permeabil-
the moduli of rigid, phase-separated, porous D) deviate substantially from their equilib- ity and mechanical stiffness and thus greater
membranes (e.g., zeolites, macroporous poly- rium values. A systematic perturbation solution control over the limiting flux.
mers, or track-etched membranes) are not set to the unidirectional permeation problem, de- The FS model is primed to determine the
by balancing the elastic stresses of the network scribed in the SM, reveals the general first dynamics of solutes and solvents in hydrogel
against the mixing and solvating stresses of the correction to the flux relation to be systems placed under mechanical stresses
solvent but instead reflect the stiffness of its or pressures—for example, hydrogels injected
DDp
elements and the morphology of the network. M0 L
into the human body for drug delivery (45, 46),
In both cases, the osmotic (free-draining) moduli Js ¼ h i ð12Þ tissue scaffolds (47), and chemomechanic
f0
1 þ 2M
Dp
f0p ðlnx0 Þ′ þ fp0 1
of the networks are little affected by the presence 0 s sensors (48, 49). The FS description of these
(or absence) of solvent, breaking the inter- bioengineered hydrogels connects system
relation between hydraulic and diffusive trans- where ðlnx0 Þ′ ¼ dlnx=dfs jf0s < 0 captures the performance with material chemistry. Al-
port captured by the FS model. Whether the decrease in friction with solvent swelling. The though we have specifically focused on binary
nanoscale heterogeneities observed in PARO second term in brackets reflects changes in solvent-polymer systems here, it is natural to
membranes (36) violate the homogeneity as- flow velocity due to increased (compressed) expand the FS formulation to account for ad-
sumed by SD and FS depends on whether they polymer fraction fp, and the third reflects ditional solute or solvent concentration fields,
are osmotically active when hydrated and can the decreased thickness of the compressed which occur both in membranes and more
therefore be treated as a swellable elastic membrane. generally.
network that is statistically homogeneous The FS limiting flux relation (Eq. 12) justifies More broadly, the FS model lays out a frame-
on length scales that are larger than the the empirical expression aDp/(1 + bDp) found work for the design of membrane materials on
heterogeneities but smaller than membrane by Paul and Ebra-Lima to fit nonlinear flux the basis of ex
situ
characterizations. The degree
dimensions. measurements (5). Additionally, Eq. 12 predicts of swelling f0s and mechanical moduli (G, K,
Another process involves high-pressure mem- the physical form of the fitting parameters and M) may be measured macroscopically for a
brane performance under extreme pressures, as a and b and thus provides design guidance swollen gel of interest, and the collective dif-
found in organic solvent nanofiltration (OSN) for future membranes. A key feature is that fusivity may be measured using dynamic light
(4, 5) and high-pressure reverse osmosis, for the longitudinal osmotic modulus M0 sets scattering, following Tanaka et al. (28), to give
example, as is used for brine treatment (39–41). the pressure scale above which the permeation all the needed parameters for FS description.
In these and other cases, membrane permeance flux starts to saturate. Although M0 was not Moreover, performing such experiments while
decreases at high pressures, which has been measured alongside the flux measurements, varying the osmotic pressure of the swelling
hypothesized to result from membrane compac- FS predicts the limiting (ceiling) flux at large solution could elucidate the dependence of
tion (4, 39, 41). pressure M, x, and fs on osmotic or transmembrane
Our analysis thus far has focused on lower- 1 pressure, providing a complete set of parame-
pressure permeation, where flux grows linearly 2D f0p ters for nonlinear flux relations such as Eq. 12.
Jsceiling ∼ f0p ðlnx0 Þ′ þ 0 1 ð13Þ
with applied pressure. In this linear-response L fs Finally, measuring these parameters as func-
limit, the physical properties of the membrane tions of the composition of the swelling solu-
change very little and are well approximated to be independent of M0, which can be compared tion (solvent mixtures, solute concentrations)
by their equilibrium values. Because the local to the measured ratio a/b. Paul and Ebra-Lima would provide the parameters needed to
friction coefficient x, longitudinal osmotic mod- tabulated a, b, D, L, fp, and fs for 12 organic predict multicomponent transport through
ulus M, and solvent velocity vs = js/fs all solvents in cross-linked rubber membranes (5); membranes.
depend on fs, concentration gradients can af- only (lnx0)′ was not determined. Basing x on Here, we have shown that the pore-flow
fect the flux expressions beyond their linear- Zick and Homsy’s calculation of fluid flow and solution-diffusion approaches both provide
response forms (Eq. 10). through periodic arrays of rigid spheres (19) correct and consistent descriptions of solvent
The FS model provides a natural framework reveals Eq. 13 to predict measurements to transport through nonporous, swollen mem-
to understand, analyze, and predict nonlinear within a factor of four for all 12 solvents (SM). branes, despite the apparent discrepancies in
pressure-flux relations that are expected to Furthermore, the fluid-solid model lays out their predictions of pressure and concentration
arise in high-pressure reverse osmosis brine a path to treat membranes for which neither profiles. These two approaches represent com-
treatment (39–41) and OSN (4, 5). Namely, the PF nor the SD approach is appropriate. For plementary descriptions of the same process
the permeation flux stops increasing linearly example, hydrogel pore-filled composite mem- and ultimately yield identical predictions. Nei-
at sufficiently high pressures, asymptotically branes (42) immobilize a homogeneous, non- ther approach is incorrect; instead, the predic-
approaching a limiting flux, as identified ex- porous polymer matrix within the pores of a tions of one model are in perfect quantitative
perimentally by Paul and Ebra-Lima (5) and in rigid membrane support. The rigid pore walls agreement with those of the other. After all,
the OSN context (4). Hints of a limiting flux constrain the deformations of the nonporous momentum and mass must both be conserved,
already appear in the leading-order FS solu- gel in a way that would be difficult to capture mechanical and thermodynamic balances must
tions (Fig. 4), because up →L once
Dp ~ M. The with SD, yet straightforward with FS. More both be respected. A subtle distinction between
weak compression limit dfp ≪f0p is therefore generally, the FS approach provides a natural the two approaches lies in the phases they treat:
only valid when Dp is much smaller than the framework to treat mechanically heteroge- Pore flow focuses exclusively on the liquid
longitudinal osmotic modulus M, which emerges neous membranes with rigid structural ele- phase, whereas solution diffusion treats the
as a natural scale for the transmembrane pres- ments that pin or restrict the deformation of polymer and solvent as a single, inseparable
sure. When Dp ≪ M, polymer and solvent con- nonporous networks. Double network hydro- continuum phase. The fluid-solid model is
centrations are perturbed only slightly, and gels (43, 44), composed of an expanded rigid general enough to enable transport modeling
the solvent flux varies linearly with Dp. When polymer mesh interpenetrated by an elastic for membrane materials that neither of the
Dp approaches or exceeds M, however, the swollen polymer network, have both high classical models can handle. Given the central
network compresses more substantially, the water content and high mechanical strength role of the longitudinal osmotic modulus M0

in the fluid-solid model, it is imperative that 21. J. S. Ham, M. C. Bolen, J. K. Hughes, J. Polym. Sci. 57, 25–40 (1962). 47. J. M. Cloyd et al., Eur. Spine J. 16, 1892–1898 (2007).
it be measured as part of the standard suite of 22. S. D. Gehman, Rubber Chem. Technol. 40, 532–543 (1967). 48. H. Ding et al., Carbohydr. Polym. 248, 116797 (2020).
23. M. A. Biot, J. Appl. Phys. 12, 155–164 (1941). 49. G. Lin, S. Chang, C. H. Kuo, J. Magda, F. Solzbacher, Sens.
characterization experiments for swollen mem- 24. M. A. Biot, J. Appl. Mech. 23, 91–96 (1956). Actuators B Chem. 136, 186–195 (2009).
branes (50). 25. M. A. Biot, D. G. Willis, J. Appl. Mech. 24, 594–601 (1957). 50. J. M. Dennison, X. Xie, C. J. Murphy, D. G. Cahill, ACS Appl.
26. M. Tokita, T. Tanaka, J. Chem. Phys. 95, 4613–4619 (1991). Nano Mater. 1, 5008–5018 (2018).
RE FE RENCES AND N OT ES 27. T. Tanaka, D. J. Fillmore, J. Chem. Phys. 70, 1214–1218 (1979).
1. G. M. Geise et al., J. Polym. Sci. B Polym. Phys. 48, 1685–1718 (2010). 28. T. Tanaka, L. O. Hocker, G. B. Benedek, J. Chem. Phys. 59, AC KNOWL ED GME NTS
2. G. M. Geise, D. R. Paul, B. D. Freeman, Prog. Polym. Sci. 39, 5151–5159 (1973).
Funding: This work was supported as part of the Center for
1–42 (2014). 29. M. Doi, J. Phys. Soc. Jpn. 78, 052001 (2009).
Materials for Water and Energy Systems (M-WET), an Energy
3. J. G. Wijmans, R. W. Baker, J. Membr. Sci. 107, 1–21 (1995). 30. C. Y. Hui, V. Muralidharan, J. Chem. Phys. 123, 154905 (2005).
Frontier Research Center funded by the US Department of Energy,
4. H. Y. Jang et al., AIChE J. 65, e16757 (2019). 31. J. Yoon, S. C. Cai, Z. Suo, R. C. Hayward, Soft Matter 6,
Office of Science, Basic Energy Sciences under award DE-
5. D. R. Paul, O. M. Ebra‐Lima, J. Appl. Polym. Sci. 14, 2201–2224 6004–6012 (2010).
SC0019272. Author contributions: V.H.H.: Conceptualization,
(1970). 32. D. Chen, J. Yoon, D. Chandra, A. J. Crosby, R. C. Hayward,
Formal analysis, Investigation, Methodology, Validation,
6. L. F. Greenlee, D. F. Lawler, B. D. Freeman, B. Marrot, P. Moulin, J. Polym. Sci. B Polym. Phys. 52, 1441–1461 (2014).
Writing – original draft, and Writing – review & editing. M.F.D.:
Water Res. 43, 2317–2348 (2009). 33. R. B. Bird, W. E. Stewart, E. N. Lightfoot, Transport Phenomena,
Conceptualization, Writing – review & editing, Supervision, Project
7. D. R. Paul, J. Polym. Sci. Polym. Phys. Ed. 11, 289–296 (1973). Revised 2nd Edition (John Wiley & Sons, 2006).
administration, and Funding acquisition. T.M.S.: Conceptualization,
8. D. R. Paul, J. Polym. Sci. Polym. Phys. Ed. 12, 1221–1230 (1974). 34. L. D. Landau, E. M. Lifshitz, Theory of Elasticity, vol. 7 of Course
Formal analysis, Investigation, Methodology, Validation, Writing –
9. D. R. Paul, Separ. Purif. Methods 5, 33–50 (1976). of Theoretical Physics, J. B. Sykes, W. H. Reid, Trans. (Addison-
original draft, Writing – review & editing, Supervision, Project
10. M. L. White, J. Phys. Chem. 64, 1563–1565 (1960). Wesley, 1959).
administration, and Funding acquisition. Competing interests:
11. H. Yasuda, C. E. Lamaze, A. Peterlin, J. Polym. Sci. A-2 35. T. Tanaka, S. Ishiwata, C. Ishimoto, Phys. Rev. Lett. 38,
None declared. Data and materials availability: All data needed
Polym. Phys. 9, 1117–1131 (1971). 771–774 (1977).
to evaluate the conclusions in the paper are present in the paper or
12. A. Peterlin, H. Yasuda, J. Polym. Sci. Polym. Phys. Ed. 12, 36. T. E. Culp et al., Science 371, 72–75 (2021).
the supplementary materials. License information: Copyright ©
1215–1220 (1974). 37. Y. Hu, X. Zhao, J. J. Vlassak, Z. Suo, Appl. Phys. Lett. 96,
2022 the authors, some rights reserved; exclusive licensee
13. M. Qasim, M. Badrelzaman, N. N. Darwish, N. A. Darwish, 121904 (2010).
American Association for the Advancement of Science. No claim to
N. Hilal, Desalination 459, 59–104 (2019). 38. M. G. Shin et al., J. Membr. Sci. 578, 220–229 (2019).
original US government works. https://www.science.org/about/
14. P. M. Biesheuvel, S. Porada, M. Elimelech, J. E. Dykstra, 39. M. T. M. Pendergast, J. M. Nygaard, A. K. Ghosh, E. M. Hoek,
science-licenses-journal-article-reuse
J. Membr. Sci. 647, 120221 (2022). Desalination 261, 255–263 (2010).
15. L. Song, M. Heiranian, M. Elimelech, Desalination 520, 115360 40. D. M. Davenport et al., J. Membr. Sci. 610, 118268 (2020).
(2021). 41. D. M. Davenport, A. Deshmukh, J. R. Werber, M. Elimelech, SUPPLEMENTARY MATERIALS
16. W. K. Kim, S. Milster, R. Roa, M. Kanduč, J. Dzubiella, Environ. Sci. Technol. Lett. 5, 467–475 (2018). science.org/doi/10.1126/science.abm7192
Macromolecules 10.1021/acs.macromol.2c00605 (2022). 42. N. Adrus, M. Ulbricht, J. Mater. Chem. 22, 3088–3098 (2012). Supplementary Text
17. E. P. Chan et al., Macromolecules 53, 1443–1450 (2020). Table S1
43. J. P. Gong, Soft Matter 6, 2583–2590 (2010).
18. E. M. Furst, T. M. Squires, Microrheology (Oxford Univ. Press, Nomenclature
2018). 44. S. Ahmed, T. Nakajima, T. Kurokawa, M. Anamul Haque,
J. P. Gong, Polymer 55, 914–923 (2014). References
19. A. A. Zick, G. M. Homsy, J. Fluid Mech. 115, 13–26 (1982).
20. S. Rosenbaum, O. Cotton, J. Polym. Sci. A-1 Polym. Chem. 7, 45. T. R. Hoare, D. S. Kohane, Polymer 49, 1993–2007 (2008). Submitted 6 October 2021; accepted 17 May 2022
101–109 (1969). 46. J. Li, D. J. Mooney, Nat. Rev. Mater. 1, 16071 (2016). 10.1126/science.abm7192

RES EARCH
GENDER-BASED VIOLENCE To fill these gaps, we employ what is to our

knowledge the largest randomized controlled
Policing in patriarchy: An experimental evaluation trial of a police reform to date, implemented
across 180 stations serving a population of
of reforms to improve police responsiveness 23.4 million. The trial evaluates the impact of
introducing Women’s Help Desks (WHDs):
to women in India dedicated spaces mandated to respond to
women’s cases, located within regular (mixed
Sandip Sukhtankar1,2*, Gabriele Kruks-Wisner3, Akshay Mangla4 gender) police stations, staffed by officers who
are trained in assisting women, and supported
Gender-targeted police reforms are frequently proposed to tackle the global problem of rising yet through routine monitoring, coupled with com-
under-reported gender-based violence (GBV)—but with mixed and often disappointing results. We munity outreach. We find that officers in sta-
explore this issue in India, a country with alarming rates of GBV and limited police capacity, by tions with WHDs are more likely to register cases
studying the impact of Women’s Help Desks (WHDs): dedicated spaces for women in local police of GBV and other complaints filed by women,
stations, staffed by trained officers. Drawing on the largest randomized controlled trial of a police particularly where female officers are (randomly)
reform to date (180 police stations serving 23.4 million people), we find that officers in stations with assigned to run the help desks. This suggests
WHDs are more likely to register cases of GBV, particularly where female officers run the desks. that, even in resource-constrained and patri-
This suggests that even in resource-constrained and patriarchal environments, police responsiveness archal environments, police responsiveness can
can be improved by focusing and mainstreaming attention to women’s cases and by greater gender be improved by mainstreaming and focusing
representation within the police. attention on women’s cases and by greater
gender representation within the police. We
C
explore the conditions under which these
ountries worldwide grapple with the dual like their male counterparts, often blame victims changes occurred, as well as the limits of the
problems of rising gender-based violence or dismiss their claims (12, 13). reform—which influenced officer behaviors
(GBV) and weak law enforcement. The A second set of reforms involve gender- and interactions with complainants but had
ability to register crimes with the police segregated services such as all-women police little observable effect on the overall rates at
is one essential step in accessing the jus- stations, established in countries as diverse as which women turn to the police.
tice system and promoting women’s security. Brazil, Ghana, India, Kosovo, and the Philippines.
Yet social barriers, including stigma, can deter These women-only stations, like other gender- Reforming police under patriarchy
women from reporting crimes, as can low segregated spaces such as women’s train car- India, which was recently labeled the “world’s
trust in the police (1, 2). The police, moreover, riages or girls’ schools (14, 15), operate on the most dangerous country for women” owing to
are often unresponsive to women’s concerns, theory that women may be more comfortable high rates of sexual violence (19), is an in-
leading to both under-reporting and under- in the absence of men. The assumption is that structive place to study gender-targeted police
recording of crime and violence (3). The result female victims of crime will be more likely to reforms. India ranks 140 of 156 countries on
is a pronounced gap between the incidence report cases in women-only police stations, the Global Gender Gap Index—an international
of crime and the rates at which such crime is and that female officers in those stations will measure of gender inequality. This inequality
formally recognized. be less constrained by patriarchal policing is visible in India’s skewed sex ratio at birth
Gender-targeted police reforms, which seek cultures (16). However, recent research sug- (929 girls/1000 boys), low female labor force
to increase the accessibility and accountabil- gests that separating women’s cases from other participation rate (22.3%) (20), and high rates
ity of the police to women, are regularly pro- police work makes it less likely that officers in of GBV. An estimated 4 in 10 Indian women
posed to tackle these problems. One set of regular, mixed-gender police stations will file report experiencing domestic violence in their
reforms have worked to increase women’s cases related to GBV, creating barriers to access lifetimes (21).
representation within police forces. These that further marginalize women (11, 13, 17). GBV is a deeply rooted problem reflecting
initiatives build on theories of representa- In this article, we examine a large-scale a nexus of social, economic, and political fac-
tive bureaucracy, which hold that the pres- gender-targeted police intervention that seeks tors, and state-led interventions—including
ence of members of marginalized groups within to mainstream rather than segregate attention policing—are only one slice of the necessary
a public agency improves performance with to women’s cases within routine police work, responses. Policing is a particularly crucial
respect to those groups (4, 5). Evidence on while also testing whether increased represen- but fraught arena in which to consider efforts
the efficacy of these reforms, however, re- tation of female officers better meets wom- to address GBV. The police are the primary
mains mixed. Some studies find that reporting en’s needs. Our setting is India, a country institution for public security, as well as the
and arrests for sexual assault increase with plagued by some of the world’s highest rates of most visible arm of the state’s coercive ap-
the presence of female officers (6, 7). Other violence against women. Home to entrenched paratus. The police are charged with protecting
research suggests that gender differentials patriarchal norms and chronically weak state women at the same time that law enforce-
in police behavior are diminished as female capacity (18), India is both a critical case and ment itself can play a role in perpetrating
officers operate within male-dominated policing a representative one, given the prevalence of violence—including violence against women
cultures (8, 9), noting a lack of increased gender such conditions in settings throughout the (22). The police are also critical gatekeepers
sensitivity among female officers (10, 11) who, Global South. Theoretical predictions about to the broader justice system, yet where em-
whether an intervention of this kind ought bedded in patriarchal norms can serve to deter
to have an impact are ambiguous, given the or block women’s legal claims. We focus on
1
Department of Economics, University of Virginia, Charlottesville, mixed evidence of existing research, in addi- police registration of cases of GBV as a pre-
VA 22904, USA. 2Abdul Lateef Jameel Poverty Action Lab tion to the difficulty of implementing such liminary step in a long chain to justice, in-
(J-PAL), Cambridge, MA, USA. 3Department of Politics, reforms. Moreover, efforts to mainstream cluding case investigation, arrests, judicial
University of Virginia, Charlottesville, VA 22904, USA. 4Saïd
Business School, University of Oxford, Oxford OX1 1HP, UK. rather than segregate attention to women’s action, and referrals to social services. Although
*Corresponding author. Email: srs8yk@virginia.edu cases have not been rigorously studied. case registration is just one step, it is a critical
Sukhtankar et al., Science 377, 191–198 (2022) 8 July 2022 1 of 7

one: If police cases are not filed, further legal of a bundle of gender-targeted reforms de- four largest cities and selecting the remainder
action cannot occur. signed to make the police both more accessible to be representative of geography, demograph-
In India, registration takes the form of a and accountable to female complainants (35). ics, and socioeconomic conditions across the
First Information Report (FIR). Lodged at a Our study setting is Madhya Pradesh, a large state (39). We excluded specialized police sta-
police station, the FIR records information (population 81 million) and ethnically diverse tions (state or district headquarters, cyber cells,
about a “cognizable” offense (classified in the state in north-central India. Madhya Pradesh or all-women stations), as well as outposts
Indian Penal Code as offenses of a serious is representative, in socioeconomic indicators (smaller than a station). MPP also excluded
nature for which a police officer has authority and gender norms, of much of northern India. stations that were remote, servicing entirely
to make an arrest without a warrant and to It is also an illustrative setting in which to rural populations, because they did not con-
start investigation without permission from explore the problem of under-reported GBV. A sider traffic to those stations high enough to
a court) (23). The FIR initiates case investiga- report from the state’s four largest cities, for justify a continuously staffed desk. The result-
tion and criminal proceedings, as stipulated example, found that only 1% of women who ing sample of 180 police stations should thus
under the Indian Penal Code. Domestic vio- had experienced violence had reported it to be viewed as being representative of primarily
lence can also be registered in a Domestic the police (36). (although not exclusively) urban and large
Incident Report (DIR), a complaint mechanism India’s federal architecture assigns state police stations. Within each district, we strati-
created under the Domestic Violence Act of governments the responsibility for policing, fied these 180 stations by geography (those that
2005. Filing a DIR initiates civil proceedings as well as social programming for women’s were fully urban, and those that also included
and referrals to social services, allows for pro- health and security. The WHD intervention rural areas) and by the first principal com-
tection orders and economic support, and may was designed by the Madhya Pradesh Police ponent of a vector of police station size char-
also lead to criminal proceedings. Unlike FIRs, (MPP) Research and Training Department, in acteristics (35). Within each of these strata,
which are filed at police stations, the DIR is consultation with our research team, lawyers, we then randomly assigned the stations into
filed with the local magistrate (judge). In the and GBV experts from civil society, with the three groups: 61 to the first treatment arm
latter instance, the police can serve as de- goal of overcoming barriers to both the report- (regular WHDs), 59 to the second (woman-run
signated Protection Officers who compile the ing of crimes by women and the recording of WHDs), and 60 to the control arm, which
DIR and bring it before the magistrate. More such cases by the police. received no help desk intervention.
details on FIRs and DIRs are in Appendix S1. Undertaking research on GBV and policing is We consider May 2019 as the start date of the
Women face major obstacles to case reg- a fraught endeavor, which requires careful full intervention. Training began as early as July
istration, reflecting barriers to both demand ethical consideration. We recognize that our 2018, although state elections in December 2018
(that inhibit the reporting of crime) and sup- roles as researchers are not and never can be meant that training did not fully ramp up
ply (that inhibit officers from recording cases) fully neutral and that there is the potential, until March 2019. Our field teams monitored
(24–26). Even when a woman overcomes so- through research, to introduce risk, activate implementation in December 2019, by which
cial and familial pressures to report a case, trauma, and create harm. We discuss our point all treatment stations had at least some
officers often resist officially recording it— efforts to minimize such risks in Appendix S2. training, 94% had set aside requisite space,
despite their legal obligation to do so (27). We are guided by recommendations devel- 87% had conducted a community outreach
The police’s hesitancy to file cases is driven oped specifically for research on GBV (37, 38), event, and 90% of women-run WHDs had a
in part by acute resource and capacity con- as well as by the insights of our local partners, female officer assigned. However, only 67% had
straints, which push officers to lessen their which enabled us to ground global best practices full training at the station level, and community
caseloads (28), as well as by political pres- in our study context. outreach was limited overall given the huge
sure to show lower official crime rates (29). The WHD intervention consists of four com- jurisdictions these stations served. Appendix S4
Patriarchal policing norms also push against ponents: (i) private spaces (such as a room or details implementation efforts and timelines.
case registration: Officers are encouraged cubicle) for female complaints within police
to “protect families” by promoting reconcil- stations; (ii) standard operating procedures Data sources
iation rather than the legal rights of women on how to register cases and assist women Our aim was to evaluate whether the estab-
(27, 30–32) and often blame victims of sex- visitors, along with officer training on those lishment of the WHDs, as well as assignment
ual assault or question the validity of their procedures and routine implementation moni- of female personnel to the WHDs, improved
claims. A recent report, for example, found toring; (iii) outreach to local women’s and com- the responsiveness of police officers to women
that 39% of officers believe that complaints munity safety networks; and (iv) assignment (40). To that end, we gathered data from five
of GBV are unfounded (33). The same report of female officers (at the rank of Assistant Sub- main sources.
also highlighted patriarchy within India’s Inspector or higher) to run the WHDs. 1) Administrative data on crimes registered
police stations. Women make up just 7% of Police stations—the unit of randomization— by the police at study police stations from
the force nationally and face heavy work bur- assigned to the first treatment arm (“regular” May 2018 through March 2020. Our data are
dens, as well as workplace discrimination. Fe- WHDs) received the first three components, aggregated with no individual or identifia-
male officers, operating in these highly masculine whereas those assigned to the second arm ble case details, but FIR categories indicate
settings, may feel pressure to act as “one of (woman-run WHDs) received all four com- (i) the number of “crimes against women”
the boys,” replicating and expressing patri- ponents. WHDs in the second treatment arm (CAW) cases, which include officially desig-
archal norms (34). These factors push against were directed to have designated female of- nated cases of GBV such as sexual assault, rape,
case registration: Indeed, there were fewer than ficers (with 90% compliance). Although there dowry, and other cases (details in Appendix S1);
four GBV-related FIRs registered per police was no prohibition on assigning female officers and (ii) whether the case was filed by a woman,
station (serving 130,000 people on average) to WHDs in the first treatment arm, most (72%) the latter also incorporating nonviolent crimes.
per month at baseline in our data. were run by male officers. The control group 2) CCTV data from the video feeds of
continued with business as usual. cameras, present in all police stations, focused
Intervention and randomization For the study, the MPP purposively selected on the station entrance. MPP provided us with
With these barriers to case registration in mind, 12 districts (out of 51) across the state, auto- a week’s worth of data from each study station,
we conducted an experiment to test the impact matically including those home to the state’s for the hours of 10 a.m. to 10 p.m. each day, at

Table 1. Primary outcomes from administrative data. Each observation represents data at the police station–month level. DIRs are Domestic Incident
Reports representing civil complaints of domestic violence (columns 1 and 5). FIRs are First Information Reports, either in cases of Crimes Against
Women (CAW) filed by anyone (columns 2 and 6) or in all criminal cases filed by women (columns 3 and 7). Arrests correspond to arrests in CAW cases
in a given month (columns 4 and 8). Regular WHD refers to Women's Help Desks without an assigned female officer; woman-run WHDs include an
assigned female officer. Strata FE refers to fixed effects for district × urban/rural strata. Standard errors clustered by police station are in parentheses.
*p < 0.10, **p < 0.05, ***p < 0.01.
(1) (2) (3) (4) (5) (6) (7) (8)

FIRs filed FIRs filed FIRs filed FIRs filed
DIRs filed in CAW cases by women Arrests DIRs filed in CAW cases by women Arrests
Treatment 1.452*** 0.542* 0.269 0.265
............................................................................................................................................................................................................................................................................................................................................
(0.180) (0.308) (0.168) (0.433)
............................................................................................................................................................................................................................................................................................................................................
Regular WHD 1.492*** 0.095 0.073 0.126
............................................................................................................................................................................................................................................................................................................................................
(0.250) (0.339) (0.188) (0.457)
............................................................................................................................................................................................................................................................................................................................................
Woman-run WHD 1.410*** 1.014** 0.476** 0.412
............................................................................................................................................................................................................................................................................................................................................
(0.205) (0.400) (0.207) (0.506)
............................................................................................................................................................................................................................................................................................................................................
Baseline control 0.192 0.409*** 0.344*** 0.210*** 0.191 0.408*** 0.344*** 0.209***
............................................................................................................................................................................................................................................................................................................................................
(0.268) (0.041) (0.036) (0.041) (0.265) (0.040) (0.035) (0.040)
............................................................................................................................................................................................................................................................................................................................................
Observed 1980 1980 1980 1980 1980 1980 1980 1980

............................................................................................................................................................................................................................................................................................................................................
Adjusted R2 0.179 0.270 0.254 0.164 0.179 0.279 0.259 0.164
............................................................................................................................................................................................................................................................................................................................................
Control mean 0.047 3.847 2.577 3.433 0.047 3.847 2.577 3.433
............................................................................................................................................................................................................................................................................................................................................
Strata FE Yes Yes Yes Yes Yes Yes Yes Yes
............................................................................................................................................................................................................................................................................................................................................
p value: T1 = T2 0.772 0.030 0.058 0.498
........................................................................................................................................................................................................................................................................................................................................
both baseline (n = 12,537, January to March

2019) and endline (n = 9757, February to
March 2020). This footage enables us to
measure the number of men and women
entering or exiting the police station.
3) A user survey of members of the public
who had visited study police stations, asking
about their satisfaction with their visit. This was
conducted at endline only (n = 3251, February to
March 2020), drawing from a random selection
of visitors within the span of 1 week.
4) A police survey, carried out at baseline
(n =1950, September to October 2018) and
endline (n = 1961, February to March 2020)
of personnel in different roles and ranks in
study stations. The survey—a representative
individual panel with equivalent rank replace-
ments for any transferred officers—captured
police perceptions and attitudes on crimes against
women.
5) A survey of citizens, carried out at baseline Fig. 1. Domestic incident registration increased sharply. Source: Raw official administrative data.
(n = 5648 women, 871 men, November 2018
and March to April 2019) and endline (n = Statistical methods ities to make the outcomes representative at
3376, July to December 2020, phone survey). We report Intent to Treat estimates compar- the police station level. Where available, we
The survey—an individual panel sampled from ing average outcomes in treated stations (with include the baseline value of the outcome as
all adult residents living in study police sta- either regular or woman-run WHDs) to out- a control. We cluster standard errors at the
tion jurisdictions—asked about perceptions comes in the control stations, as well as out- police station, the level of randomization. Our
of safety, opinions of and contact with the comes between the two treatment groups, using preferred specifications are
police, and experiences of crimes. the most disaggregated measure available. All (1) Y_ips = alpha + beta*treatment_ps +
Concurrent to this data collection, we also regressions include district-geographic stratum gamma*Y0_ips + delta_s + pc_ps + epsilon_ips
carried out qualitative research—observation fixed effects (the level at which treatment (2) Y_ips = alpha + beta1*treat1_ps +
and interviews with officers of various ranks— probabilities are equal), with a control for the beta2*treat2_ps + gamma*Y0_ips + delta_s +
in eight police stations, selected to represent first principal component of variables used for pc_ps + epsilon_ipswhere i is the individual,
each arm of the study in two purposively further stratification. Sampled observations are p is the police station, s is the stratum, Y is
selected districts (41). weighted using inverse sampling probabil- endline and Y0 is the baseline outcome. Using

baseline outcomes, we find no more imbal- increase in DIRs after the official launch of efficient is positive but standard errors are large
ances across treatment groups than would the program in May 2019, although some cases [95% confidence interval (CI) −0.58, 1.11] (43).
be expected by chance (tables S1 to S3) (42). were registered soon after training started Separating the analysis by treatment group,
in treatment stations as early as July 2018. we see that officers in stations with both reg-
Results Second, we also observe sizable increases in ular and woman-run WHDs filed increased
The WHD intervention led to increased reg- the number of FIRs filed in CAW cases (14.1%, numbers of DIRs, with the coefficients statis-
istration of women’s cases (Table 1). First, we p = 0.08), as well as FIRs filed by women tically indistinguishable from each other. How-
observe an increase in the number of DIRs (10.4%, p = 0.1). These results reflect an addi- ever, the increase in FIRs is entirely driven
filed, from practically zero in the control group tional 1905 DIRs and 3360 FIRs registered in by the woman-run WHDs (Fig. 2). In regular
(where officers had little to no knowledge of treatment stations over the 11 months of the WHDs, the treatment effects are small and
the DIR) to 1.5 monthly cases in the treatment intervention. There were no significant changes statistically indistinguishable from zero. In the
group (p < 0.01). Figure 1 shows the sharp in the number of arrests in CAW cases; the co- stations with woman-run WHDs, by contrast,
FIRs in CAW cases increased by 26.4%, whereas
FIRs filed by women rose by 18.5%; these treat-
ment effects are statistically distinguishable
from those in regular WHDs (p = 0.03, 0.06;
Table 1). Meanwhile, implementation quality
and officer training both have a strong and
significant impact on the number of cases
registered for all outcomes (table S4). Finally,
increases in registration of CAW cases do not
come at the cost of reductions in other cases,
with no discernible spillover effect on the over-
all number of FIRs or on other kinds of police
reports (table S5).
These results are driven by changes in police
behavior, reflected in a greater likelihood of
registering a case once a woman has reported
it, but not by any observable changes in the
rates at which women report cases or approach
the police. There is no impact on the overall
CAW rates reported by women in our citizen
survey (table S17). Neither is there an observ-
Fig. 2. Registration of crimes against women increased in women-run WHDs. Source: Raw official able impact on the rates at which women vis-
administrative data. ited police stations, based on analysis of CCTV
Table 2. Primary outcomes from CCTV data. Observations are at time window–day–police station level, as described in Appendix S1. The number of female
visitors is the per-hour count of all women who entered a police station at any time between 10 a.m. and 10 p.m. (columns 1 and 3). The proportion of female visitors is the
number of female visitors divided by the number of all visitors captured by CCTV camera in the same hourly durations (columns 2 and 4). Regular WHD refers to
Women's Help Desks without an assigned female officer; woman-run WHDs include an assigned female officer. We use district FE instead of strata FE owing to the
unavailability of data from 41 police stations, which results in an unbalanced distribution of treatment and control police stations across some strata. Though not shown in
the table, the regressions control for the number of female officers in each police station at endline; fixed effects for time window of day and day of week; the average
number of frames per second in the video; and the average number of pixels in the video.
(1) (2) (3) (4)

No. of female visitors Percent female visitors No. of female visitors Percent female visitors
Treatment −0.590 0.005
............................................................................................................................................................................................................................................................................................................................................
(1.656) (0.011)
............................................................................................................................................................................................................................................................................................................................................
Regular WHD −0.811 0.009
............................................................................................................................................................................................................................................................................................................................................
(1.894) (0.011)
............................................................................................................................................................................................................................................................................................................................................
Woman-run WHD −0.402 0.001
............................................................................................................................................................................................................................................................................................................................................
(2.097) (0.012)
............................................................................................................................................................................................................................................................................................................................................
Baseline control 0.226*** 0.359*** 0.227*** 0.359***
............................................................................................................................................................................................................................................................................................................................................
(0.052) (0.062) (0.052) (0.061)
............................................................................................................................................................................................................................................................................................................................................
Observed 1832 1831 1832 1831

............................................................................................................................................................................................................................................................................................................................................
Adjusted R2 0.162 0.168 0.162 0.169
............................................................................................................................................................................................................................................................................................................................................
Control mean 13.706 0.092 13.706 0.092
............................................................................................................................................................................................................................................................................................................................................
District FE Yes Yes Yes Yes
............................................................................................................................................................................................................................................................................................................................................
p value: T1 = T2 0.857 0.403
............................................................................................................................................................................................................................................................................................................................................

with better implementation and more training

(table S8). In addition, both female and male
visitors to treatment stations were signifi-
cantly more likely to profess satisfaction with
the physical infrastructure and conditions of
the station, compared to visitors to control
stations. Overall, these results suggest modest
improvements in citizen perceptions of the
police given the presence of a WHD. Meanwhile,
deep-seated police attitudes about gender
did not appear to shift overall; there was no
significant change in the rates at which of-
ficers reported the belief that women file so-
called “false cases” against men (an indicator
of whether they are inclined to believe women
or dismiss their claims) (Fig. 4) (45). Notably,
however, female officers in both treatment
arms were less likely than those in control
Fig. 3. Women visiting treatment stations were slightly more satisfied. Source: User Survey; stations to ascribe to the narrative of false cases,
coefficients and notes in table S7. though there was no such effect on male of-
ficers; this suggests that police attitudes shifted
more for women. There is some evidence of a
shift in officers’ awareness of the general
inadequacies of policing with regard to women’s
cases (Fig. 4). Both male and female officers
in treatment stations were significantly less
likely to describe the police as helpful to
victims of crimes against women and were
also less likely to believe that the police pay
sufficient attention to women’s cases relative
to other law and order issues (this last result is
borderline significant; p = 0.12, table S13). We
do not interpret these findings as evidence
of a shift in gender attitudes, but rather of
increased cognizance of the gap between how
women’s cases should be handled and how
they are handled in practice—a likely conse-
quence of WHD training. Both male and female
officers in treatment stations were also more
likely to state (unprompted) that cases of crimes
Fig. 4. Police attitudes and awareness. Source: Police survey; coefficients and notes in tables S13 and S14. against women are among the top priorities
within their stations—a likely effect of the
attention-focusing presence of the WHDs.
data (Table 2), although these latter results minor changes in the satisfaction of female
must be viewed with caution given missing visitors to police stations, expressed in exit Discussion
data issues (44). These non-impacts are not a interviews (Fig. 3). In control stations, re- Our findings suggest that even in resource-
result of these variables being uninformative spondents generally expressed high levels of constrained and patriarchal environments, ef-
or unpredictive of outcomes in general: Survey satisfaction with their visits (3.16 on a 4-point forts that focus attention on women’s cases can
CAW numbers are predictive of FIRs registered, scale), agreed they were treated respectfully have a meaningful impact on police behavior,
baseline CCTV counts are predictive of endline (3.31), and that they felt comfortable discus- making officers more responsive to women’s
CCTV counts, and the number of female staff sing their concerns (3.3). The high levels of security concerns. This is visible in the higher
(from administrative data) is predictive of the reported satisfaction may reflect desirability registration of both FIRs and DIRs (respectively,
number of female visitors in the CCTV data. bias among respondents hesitant to critique criminal and civil complaints). The rise in DIRs
We do not observe any broad changes in citizen the police at the station, and given top-coded across both treatment arms is pronounced,
behavior, although it is possible that there values, there may be limited scope to observe reflecting the adoption of a relatively new
were changes in complainant behavior within any improvement due to the WHD interven- practice (since 2005) that remains largely
the police stations, as women decide whether tion. It follows that there are few statistically unknown to the police without WHD training,
or not to pursue cases once at the station and distinguishable differences in satisfaction be- and so is all but absent at baseline and in
in interaction with officers. Our research de- tween treatment and control stations; only the control stations. Police officers are required to
sign and ethical parameters precluded us from “comfort” variable is significantly improved assist women in creating the report and must
observing any such interactions, but we posit given the presence of a WHD (p = 0.07, table also ensure that the form is lodged with a local
that changes in complainant behavior are likely S7). Notably, however, all three measures of magistrate and that social services (such as
supported by changes in police behavior that we user satisfaction were significantly higher, com- shelter homes) are accessible—activities that
do observe. Among complainants, we do find paring among treatment stations, in those extend beyond conventional police work. Through

WHD training, officers gained knowledge of Second, the WHD intervention worked to police, moreover, did not shift, although this
the DIR and learned to coordinate with other mainstream gender-responsive policing practices. may reflect the short duration of the inter-
state and civil society agencies. By locating the help desks in regular police vention; beliefs about the police may shift
The increase in FIRs for women’s cases is stations, and by training both male and female over time if the intervention becomes insti-
also notable, as is the fact that it is driven almost officers on the mission and operation of the tutionalized and more visible to citizens. This,
entirely by woman-run WHDs. This is not sim- desks, the WHDs helped to give visibility and however, will depend not only on the ease with
ply a function of increased personnel assigned ascribe value to work on women’s cases, rather which women are able to register cases, but
to the woman-run WHDs: Both woman-run than casting such work as peripheral to and also on the functioning of the broader crim-
WHDs and regular WHDs received additional therefore of lesser importance than other crime inal justice system. Last, the WHD interven-
high-ranked officers to operate the help desk prevention tasks. Further study is required tion did little to address deep-seated social
(table S9). It appears, then, that the presence to examine the effects of this mainstreaming, and economic structures that both drive vio-
of additional female officers is critical for over- which—in creating an enabling environment lence against women and inhibit women’s
coming barriers to FIR registration. This gender- in which to focus on women’s cases—may help access to justice. Efforts to address GBV nec-
differentiated effect for FIRs but not DIRs, us understand why female officers were partic- essarily require broader, multipronged ap-
we suggest, reflects the higher costs to officers ularly receptive to WHD training and protocols. proaches that extend beyond police reforms.
of filing an FIR. Unlike the DIR, the FIR auto- These observations carry important insights
matically initiates a criminal case, requiring for debates over representative bureaucracy Conclusion
substantial investments of police time for in- and the question of whether the presence of The urgent challenge of women’s security has
vestigation and in court proceedings. More- members of marginalized groups within a prompted police reform proposals worldwide,
over, to file an FIR, officers must push against public agency, such as the police, improves calling for increased female representation
strong norms within the police—articulated in performance (4, 5, 30). Our findings suggest among officers, gender-sensitization training,
our qualitative research—that prioritize “pro- that descriptive representation does matter; and specialized services for women. To date,
tecting families” by avoiding legal proceedings, female officers played a critical role in shaping however, the limited evidence available on the
in addition to dismissive narratives about “false the impact of the help desks. This, however, efficacy of these reforms has been mixed, if not
cases.” The DIR, though certainly not trivial, re- was not simply a matter of “feminizing” a po- pessimistic, especially within patriarchal set-
quires less from officers, in part because it ini- lice station by adding female officers. Far from it tings. Moreover, efforts aimed at mainstream-
tiates a civil rather than criminal case, and in the agency of female officers must be under- ing (rather than segregating) these reforms
part because it passes some of the burden to stood as part of the full WHD bundle, including have not been rigorously studied. Our exper-
other government agencies. It is also a relatively training, infrastructure, and higher-level sup- iment suggests that cautious optimism may
new practice that does not have the same norms ports that enabled those officers to work—both be warranted, providing evidence that gender-
pushing against it. Filing an FIR, in sum, requires as women and for women. Efforts to enhance targeted reforms can, under certain conditions,
higher levels of officer commitment, which we bureaucratic representation, this suggests, may improve police responsiveness to women.
see primarily within woman-run WHDs. hinge on institutional supports that activate the In addition to furthering our understanding
Questions remain about the sources of com- agency of underrepresented public personnel. In of police reform and representative bureauc-
mitment and the agency of female officers. the case of the WHDs, the mainstreamed nature racy, our results also carry policy implications
More research is required to explore the mech- of the intervention (housed in regular, mixed- as governments in India and elsewhere con-
anisms through which female officers affected gender police stations) may have played a cri- sider gender-targeted reforms. Indeed, informed
change through the WHDs, as well as on how tical role. Although we are not able to directly in part by these results, MPP has begun scaling
the WHD bundle affected female officers. How- compare the WHDs to all-women stations, up the WHD intervention to 700 police stations
ever, our qualitative research suggests two recent research suggests that the act of “seg- across the state. Studying this scale-up may help
avenues that, together, appear to empower regating” women’s cases may have the perverse us understand whether the observed changes in
female officers and increase their responsive- effect of marginalizing female complainants police behavior can be sustained, as well as
ness to women. First, the intervention worked (creating barriers and displacing cases) as well whether and how attitudinal change among
to build station-level capacity for action on as female officers (who are isolated and both police and citizens can take hold.
women’s cases, channeling resources to treat- sidelined from broader policing structures)
ment stations, while also introducing manuals (13, 30, 46). Locating women’s help desks in REFERENCES AND NOTES
to clarify complex legal procedures, alongside regular stations, by contrast, appears to have 1. G. Blair et al., Science 374, eabd3446 (2021).
intensive training. These investments in officer worked to increase attention to women’s 2. Human Rights Watch, “Everyone Blames Me”: Barriers to Justice
capacity may have been most acutely felt and security within “normal” police work. Future and Support Systems for Sexual Assault Survivors in India
(2017).
utilized by female officers, who tend to be more research is required to systematically explore 3. R. Manjoo, “Report of the Special Rapporteur on violence
overburdened—given their small numbers and such mainstreaming attempts, which may against women, its causes and consequences” (United Nations
the many tasks assigned to them—relative to prove critical to broader efforts to build and Office of the High Commissioner for Human Rights, 2012).
4. B. A. Ba, D. Knox, J. Mummolo, R. Rivera, Science 371,
male officers (33, 34). Female officers appear to support more representative bureaucracies. 696–702 (2021).
have been particularly responsive to WHD train- Our study also highlights the limitations of 5. L. Keiser, V. M. Wilkins, K. J. Meier, C. A. Holland, Am. Polit.
ing, which emphasized the legal requirement to police-centered reforms. First, gender attitudes Sci. Rev. 96, 553–564 (2002).
6. K. Meier, J. Nicholson-Crotty, Public Adm. Rev. 66, 850–860 (2006).
file FIRs in the case of cognizable offenses, and among the police are difficult to move (although 7. A. Miller, C. Segal, Rev. Econ. Stud. 86, 2220–2247 (2019).
which also contained gender sensitization mod- perhaps less so for female officers). Second, even 8. M. Morash, R. Haarr, Fem. Criminol. 7, 3–23 (2012).
ules that urged officers to listen to women’s as the WHD program increased the likelihood 9. S. Karim, Am. Polit. Sci. Rev. 114, 536–551 (2020).
10. S. Karim, M. J. Gilligan, R. Blair, K. Beardsley, Int. Stud. Q. 62,
claims and not dismiss them out of hand. The of the police recording crimes against women,
618–631 (2018).
impact of this training is visible in the gender- barriers to women reporting such crimes 11. N. Wagner, M. Rieger, A. Bedi, W. Hout, J. Afr. Econ. 26,
differentiated rates of filing FIRs (more likely remain. This may reflect limited community 492–515 (2017).
in woman-run stations) and shifts in beliefs outreach implementation, with only 10% of 12. C. Santos, Women’s Police Stations: Gender, Violence,
and Justice in São Paulo, Brazil (Palgrave Macmillan, 2005).
concerning “false cases” (significant for female women in our citizen survey aware of WHDs 13. N. Jassal, Am. Polit. Sci. Rev. 114, 1035–1054 (2020).
but not male officers). (table S20). Citizens’ attitudes toward the 14. C. Jackson, J. Public Econ. 96, 173–187 (2012).

15. V. Ceccato, A. Loukaitou-Sideris, Transit Crime and Sexual 38. World Health Organization, Putting Women First: Ethical and AC KNOWL ED GME NTS
Violence in Cities: International Evidence and Prevention Safety Recommendations for Research on Domestic Violence This study would not have been possible without the cooperation
(Routledge, 2020). Against Women (WHO, 2003). of the Madhya Pradesh Police department and its officials, in
16. E. Perova, S. A. Reynolds, Soc. Sci. Med. 174, 188–196 (2017). 39. Bhopal, Gwalior, Indore, and Jabalpur districts contain the particular V. Kapoor, A. Shankar, R. Kumar Shukla, S. Rana, and
17. S. Amaral, S. Bhalotra, N. Prakash, “Gender, crime and (eponymous) four largest cities in MP; the other districts are R. Tiwari; as well as the enormous efforts of the JPAL-SA research
punishment: Evidence from women police stations in India” Baitul, Balaghat, Morena, Panna, Ratlam, Rewa, Seoni, and Vidisha. team, in particular S. Bhattacharya, S. Suganya, S. Chandrasekhar,
Working paper, 2021. 40. The registration of crime must be distinguished from the M. Tikku, M. Walia, P. Kumar, P. Baruah, K. Ravishankar,
18. D. Kapur, J. Econ. Perspect. 34, 31–54 (2020). actual incidence of crime which, by nature, is not observable L. Diaz-Martin, S. Ramesh, and A. Bhargava; and the analysis team
19. Thomas Reuters Foundation, “The World’s Most Dangerous through official data given the many deterrents to both of J. Le, E. Robertson, and A. Kumar at UVA. We thank J. Doleac,
Countries for Women – 2018”; https://www.reuters.com/ reporting and recording cases (1). O. Dube, S. Karim, S. Lakhtakia, A. Miller, P. Niehaus, S. Ostermann,
article/us-women-dangerous-poll-exclusive/exclusive-india- 41. Interviews took place over repeated station visits during a N. Prakash, S. Sekhri, M. Williams, four anonymous referees,
most-dangerous-country-for-women-with-sexual-violence-rife- 10-month period and were carried out by two Research and participants in various seminars and conferences for
global-poll-idUSKBN1JM01X. Assistants (RAs) trained in qualitative methods. RAs visited comments. IRB: The study received IRB approval from the
20. The Global Gender Gap Report 2021 (World Economic Forum, each station approximately twice per month. Each visit lasted University of Virginia (#2548), the Institute for Financial
2021). for a full day, yielding an average of 21.25 day-long visits per Management and Research in India (#7107), and Oxford University
21. A. Kalokhe et al., Glob. Public Health 12, 498–513 (2017). station. RAs interviewed four to six officers of different ranks in (SSH_SBS_C1A_18_065). Registration: The trial is registered (with
22. M. Berry, M. Lake, To End Violence Against Women, We Need to each station (including station chiefs, assigned help desk Pre-Analysis Plans) in the American Economic Association’s
Imagine New Futures (IPI Global Observatory, 2021). officers, and female officers where present). Interviews lasted randomized control trial registry under ID AEARCTR-0003357.
23. Noncognizable offences, by contrast, require police officers to for 1 hour on average, and most officers were interviewed Funding: This study was funded primarily by the Crime and
obtain warrants before arrests and court orders before multiple times. The qualitative research design is described in Violence Initiative (CVI) of the Jameel Poverty Action Lab (JPAL),
initiating investigation. our Pre-Analysis Plan, Section 8 (35). as well as the World Bank’s Sexual Violence Research Initiative
24. P. Roychowdhury, Capable Women, Incapable States: Negotiating (SVRI), with supplemental funding provided by the University of
42. DIRs filed are significantly higher if data until April 2019
Violence and Rights in India (Oxford Univ. Press. 2020). Virginia’s Center for Global Inquiry and Innovation (CGII) and the
are considered, which is not surprising because many
25. A. Banerjee, R. Chattopadhyay, E. Duflo, D. Keniston, N. Singh, University of Oxford. Author contributions: Authors contributed
treatment stations had received training by that time. There is
Am. Econ. J. Econ. Policy 13, 36–66 (2021). equally to conceptualization, design, and writing. S.S. was
no difference in DIRs when including only data before
26. L. Iyer, A. Mani, P. Mishra, P. Topalova, Am. Econ. J. Appl. Econ. responsible for randomization and data analysis. Competing
July 2018 (prior to training). Excluding DIRs, out of 45
4, 165–193 (2012). interests: Authors declare that they have no competing interests.
coefficients only one is significant at the 5% and one at the
27. Commonwealth Human Rights Initiative, Barriers in Accessing Data and materials availability: A Pre-Analysis Plan was
10% level.
Justice (CHRI, 2020). registered with the American Economic Association’s RCT Registry
43. This lack of movement on arrests is unsurprising given
28. Nationally, there are just 1.2 officers per 1000 people in India. at: https://www.socialscienceregistry.org/trials/3357. All code,
the complex mix of factors that shape whether an arrest is
The United States, by comparison, has 2.3 police officers per and survey and administrative data used in the analysis are
appropriate (given the nature of a case), as well as the fact that
1000 people, according to the UN Office on Drugs and Crime. available (with PII and police station names extracted to preserve
some complainants, facing family pressure, may not want
29. On politicization of the police in India, see (47). anonymity as required by IRB) on the JPAL dataverse hosted by
perpetrators arrested. Lack of increased arrests, therefore,
30. N. Jassal, J. Asian Stud. 80, 631–661 (2021). Harvard University (48). The raw CCTV videos will not be available
should not be interpreted solely as an indicator of police inaction.
31. CSDS, Status of Policing in India Report 2018: A Study of given privacy issues, but the final data used for analysis are
Performance and Perceptions (Centre for the Study of 44. Because of data collection issues (COVID-related and other-
available. All data needed to evaluate the conclusions in the paper are
Development Societies, New Delhi, 2018). wise) with the CCTV data, nearly one-third (55) of our study
present in the paper or the Supplementary Materials. License
32. The practice of reconciliation builds on Indian legislation, including stations are not included in our endline analysis of these
information: Copyright © 2022 the authors, some rights reserved;
the 1984 Family Courts Act, and various Supreme Court directives, outcomes, so these results must be interpreted with caution.
exclusive licensee American Association for the Advancement of
which advise officers to provide “counseling” to preserve families. 45. The relatively tight standard error bounds visible in the figure
Science. No claim to original US government works. https://www.
33. Common Cause et al., Status of Policing in India 2019: Police suggest that these are precise null effects, which is
science.org/about/science-licenses-journal-article-reuse
Adequacy and Working Conditions (Common Cause and Lokniti, unsurprising given deep-rooted attitudes.
2019). 46. Our aim and empirical strategy did not include a direct
34. Commonwealth Human Rights Initiative, Rough Roads to comparison of the WHD intervention and all-women stations, SUPPLEMENTARY MATERIALS
Equality: Women Police in South Asia (CHRI, 2015). which are entirely different models of policing. Moreover, there science.org/doi/10.1126/science.abm7387
35. Our Pre-Analysis Plans were registered with the American were only six all-women police stations in our study districts in Appendix S1: Clarifications and further background
Economic Association’s RCT Registry at https://www. MP during our study period. Appendix S2: Ethics statement
socialscienceregistry.org/trials/3357. 47. S. K. Mitra, “Policing and strategic reform” in The puzzle of Appendix S3: Tables S1 to S24
36. S. Neville et al., “Evaluation of the Madhya Pradesh Safe Cities India's Governance: Culture, Context and Comparative Theory
Appendix S4: Implementation timeline and details, table S25, and
Initiative. Baseline Report” (DFID, London, 2015). (Routledge, 2007).
figs. S1 to S6
37. C. Bradbury-Jones et al., Best Practice Guidelines in Relation to 48. S. Sukhtankar, G. Kruks-Wisner, A. Mangla, ‘Policing in
References (49–65)
Gender-Based Violence Research in Low- and Middle-Income patriarchy: An experimental evaluation of reforms to improve
Countries (Univ. of Birmingham/Institute for Global Innovation, police responsiveness to women in India,’ Harvard Dataverse, Submitted 7 October 2021; accepted 4 May 2022
2020). Version 2 (2022). https://doi.org/10.7910/DVN/R75XVZ. 10.1126/science.abm7387

RES EARCH
NEUROSCIENCE difference relative to ambient noise (i.e., the

SNR) increased the nociceptive threshold, we
Sound induces analgesia through delivered 62-, 67-, 72-, or 77-dB SPL white noise
to CFA mice under conditions of 57-dB SPL
corticothalamic circuits ambient noise. The mechanical nociceptive
threshold was elevated only after exposure to
Wenjie Zhou1†, Chonghuan Ye1†, Haitao Wang2,3†, Yu Mao1,4†, Weijia Zhang1, An Liu5, 62-dB (5-dB SNR) SPL white noise (Fig. 1E).
Chen-Ling Yang5, Tianming Li6, Lauren Hayashi6, Wan Zhao7, Lin Chen2, Yuanyuan Liu6*‡, In addition, 35-dB (5-dB SNR) SPL white noise,
Wenjuan Tao5*‡, Zhi Zhang1*‡ but not 40-, 45-, or 50-dB SPL, produced sim-
ilar effects in an environment with 30-dB SPL
Sound—including music and noise—can relieve pain in humans, but the underlying neural mechanisms ambient noise (fig. S1C). We also found that
remain unknown. We discovered that analgesic effects of sound depended on a low (5-decibel) signal-to- 5-dB SNR sound elevated the thermal nocicep-
noise ratio (SNR) relative to ambient noise in mice. Viral tracing, microendoscopic calcium imaging, tive threshold in the Hargreaves test (Fig. 1F
and multitetrode recordings in freely moving mice showed that low-SNR sounds inhibited glutamatergic and fig. S1D). Furthermore, the effects of 5-dB
inputs from the auditory cortex (ACxGlu) to the thalamic posterior (PO) and ventral posterior (VP) SNR sound on pain hypersensitivity were ob-
nuclei. Optogenetic or chemogenetic inhibition of the ACxGlu→PO and ACxGlu→VP circuits mimicked served in mice with neuropathic pain induced
the low-SNR sound–induced analgesia in inflamed hindpaws and forepaws, respectively. Artificial by spared nerve injury (SNI) and tonic pain in-
activation of these two circuits abolished the sound-induced analgesia. Our study reveals the duced by capsaicin in both males and females
corticothalamic circuits underlying sound-promoted analgesia by deciphering the role of the auditory (figs. S2 to S4). On the basis of these findings,
system in pain processing. we refer to 5-dB SNR as low SNR; henceforth,
the ambient noise level used was 45-dB SPL, un-
A
less otherwise stated.
s early as 1960, there were accounts from associated regions to process sensory dis- We next conducted conditioned place aver-
dental operations showing that music crimination and affective-motivation of pain sion (CPA) and conditioned place preference
and noise can induce analgesic effects through diverse subregions (16–19). Notably, (CPP) tests to assay the potential effects of
(1). Music delivered as an intervention the activity of the somatosensory thalamus can sound on the affective component of pain.
can alleviate postoperative and procedu- be affected by music and innocuous salient au- Five-dB, but not 15-dB, SNR white noise abol-
ral pain (2–4) and even refractory pain in the ditory stimuli (15, 20). Thus, the thalamus might ished the subthreshold von Frey (0.04 g) stimuli-
clinic (5, 6). Because diverse genres of music function as a bridge for audio-somatosensory induced place aversion in the CPA test and
and even nature sounds can relieve pain to an processing. However, the precise cell type– evoked a preference for the sound-delivery
equal extent (7), the inherent characteristics specific organization and the function(s) of the side in the CPP test (Fig. 1, G to J, and fig. S5).
of music or contextual factors—that is, not only thalamic circuits mediating sound-induced an- Taken together, these findings support that
to music per se—have been hypothesized to algesia remain largely unknown. low-SNR sound induces analgesia.
drive these analgesic effects (8). However, it is Pain perception is affected by emotion and
still unknown how it works. Sound-induced analgesia depends on a low stress (22, 23). Behavioral tests, including open
Functional magnetic resonance imaging (fMRI) signal-to-noise ratio (SNR) relative to ambient field, light-dark box, and elevated plus maze,
studies have implicated changes in the activity noise in mice showed that neither 5-dB nor 15-dB SNR
of multiple brain areas mediating pain process- Inspired by findings in humans, we investi- white noise evoked anxiety-like behaviors in
ing in humans exposed to music (9–12). To date, gated whether consonant sound (e.g., pleasant mice under acute pain conditions (CFA 3 days
the neural substrates underpinning cross-modal music for humans) may elicit analgesic effects or SNI 7 days) (fig. S6, A to F), and neither
audio-somatosensory interactions remain unclear. in mice with inflammatory pain induced by sound reduced anxiety under chronic pain
The thalamus relays multimodal sensory infor- hindpaw injection of complete Freund’s adju- conditions (CFA 21 days or SNI 42 days) (fig. S6,
mation (including auditory and somatosensory) vant (CFA) (Fig. 1A and fig. S1A). Because loud G to L). Moreover, no stress effects were in-
(13–15) and reciprocally connects with pain- noises (above ~75 dB) induce escape behavior volved in the sound-induced analgesia: (i) The
in mice (21), we delivered consonant sound to serum corticosterone level was unaffected by
1 CFA mice at a sound pressure level (SPL) of 5-dB SNR sound exposure (fig. S7), and (ii)
Department of Anesthesiology and Pain Medicine, The First
Affiliated Hospital of USTC, Hefei National Laboratory for ~50 or ~60 dB through a closely positioned intrathecal application of the m opioid recep-
Physical Sciences at the Microscale, Division of Life Sciences audio speaker (ambient noise at 45-dB SPL). tor antagonist naloxone had no effect on 5-dB
and Medicine, University of Science and Technology of China, Using von Frey filaments to assess me- SNR sound–induced increase in nociceptive
Hefei, PR China. 2Auditory Research Laboratory, Department of
Neurobiology and Biophysics, Division of Life Sciences and chanical sensitivity, we found that delivery thresholds (fig. S8).
Medicine, University of Science and Technology of China, Hefei, of consonant sound at a SPL of ~50 dB, but
PR China. 3School of Integrated Chinese and Western Medicine, not of ~60 dB, elevated the inflamed hindpaw Glutamatergic neurons in the auditory cortex
Anhui University of Chinese Medicine, Hefei, PR China. (ACx) are functionally connected to the
4
Department of Anesthesiology and Pain Management, The
nociceptive threshold compared with CFA mice
First Affiliated Hospital of Anhui Medical University, Hefei, PR exposed only to ambient noise (Fig. 1B). Not- thalamic posterior (PO) and ventral posterior
China. 5Department of Physiology, School of Basic Medical ably, this effect lasted for at least 2 days after (VP) nuclei
Sciences, Anhui Medical University, Hefei, PR China.
6
Somatosensation and Pain Unit, National Institute of Dental
repeated exposure to ~50-dB SPL consonant We next investigated the neural circuit(s) un-
and Craniofacial Research (NIDCR), National Center for sound for 3 days (20 min per day). Notably, no derlying the observed low-SNR sound–induced
Complementary and Integrative Health (NCCIH), National difference was detected in the nociceptive analgesia. In the brain, sound signals are gated
Institutes of Health (NIH), Bethesda, MD, USA. 7Department of thresholds among CFA mice exposed to ~50-dB by the medial geniculate body (MGB) before
Otolaryngology, The First Affiliated Hospital of USTC, Division of
Life Sciences and Medicine, University of Science and SPL consonant sound, dissonant sound, or reaching the ACx, which serves as the conver-
Technology of China, Hefei, PR China. white noise (Fig. 1, C and D, and fig. S1B). gence point for detailed analysis of sound
*Corresponding author. Email: zhizhang@ustc.edu.cn (Z.Z.); Music-induced analgesia in humans may be (24, 25). Because both of these nuclei are pri-
wjtao01@ahmu.edu.cn (W.T.); yuanyuan.liu@nih.gov (Y.L.)
†These authors contributed equally to this work. attributable to the treatment environment (8). marily composed of glutamatergic neurons (26),
‡These authors contributed equally to this work. To test whether 50-dB SPL or the intensity we examined the presynaptic outputs from the
Zhou et al., Science 377, 198–204 (2022) 8 July 2022 1 of 7

Fig. 1. Low-intensity sound A B Ambient noise 50 dB SPL CS
Paw withdrawal threshold (g)

relative to ambient noise Saline CFA Sound 0.6 60 dB SPL CS
induces analgesia in mice. CS
CS
(A) Schematic for inducing
0.4
inflammatory pain and the von
Frey test to assess the mechanical
CS
nociceptive threshold. Sound 0.2
refers to delivered consonant
sound (CS), dissonant sound (DS),
0.0
or white noise (WN) given at the
von Frey test
10 BL
20 in
30 in
40 in
50 in
in
H
indicated SPLs. (B to D) The Saline or CFA
m
m
m
m
m
24
48
72
96
mechanical nociceptive threshold
C D
in CFA mice treated with or

0.6 Ambient noise 50 dB SPL DS 0.6 Ambient noise 50 dB SPL WN
without consonant sound (ambi- 60 dB SPL DS 60 dB SPL WN
ent noise, n = 10 mice; 50-dB SPL, WN
DS WN
n = 10 mice; 60-dB SPL, n = 8 0.4 DS 0.4
mice; P < 0.0001) (B), dissonant
sound (n = 10 mice each group; WN
0.2 DS 0.2
P < 0.0001) (C), and white noise
(ambient noise, n = 10 mice;
50-dB SPL, n = 10 mice; 60-dB 0.0 0.0
SPL, n = 8 mice; P < 0.0001) (D)
10 BL
20 in
30 in
40 in
50 in
in
H
10 BL
20 in
30 in
40 in
50 in
in
m
m
m
m
m
24
48
72
96
m
m
m
m
m
24
48
72
96
in an environment with an ambient
noise at 45-dB SPL. BL, baseline. E Ambient noise at 57 dB SPL F
Ambient noise 5-dB SNR WN
(E) The mechanical nociceptive 0.4 62 dB SPL WN 72 dB SPL WN 15
Latency to withdrawal (s)

threshold of CFA mice exposed 67 dB SPL WN 77 dB SPL WN WN 15-dB SNR WN
WN
to white noise at different 0.3
intensities in an environment 10
WN
with ambient noise at 57-dB SPL 0.2
n.s.
(62-dB SPL, n = 10 mice; n.s. n.s. 5
67-dB SPL, n = 10 mice; 0.1
72-dB SPL, n = 8 mice; 77-dB SPL,
n = 10 mice; P < 0.0001). (F) The 0.0 0
in
10 BL
20 in
30 in
40 in
50 in
H
Du e
Du e
Du e
Du e
g
thermal nociceptive threshold

Pr
Pr
Pr
Pr
rin
rin
rin
rin
m
m
m
m
m
24
48
72
96
assessed by the Hargreaves test in
CFA mice exposed to different G Sound H Ambient noise I J Ambient noise
SNR white noise (n = 10 mice Pre During 5-dB SNR WN Sound 5-dB SNR WN
each group; P < 0.0001). 3.0 15-dB SNR WN 2.5 15-dB SNR WN
Preference ratio (Post/Pre)

15 10 15 n.s.
Aversion for dark side
(G) Schematic for the CPA test. Time (min) n.s. 2.0
(H) Summarized data for the von
g
(During/Pre)
g
le
2.0
l le
Frey filament stimulus-induced 1.5

CF
ro
nt
Co
place aversion in CFA mice treated

1.0
with or without white noise 1.0 Paired side Unpaired side
l
tra
u
(ambient noise, n = 9 mice; 5-dB

Ne
0.5
SNR, n = 9 mice; 15-dB SNR, CPP
0.04 g von Frey 0 0
n = 11 mice; P = 0.0165).
(I) Schematic for the CPP test.
(J) Summarized data for preference ratio for the sound-delivery side in CFA mice from the indicated group (ambient noise, n = 11 mice; 5-dB SNR, n = 10 mice;
15-dB SNR, n = 8 mice; P = 0.0015). The data are expressed as the means ± SEMs. *P < 0.05; **P < 0.01; ***P < 0.001; n.s., not significant. Details of the
statistical analyses are presented in table S1.
ACxGlu and MGBGlu neurons into the brain regulate cortices, the primary and secondary neurons in any of these nuclei (fig. S10). Be-
gions tightly linked to pain processing (27–29). somatosensory cortices, the nucleus accumbens, cause the PO and VP receive prominent inputs
Specifically, we injected an adeno-associated the ventrolateral periaqueductal gray, the dorsal from the ACx and are known to relay noci-
viral vector (AAV)–DIO–membrane-bound green raphe nucleus, the rostroventral medulla, the ceptive information (30–32), we subsequently
fluorescent protein (mGFP)–Synaptophysin-mRuby parabrachial nucleus, and the basolateral and focused on the role of ACx→PO and ACx→VP
into these nuclei in Ca2+/calmodulin-dependent central amygdala, and no signals were de- circuits in sound-induced analgesia.
protein kinase II (CaMKII)–Cre mice (fig. S9, A tected in the spinal cord (fig. S9). We de- In vivo multitetrode recordings in freely
and B, and fig. S10, A and B). tected abundant mRuby signals in thalamic moving CFA mice showed that the spontane-
mRuby+ terminals originating from ACxGlu nuclei, including the PO and VP, but not in ous neuronal activity in the ACx was decreased
neurons were observed in the insular cortex the mediodorsal and central medial nuclei in response to white noise at 5-dB, but not 15-dB,
(ICx) but were very scarce in the medial pre- (fig. S9, M to P and R). Notably, we detected SNR (Fig. 2, A to C). Five-dB SNR white noise–
frontal cortex, the anterior and posterior cin- no mRuby signals originating from MGBGlu induced neuronal inhibition was mimicked on

the basis of (i) ACx infusion of Cre-dependent von Frey, Hargraves, CPA, and CPP tests. This neuronal activity was significantly increased
chemogenetic inhibitory hM4Di-mCherry vi- selective inhibition mimicked the effects from in CFA mice (fig. S17, I and J), and this ele-
ruses (AAV-DIO-hM4Di-mCherry) and (ii) intra- the 5-dB SNR white noise in both male and vation in neuronal activity was attenuated by
peritoneal injection with the hM4Di-agent female mice (Fig. 3, G to J; fig. S13J; and fig. exposure to 5-dB SNR white noise, but not
clozapine-N-oxide (CNO) in CaMKII-Cre mice. S14). By contrast, optical activation of the 15-dB SNR (Fig. 4, D and E, and fig. S17K).
The nociceptive thresholds were elevated upon ACxGlu→PO circuit abolished both neuronal In both male and female mice, bilateral
selective inactivation of ACxGlu neurons; condi- inhibition in the PO and analgesia induced optical inhibition of the eNpHR3.0-EYFP–
tioned aversion was abolished upon such in- by 5-dB SNR sound (fig. S15). containing ACxGlu terminals in the VP mimicked
hibition (Fig. 2, D and E). Conversely, bilateral To selectively monitor the response of PO- 5-dB SNR white noise–induced VP neuronal in-
optical activation of ACxGlu neurons expressing projecting ACx neurons to 5-dB SNR sound at hibition (Fig. 4, F and G) and analgesia (Fig. 4,
AAV-DIO-ChR2-mCherry abolished the 5-dB single-neuron resolution, we infused a retroAAV- H to J; fig. S17, L and M; and fig. S18). Optical
SNR white noise–induced analgesia in CaMKII- Cre virus into the PO and an AAV expressing activation of the ACxGlu→VP circuit abolished
Cre mice (fig. S11). Cre-dependent fluorescent Ca2+ indicator both elevation of nociceptive thresholds (fig.
We then characterized both ACx→PO and GCaMP6m (AAV-DIO-GCaMP6m) into the ipsi- S19, A to C) and neuronal inhibition in the VP
ACx→VP circuits in greater detail. We per- lateral ACx, accompanied with the mounting of induced by 5-dB SNR sound (fig. S19, D to F).
formed anterograde monosynaptic tracing by a microendoscopic gradient index (GRIN) lens Microendoscopic calcium imaging showed
ACx injection with AAV1-Cre–enhanced green at the top of the ACx (fig. S16A). The Ca2+ that the spontaneous Ca2+ transient frequency
fluorescent protein (EGFP) virus along with transient frequency of these ACx neurons was of VP-projecting ACx neurons was attenuated
ipsilateral PO and VP injection of Cre-dependent decreased after exposure to 5-dB, but not 15-dB, during exposure to 5-dB, but not 15-dB, SNR
DIO-EGFP, which revealed numerous EGFP+ SNR white noise in freely moving CFA mice white noise (fig. S20, A to D). In contrast to
neurons in the PO and VP (Fig. 2F). These PO (fig. S16, B and C). The activity of PO neurons that of POACx neurons, the Ca2+ transient fre-
and VP EGFP+ neurons colocalized with gluta- receiving ACx projections (POACx) was mea- quency of VP neurons receiving ACx projec-
matergic neurons but not with g-aminobutyric sured in mice with ACx infusion of AAV1-Cre tions (VPACx) was rapidly increased by punctate
acid–releasing (GABAergic) neurons (Fig. 2, G virus and ipsilateral PO infusion of AAV-DIO- mechanical stimulation of inflamed forepaws,
and H, and fig. S12, A to C). GCaMP6m (Fig. 3, K and L). The Ca2+ tran- but not hindpaws (movies S7 and S8). The
Retrograde tracing by PO injection with sient frequency of POACx neurons was rapidly spontaneous Ca2+ transient frequency of these
a retroAAV expressing tdTomato (retroAAV- increased by punctate mechanical stimuli neurons was enhanced in CFA mice compared
tdTomato) and VP injection with a retroAAV (movies S3 and S4). The spontaneous Ca 2+ with saline mice (fig. S20, E and F), which was
expressing EGFP (retroAAV-EGFP) showed transient frequency of these neurons was ele- attenuated by exposure to 5-dB, but not 15-dB,
that retrogradely labeled tdTomato+ neurons vated in CFA mice compared with saline mice SNR white noise (Fig. 4, K to M).
were abundant in ACx layer 6 (L6), whereas (fig. S16, D to F), and the increased frequency The 5-dB SNR sound–induced elevation of
EGFP+ neurons were abundant in ACx L5 was attenuated during exposure to 5-dB, but the forepaw nociceptive threshold was mimicked
(Fig. 2, I and J). Both the PO- and VP-projecting not 15-dB, SNR white noise (Fig. 3, M and N). upon chemogenetic inhibition of VPACx neu-
ACx neurons colocalized with glutamatergic The POACx neurons were then selectively rons and was blocked upon activation of these
rather than GABAergic neurons (Fig. 2, K and manipulated to experimentally validate their neurons (fig. S20, G to I). Notably, the hindpaw
L, and fig. S12, D to F). Whole-cell recordings participation in sound-induced analgesia. We nociceptive threshold was not affected by op-
combined with optogenetics in brain slices injected AAV1-Cre virus into the ACx while tical activation or inhibition of the ACxGlu→VP
showed that brief light stimulation of ChR2- concurrently injecting a viral vector express- circuit, and neither optical activation nor in-
containing ACxGlu terminals in the PO or VP ing either Cre-dependent hM3Dq-mCherry or hibition of the ACxGlu→PO circuit affected the
reliably elicited glutamate-mediated excitatory hM4Di-mCherry into the PO (fig. S16G). We forepaw nociceptive threshold (fig. S21). Com-
postsynaptic currents (EPSCs) (Fig. 2, M to P). found bilateral chemogenetic inhibition of parable results were obtained upon chemo-
POACx neurons recapitulated 5-dB SNR sound– genetic manipulations of VPACx or POACx neurons
Inhibition of the ACxGlu→PO circuit mediates induced analgesia (fig. S16H). By contrast, chemo- (fig. S22).
sound-induced hindpaw analgesia genetic activation of these neurons blocked the Given that the ICx, a higher-order cortex
We conducted in vivo multitetrode and fiber 5-dB SNR sound–induced analgesia (fig. S16I). influenced by music in humans (9–11), receives
photometry recordings in freely moving CFA ACx projections (fig. S9C), we examined the
mice. PO rather than VP neurons were rapidly Sound-induced forepaw analgesia mediated by potential role of the ACxGlu→ICx circuit in
activated by punctate mechanical stimulation inhibition of the ACxGlu→VP circuit sound-induced analgesia. Optical activation of
of inflamed hindpaws (Fig. 3, A to C; fig. S13, A The VP mediates somatosensation of the upper ACxGlu terminals in the ICx did not affect 5-dB
to D; and movies S1 and S2). PO neuronal ac- limbs (33). Given our findings of a functional SNR sound–induced elevation of nociceptive
tivity was significantly enhanced in CFA mice ACxGlu→VP connection, we further investi- thresholds of inflamed paws, whereas optical
compared with saline-treated controls (fig. S13, gated whether this circuit participates in 5-dB inhibition of this circuit in the absence of sound
E and F), which was attenuated by 5-dB, but SNR sound–induced analgesia. We observed a had no effects on pain hypersensitivity (fig. S23).
not 15-dB, SNR white noise (Fig. 3, D and E, significant increase in nociceptive thresholds Taken together, our results demonstrate that the
and fig. S13G). Similar sound-induced inhib- of the CFA-injected forepaw along with a observed analgesia from 5-dB SNR sound is me-
itory effects on PO neurons were observed reduction in place aversion upon 5-dB, but diated by the ACx→PO circuit for hindpaws and
after optical inhibition of the ACxGlu termi- not 15-dB, SNR white noise (fig. S17, A to E). by the ACx→VP circuit for forepaws (fig. S24).
nals in the PO of CaMKII-Cre mice treated Both in vivo multitetrode and fiber photome-
with ACx infusion of a Cre-dependent AAV try recordings in freely moving mice revealed Discussion
carrying eNpHR3.0–enhanced yellow fluores- that neuronal activity in the VP—but not the The neural circuit responsible for processing
cent protein (EYFP) (AAV-DIO–eNpHR3.0– PO—was rapidly enhanced by punctate me- sound-induced analgesia has long remained
EYFP) (Fig. 3F and fig. S13, H and I). chanical stimulation of inflamed forepaws elusive. Here, we describe the analgesic effects
We then performed bilateral optical inhibi- (Fig. 4, A to C; fig. S17, F to H; and movies S5 of sound at low SNR through inhibition of
tion of the ACxGlu→PO circuit and conducted and S6). Compared with saline mice, the VP audio-somatosensory corticothalamic circuits.

A Amplifier B C D
ACx in vivo recording 5-dB SNR ACxGlu inactivation
60 n.s.
15-dB SNR

5-dB SNR 15-dB SNR mCherry hM4Di-mCherry
Pre-noise
1.5 1.5
Firing Rate (Hz)

In vivo
0.1 mV

recording 40
1
(During/Pre)
1.0 1.0
1 100 ms 20
0.5 0.5
During
0 n.s.
Du e
Du re
g
g
Pr
rin
rin
0.0 0
P
BL CNO CPA
E F G H
mCherry Thalamus : EGFP PO : EGFP/Glu/DAPI VP : EGFP/Glu/DAPI
hM4Di-mCherry AAV1-Cre 100
Percentage of Glu+ (%)

15
LP 90
ACx
PO 80
10 n.s. VP
PO/VP 70
5 st
60
0 AAV-DIO-EGFP 50
BL CNO PO VP
I J K L
EGFP/tdTomato/DAPI ACx : POEGFP/Glu/DAPI ACx : VPEGFP/Glu/DAPI
rAAV2/2-tdTomato 100
Percentage of Glu+ (%)

L1
L4 L2/3 90
ACx L6 L5 80
PO
VP 70
60
rAAV2/2-EGFP 50
PO VP
M N ACSF X O ACSF F X P
DIO-ChR2-mCherry SF Q S Q ACx
DNQX AC DN DNQX AC DN L6
Light evoked response (pA)
Light evoked response (pA)
0 0
R ACx 0 mV 0 mV -100
-100 -70 mV
-70 mV
-200
PO -200 PO L5
VP -300
50 pA
50 pA 20 ms
-300 -400
20 ms
Glutamate
CaMKII-Cre mice -400 -500 VP
GluRs
Fig. 2. ACxGlu neurons project to VPGlu and POGlu neurons. (A) Schematic DAPI, 4′,6-diamidino-2-phenylindole. (I) Schematic for retrograde tracing.
for multitetrode recording in freely moving mice. (B and C) Raster plots (J) Representative images showing EGFP+ and tdTomato+ neurons in
and voltage traces of the spontaneous firings recorded in the ACx (B) the ACx. Scale bars, 100 mm. (K and L) Representative images of the
and summarized data (5-dB SNR, n = 25 cells from four mice; 15-dB SNR, colocalization of EGFP-labeled PO- and VP-projecting ACx neurons
n = 22 cells from four mice; P = 0.0053) (C). (D and E) Summarized with glutamate immunofluorescence (K) and summarized data
data for the mechanical nociceptive threshold (mCherry, n = 10 mice; (n = 4 slices) (L). Scale bars, 50 mm. (M) Schematic for viral injection
hM4Di-mCherry, n = 8 mice; BL, P = 0.3816; CNO, P < 0.0001) [(D), left], and whole-cell recordings. R, recording. (N and O) Representative traces
place aversion (n = 9 mice each group; P = 0.0006) [(D), right], and and summarized data for light-evoked postsynaptic currents recorded
thermal nociceptive threshold (n = 10 mice each group; P < 0.0001) (E) in CFA in PO neurons (n = 12 cells from four mice; P = 0.0002) (N) and VP
mice upon chemogenetic inhibition of ACxGlu neurons. (F) Schematic neurons (n = 14 cells from four mice; P < 0.0001) (O). ACSF, artificial
for anterograde tracing and representative image of EGFP-expressing cerebrospinal fluid; DNQX, 6,7-dinitroquinoxaline-2,3-dione. (P) A
neurons in the PO and VP. Scale bar, 500 mm. LP, lateral posterior thalamic model of ACxGlu→PO and ACxGlu→VP circuits. GluRs, glutamate
nucleus; st, stria terminalis. (G and H) Representative images showing the receptors. The data are expressed as the means ± SEMs. ***P < 0.001;
colocalization of EGFP-labeled neurons with glutamate (Glu) immuno- n.s., not significant. Details of the statistical analyses are presented
fluorescence (G) and summarized data (n = 4 slices) (H). Scale bars, 50 mm. in table S1.

A B VP in vivo recording C PO in vivo recording

BL 80 30 n.s. BL 80 40
20 30
In vivo
20
recording 60 10 60 10
FR BL (Hz)
FR BL (Hz)
1 0 0
40 40
0.1 mV
0.1 mV
1 100 ms
von Frey von Frey 100 ms
20 20
0 0
von Frey 0 20 40 60 80 0 20 40 60 80
FR von Frey (Hz) FR von Frey (Hz)
D PO in vivo recording
E F 60 30
60 Control
5-dB SNR 20
Control 5-dB SNR
FR Light On (Hz)
Pre-light
Firing Rate (Hz) 40 10
Pre-noise
40 n.s. 0
100 ms 20
100 ms 20
Light On
0.1 mV
During
0 0
Du e
g
Du e
g
0 20 40 60
Pr
Pr
rin
rin
FR Pre-light (Hz)
G H 15 eNpHR3.0-EYFP I 0.6 eNpHR3.0-EYFP J eNpHR3.0-EYFP
EYFP EYFP 2.0 EYFP

DIO-eNpHR3.0-EYFP

594 nm light
10 0.4 1.5
(During/Pre)
ACx n.s.
1.0
5 0.2
n.s.
PO 0.5
CaMKII-Cre mice 0 0.0

0
Lig e
ht
st
Lig e
Po t
st
Lig e
st
Lig e
Po t
st
ht
h
h
Pr
Pr
Pr
Pr
Po
Po
CPA
K L M N
Sound Scope DIO-GCaMP6m/DAPI Pre-noise During-noise 15 5-dB SNR
15-dB SNR
AAV1-Cre lens
1 min dF/F0
Events/min/cell
5-dB SNR
10 n.s.
LP
ACx GRIN lens
5
PO
15-dB SNR
PO VP
rt
AAV-DIO- 0
e
Du e
g
Pr
Pr
rin
rin
GCaMP6m
Du
Fig. 3. Low-SNR sound inhibits the ACxGlu→PO circuit to induce analgesia for optogenetic inhibition of the ACxGlu→PO circuit. (H to J) Summarized data
on hindpaws. (A) Schematic for multitetrode recording in freely moving for the thermal nociceptive threshold (EYFP, n = 10 mice; eNpHR3.0-EYFP, n = 9 mice;
mice with punctate mechanical stimulation (von Frey filament, 0.04 g). P < 0.0001) (H), mechanical nociceptive threshold (EYFP, n = 8 mice; eNpHR3.0-
(B and C) Raster plots, voltage traces, and summarized data for the EYFP, n = 9 mice; P < 0.0001) (I), and place aversion (n = 10 mice each group;
spontaneous firings recorded in VP neurons (n = 19 cells from four mice; P = 0.0001) (J) after optical inhibition of the ACxGlu→PO circuit in CFA mice.
P = 0.5079) (B) and in PO neurons (n = 27 cells from five mice; P = 0.0003) (K) Schematic for vial injection and microendoscopic calcium imaging. (L) A
(C) before and during punctate mechanical stimulation of CFA-injected typical image showing the GCaMP6m fluorescence and track of lens in
hindpaws. FR BL, firing rate baseline. (D and E) Raster plots and voltage the PO. Scale bar, 200 mm. rt, reticular thalamic nucleus. (M and N) Representative
traces of the spontaneous firings recorded in PO neurons in CFA mice traces of spontaneous Ca2+ signal transient recorded in PO neurons receiving
with or without 5-dB SNR white noise exposure (D) and summarized data ACx projections (M) and summarized data (5-dB SNR, n = 20 cells from four
(control, n = 24 cells from four mice; 5-dB SNR, n = 47 cells from eight mice; mice; 15-dB SNR, n = 15 cells from four mice; P < 0.0001) (N). dF/F0, the
P < 0.0001) (E). (F) Raster plots of the spontaneous firings recorded in PO change in fluorescence (dF) over the baseline fluorescence (F0) of calcium
neurons before and during optical inhibition of the ACxGlu→PO circuit (left) and spikes. The data are expressed as the means ± SEMs. **P < 0.01; ***P < 0.001;
summarized data (n = 71 cells from seven mice; P < 0.0001) (right). (G) Schematic n.s., not significant. Details of the statistical analyses are presented in table S1.

A In vivo B PO in vivo recording C VP in vivo recording D VP in vivo recording

recording
BL 80 20 BL 150 100 Control 5-dB SNR
Pre-noise
n.s.
60 10 50
FR BL (Hz)
100
FR BL (Hz)
0 0
0.1 mV
0.1 mV
1
40
0.1 mV
100 ms 100 ms
1 von Frey100 ms von Frey 50
20
During
0 0
0 20 40 60 80 0 50 100 150
von Frey FR von Frey (Hz) FR von Frey (Hz)
E F G H I

Control 100 60 10 eNpHR3.0-EYFP 0.4 eNpHR3.0-EYFP

120 5-dB SNR EYFP
40 EYFP
80 8
FR Light On (Hz)
Pre-light
n.s. 0.3
Firing Rate (Hz)
20
80 60 6 n.s.
0
0.2
100 ms 40 4
40 n.s.
0.1
Light On
20 2
0 0 0 0.0
Du e
g
Du re
g
0 20 40 60 80 100
Lig e
Po t
st
Lig e
Po t
st
Lig e
st
Lig e
Po t
st
Po t
Pr
h
rin
rin
Pr
Pr
h
h
P
Pr
Pr
FR Pre-light (Hz)
J EYFP
K DIO-GCaMP6m/DAPI
L Pre-noise During-noise
M
2.0 15 5-dB SNR 15-dB SNR
eNpHR3.0-EYFP GRIN LP
5-dB SNR
Events/min/cell
lens n.s.
1.5
(During/Pre)
1 min 10
rt PO
1.0
VP 5
15-dB SNR
dF/F0
0.5
0
0
g
Pr
Pr
rin
rin
CPA
Du
Du
Fig. 4. Inhibition of the ACxGlu→VP circuit mediates low-SNR sound– data (n = 67 cells from seven mice; t66 = 12.14; P < 0.0001) (G). (H to J)
induced analgesia on forepaws. (A) Schematic for multitetrode recording in Summarized data for the thermal (EYFP, n = 10 mice; eNpHR3.0-EYFP, n = 9
the VP or PO of freely moving mice. (B and C) Raster plots, voltage traces, and mice; F2,34 = 20.98; P < 0.0001) (H) and mechanical (EYFP, n = 10 mice;
summarized data for the spontaneous firings recorded in PO neurons (n = 36 eNpHR3.0-EYFP, n = 9 mice; F2,34 = 13.25; P < 0.0001) (I) nociceptive thresholds
cells from four mice; t35 = 1.749; P = 0.089) (B) and VP neurons (n = 18 cells of CFA-injected forepaws and place aversion (EYFP, n = 10 mice; eNpHR3.0-
from four mice; t17 = 7.373; P < 0.0001) (C) before and during punctate EYFP, n = 9 mice; t17 = 5.648; P < 0.0001) (J) upon optical inhibition of the
mechanical stimulation (von Frey filament, 0.02 g) of inflamed forepaws. ACxGlu→VP circuit. (K) A typical image of GCaMP6m fluorescence and track of
(D and E) Raster plots and voltage traces of the spontaneous firings recorded in the lens in the VP. Scale bar, 200 mm. (L and M) Representative traces (L) of
VP neurons from CFA mice with or without 5-dB SNR white noise exposure spontaneous Ca2+ signals recorded in VP neurons receiving ACx projections and
(D) and summarized data (control, n = 21 cells from four mice; 5-dB SNR, n = 23 summarized data (5-dB SNR, n = 35 cells from four mice; 15-dB SNR, n = 36
cells from four mice; F1,42 = 24.18; P < 0.0001) (E). (F and G) Raster plots cells from four mice; F1,69 = 24.24; P < 0.0001) (M). The data are expressed as
of the spontaneous activity recorded in VP neurons before and during optical the means ± SEMs. *P < 0.05; ***P < 0.001; n.s., not significant. Details of
inhibition of the ACxGlu→VP circuit in CFA-treated mice (F) and summarized the statistical analyses are presented in table S1.
Specifically, we reveal that the distinct roles given that the analgesic effects persisted for cated in the effect of sound on pain processing
of the ACxGlu→PO and ACxGlu→VP circuits in at least 2 days after sound withdrawal. and could expedite the study of music-induced
sound-induced analgesia depend on the phys- The neural mechanisms underlying music- analgesia. In the future, these findings could
ical location of the pain. induced analgesia in humans are doubtlessly spur the development of alternative interven-
In mice, we found that sound-induced anal- more complicated than those revealed in mice tions for treating pain.
gesia depended on its low SNR rather than (34). In humans, multiple areas that are in-
harmony, which is supported by a previous volved in pain processing, including the ICx,
hypothesis that music-induced analgesia is striatum, and midbrain dopamine system, do REFERENCES AND NOTES
attributable to contextual factors of the treat- respond to music (35–38). Therefore, whether 1. W. J. Gardner, J. C. Licklider, A. Z. Weisz, Science 132, 32–33
(1960).
ment, not only to the music per se (8). It bears other brain regions receiving ACx projections
2. A. Keenan, J. K. Keithley, Oncol. Nurs. Forum 42, E368–E375
mention that the observed low-SNR sound– (except the PO and VP) or how downstream (2015).
induced analgesia is unlikely to result from projections of the PO and VP account for music- 3. T. N. Nguyen, S. Nilsson, A.-L. Hellström, A. Bengtson,
some reduction in anxiety or stress, and it induced analgesia deserve further investigation. J. Pediatr. Oncol. Nurs. 27, 146–155 (2010).
4. L. Hartling et al., JAMA Pediatr. 167, 826–835 (2013).
probably does not directly involve attention- This study suggests a complement for the 5. C. Boyd-Brewer, R. McCaffrey, Holist. Nurs. Pract. 18, 111–118
distraction in affecting pain perception (33), classic pain sensation pathway that is impli- (2004).

6. M. P. Jensen, S. Hakimian, L. H. Sherlin, F. Fregni, J. Pain 9, 25. H. L. Read, J. A. Winer, C. E. Schreiner, Curr. Opin. Neurobiol. Natural Science Foundation of Anhui Province (KJ2020A0138,
193–199 (2008). 12, 433–440 (2002). 2008085QC115, and 2008085QC114), the University of Science and
7. E. A. Garza Villarreal, E. Brattico, L. Vase, L. Østergaard, 26. J. A. Winer, in The Mammalian Auditory Pathway: Technology of China Research Funds of the Double First-Class
P. Vuust, PLOS ONE 7, e29397 (2012). Neuroanatomy, D. B. Webster, A. N. Popper, R. R. Fay, Eds. Initiative (YD3460002001), the NIH NIDCR intramural research
8. S. J. Lunde, P. Vuust, E. A. Garza-Villarreal, L. Vase, Pain 160, (Springer, 1992), pp. 222–409. program (ZIA DE000757), and the Innovative Research Team of High-
989–993 (2019). 27. T. V. Salomons, G. D. Iannetti, M. Liang, J. N. Wood, JAMA level Local Universities in Shanghai. Author contributions: W.Zho.,
9. C. E. Dobek, M. E. Beynon, R. L. Bosma, P. W. Stroman, J. Pain Neurol. 73, 755–756 (2016). C.Y., Y.L., W.T., and Z.Z. initiated and designed the research. W.Zho.,
15, 1057–1068 (2014). 28. A. V. Apkarian, M. C. Bushnell, R. D. Treede, J. K. Zubieta, H.W., Y.L., W.T., and Z.Z. wrote the manuscript. W.Zho., C.Y., H.W.,
10. M. E. Sachs, A. Habibi, A. Damasio, J. T. Kaplan, Neuroimage Eur. J. Pain 9, 463–484 (2005). Y.M., W.Zhan., C.-L.Y., T.L., L.H., A.L., and W.Zhao performed all
218, 116512 (2020). 29. W. Zhou et al., Nat. Neurosci. 22, 1649–1658 (2019). experiments and analyzed and interpreted the results. L.C.
11. C. Usui et al., Pain Med. 21, 1546–1552 (2020). 30. D. D. Price, Mol. Interv. 2, 392–403 (2002). contributed to the discussion of the results. Competing interests:
12. L. Bonetti et al., Neuroimage 245, 118735 (2021). 31. H. Head, G. Holmes, Brain 34, 102–254 (1911). The authors declare that they have no competing interests. Data and
13. E. G. Jones, The Thalamus (Cambridge Univ. Press, ed. 2, 32. M. E. Diamond, M. Armstrong-James, M. J. Budway, F. F. Ebner, materials availability: All data needed to evaluate the conclusions in
2007). J. Comp. Neurol. 319, 66–84 (1992). the paper are present in the paper or the supplementary materials.
14. S. M. Sherman, R. W. Guillery, Phil. Trans. R. Soc. Lond. B 357, 33. K. Wiech, M. Ploner, I. Tracey, Trends Cogn. Sci. 12, 306–313 (2008). License information: Copyright © 2022 the authors, some rights
1695–1708 (2002). 34. S. Leknes, I. Tracey, Nat. Rev. Neurosci. 9, 314–320 (2008). reserved; exclusive licensee American Association for the
15. A. Mouraux, A. Diukova, M. C. Lee, R. G. Wise, G. D. Iannetti, 35. W. Ren et al., Nat. Neurosci. 19, 220–222 (2016). Advancement of Science. No claim to original US government works.
Neuroimage 54, 2237–2249 (2011). 36. M. L. Smith, N. Asada, R. C. Malenka, Science 371, 153–159 (2021). https://www.science.org/about/science-licenses-journal-article-reuse
16. B. A. Vogt, D. L. Rosene, D. N. Pandya, Science 204, 205–207 37. V. N. Salimpoor, M. Benovoy, K. Larcher, A. Dagher,
(1979). R. J. Zatorre, Nat. Neurosci. 14, 257–262 (2011).
SUPPLEMENTARY MATERIALS
17. A. N. Viaene, I. Petrof, S. M. Sherman, Proc. Natl. Acad. 38. M. N. Baliki et al., Nat. Neurosci. 15, 1117–1119 (2012).
Sci. U.S.A. 108, 18156–18161 (2011). science.org/doi/10.1126/science.abn4663
18. K. S. Meda et al., Neuron 102, 944–959.e3 (2019). ACKN OWLED GMEN TS Materials and Methods
19. X. Zhu et al., Nat. Neurosci. 24, 542–553 (2021). Figs. S1 to S24
We thank Y. Yang for providing Ai9 (RCL-tdT) mice. We thank F. Xu and
20. G. Wagner, M. Koschke, T. Leuf, R. Schlösser, K.-J. Bär, Tables S1 and S2
T. Xue for technical support, C. Weisz, and K. Suthakar for critical
Neuropsychologia 47, 980–987 (2009). References (39, 40)
reading. Funding: This work was supported by the National Key
21. X. R. Xiong et al., Nat. Commun. 6, 7224 (2015). MDAR Reproducibility Checklist
Research and Development Program of China Brain Science and
22. G. Gilam, J. J. Gross, T. D. Wager, F. J. Keefe, S. C. Mackey, Movies S1 to S8
Brain-Like Intelligence Technology (2021ZD0203100), the National
Neuron 107, 17–21 (2020). Natural Science Foundation of China (32025017, 32000717, and View/request a protocol for this paper from Bio-protocol.
23. J. W. Grau, R. L. Hyson, S. F. Maier, J. Madden4th, 82101301), the Science Fund for Creative Research Groups of the
J. D. Barchas, Science 213, 1409–1411 (1981). National Natural Science Foundation of China (32121002), the CAS Submitted 30 November 2021; accepted 18 May 2022
24. S. Koelsch, Nat. Rev. Neurosci. 15, 170–180 (2014). Project for Young Scientists in Basic Research (YSBR-013), the 10.1126/science.abn4663

RES EARCH
◥ tors. Experiments are highlighted that attempt

REVIEW SUMMARY to link the phenomenon of nonsaturating re-
sistivity to parameter-free universal physics. A
PHYSICS key experimental observation in such mate-
rials is that removing a single electron affects
Stranger than metals the spectrum at all energy scales, not just the
low-energy sector as in a Fermi liquid. It is
Philip W. Phillips*, Nigel E. Hussey, Peter Abbamonte observations of this sort that reinforce the
breakdown of the single-particle concept. On
the theoretical side, the modern accounts
BACKGROUND: Landau’s Fermi liquid theory in physics, attracting the attention of con- that borrow from the conjecture that strongly
provides the bedrock on which our under- densed matter physicists as well as string interacting physics is really about gravity are
standing of metals has developed over the theorists. One thing is clear in this regime: discussed extensively, as they have been the
past 65 years. Its basic premise is that the The particle picture breaks down. As particles most successful thus far in describing the
electrons transporting a current can be treated and locality are typically related, the strange range of physics displayed by strange metals.
as “quasiparticles”—electron-like particles metal raises the distinct possibility that its The foray into gravity models is not just a
whose effective mass has been modified, typ- resolution must abandon the basic building pipe dream because in such constructions,
ically through interactions with the atomic blocks of quantum theory. no particle interpretation is given to the
lattice and/or other electrons. For a long We review the experimental and theoretical charge density. As the breakdown of the
time, it seemed as though Landau’s theory studies that have shaped our current under- independent-particle picture is central to
could account for all the many-body inter- standing of the emergent strongly interacting the strange metal, the gravity constructions
actions that exist inside a metal, even in the physics realized in a host of strange metals, are a natural tool to make progress on this
so-called heavy fermion systems whose quasi- with a special focus on their poster-child: the problem. Possible experimental tests of this
particle mass can be up to three orders of copper oxide high-temperature superconduc- conjecture are also outlined.
magnitude heavier than the electron’s
mass. Fermi liquid theory also lay the OUTLOOK: As more strange metals emerge
foundation for the first successful micro- and their physical properties come under
scopic theory of superconductivity. the scrutiny of the vast array of ex-
In the past few decades, a number of perimental probes now at our disposal,
new metallic systems have been discovered their mysteries will be revealed and
Black hole
that violate this paradigm. The violation is their commonalities and differences
most evident in the way that the electrical cataloged. In so doing, we should be
resistivity changes with temperature or able to understand the universality of
magnetic field. In normal metals in which strange metal physics. At the same
electrons are the charge carriers, the time, the anomalous nature of their
resistivity increases with increasing tem- superconducting state will become
perature but saturates, both at low tem- apparent, offering us hope that a
peratures (because the quantized lattice new paradigm of pairing of non-
vibrations are frozen out) and at high quasiparticles will also be formalized.
temperatures (because the electron mean The correlation between the strength
free path dips below the smallest scatter- of the linear-in-temperature resistivity
Resistivity (Ω•cm)
ing pathway defined by the lattice spacing). in cuprate strange metals and their
In “strange metals,” by contrast, no satu- corresponding superfluid density, as
ration occurs, implying that the quasipar- revealed here, certainly hints at a fun-
ticle description breaks down and electrons damental link between the nature of
K)
are no longer the primary charge carriers. re ( strange metallicity and superconduc-
eratu
When the particle picture breaks down, no
Te mp tivity in the cuprates. And as the gravity-
local entity carries the current. inspired theories mature and overcome
the challenge of projecting their power-
ADVANCES: A new classification of metal- ful mathematical machinery onto the
licity is not a purely academic exercise, appropriate crystallographic lattice,
however, as strange metals tend to be the Curved spacetime with a black hole in its interior and the so too will we hope to build with con-
high-temperature phase of some of the strange metal arising on the boundary. This picture is fidence a complete theory of strange
best superconductors available. Under- based on the string theory gauge-gravity duality conjecture by metals as they emerge from the horizon
standing high-temperature superconduc-
tivity stands as a grand challenge because
J. Maldacena, which states that some strongly interacting
quantum mechanical systems can be studied by replacing them
of a black hole.
▪
its resolution is fundamentally rooted in with classical gravity in a spacetime in one higher dimension.
the physics of strong interactions, a re- The conjecture was made possible by thinking about some of the The list of author affiliations is available in the full article
gime where electrons no longer move inde- fundamental components of string theory, namely D-branes online.
*Corresponding author. Email: dimer@illinois.edu
pendently. Precisely what new emergent (the horseshoe-shaped object terminating on a flat surface in
Cite this article as P. W. Phillips et al., Science 377,
phenomena one obtains from the inter- the interior of the spacetime). A key surprise of this conjecture eabh4273 (2022). DOI: 10.1126/science.abh4273
actions that drive the electron dynamics is that aspects of condensed matter systems in which the
above the temperature where they super- electrons interact strongly—such as strange metals—can be READ THE FULL ARTICLE AT
conduct is one of the most urgent problems studied using gravity. https://doi.org/10.1126/science.abh4273
Phillips et al., Science 377, 169 (2022) 8 July 2022 1 of 1

RES EARCH
◥ if ttr = a1tP with a1 (the coefficient of T-linear

REVIEW resistivity) ~ 1. Although this state of affairs
seems to be realized in a host of correlated
PHYSICS metals, including the cuprates (7–10), there is
no consensus concerning how accurately a1 can
Stranger than metals be known and the assumptions that go into its
determination. Regardless of the possible rela-
Philip W. Phillips1*, Nigel E. Hussey2,3, Peter Abbamonte4 tionship with Planckian dissipation, what
makes T-linear resistivity in the cuprates truly
In traditional metals, the temperature (T) dependence of electrical resistivity vanishes at low or high novel is its persistence—from millikelvin tem-
T, albeit for different reasons. Here, we review a class of materials, known as “strange” metals, that can peratures (in both the electron- and hole-
violate both of these principles. In strange metals, the change in slope of the resistivity as the mean doped cuprates) (11, 12) up to 1000 K (in the
free path drops below the lattice constant, or as T → 0, can be imperceptible, suggesting continuity hole-doped cuprates) (4, 6)—and its omni-
between the charge carriers at low and high T. We focus on transport and spectroscopic data on presence, the strange metal regime dominating
candidate strange metals in an effort to isolate and identify a unifying physical principle. Special large swathes of the temperature versus doping
attention is paid to quantum criticality, Planckian dissipation, Mottness, and whether a new gauge phase diagram (13). In normal metals (14, 15) as
principle is needed to account for the nonlocal transport seen in these materials. well as some heavy fermion materials (16), the
resistivity asymptotically approaches a satura-
tion value at which the mean free path ‘ be-
T
o understand the essential tension be- no system parameters enter tP, this quantity comes comparable with the interatomic spacing
tween quantum mechanics and gravity, occupies a similar fundamental role in anal- a or more generally the Fermi wavelength lF—
simply imagine two electrons impinging ogy to the Planck length and is referred to as the minimum length over which a Bloch wave
on the event horizon of a black hole. the Planckian dissipation time. Equation 2 has and its associated Fermi velocity and wave vec-
Whereas classical gravity predicts that had previous incarnations (1, 2); in the realm tor can be defined. In many correlated metals,
they meet at the center, quantum mechanics of charge transport, relevant to this article, it collectively referred to as “bad metals,” ‘ < a at
forbids this if the electrons have the same defines the time scale for scale-invariant or high T, which violates the so-called Mott-Ioffe-
spin. In essence, classical gravity has no way Planckian dissipation (3). Scale invariance fol- Regel (MIR) limit (5, 6, 1, 16–18). Remarkably,
of preserving Pauli exclusion. Replacing clas- lows because there is no scale other than tem- no saturation occurs in these bad metals across
sical general relativity with a quantum theory perature appearing in tP. Achieving such scale the MIR threshold, implying that the whole
of gravity at small enough scales resolves the invariance necessitates a highly entangled many- notion of a Fermi velocity of quasiparticles
problem, but what is this scale? body state. Such a state would lead to the break- breaks down at high T. In certain cases, an
In 1899, Planck formulated a universal length down of a local single-particle framework and example of which is shown in Fig. 1, there is no
now regarded as the scale below which a quan- the advent of new collective nonlocal entities as discernible change in slope as the MIR limit is
tum theory of gravity supplants its classical the charge carriers. Identifying the new prop- exceeded. Although this circumstance occurs
counterpart. The Planck scale, agating degrees of freedom constitutes the key only in a narrow doping window (in cuprates)
rffiffiffiffiffiffiffi mystery of strange metals. (18), such continuity does suggest that, even
ħG Whereas the Planck scale ‘P requires high- at low T, quasiparticles (19) cannot be the
‘P ¼ ð1Þ
c3 energy accelerators much beyond anything
now in use, such is not the case with physics
obtains by pure dimensional analysis on three at the Planckian dissipation limit. Early tabletop
fundamental constants: the speed of light, experiments on cuprate superconductors, for
c, Newton’s gravitational constant, G, and the example, revealed a “strange metal” regime
quantum of uncertainty, ħ (Planck’s constant, h, defined by a robust T-linear resistivity extend-
divided by 2p). This leads naturally to a Planck ing to the highest temperatures measured (4–6)
time as the ratio of the Planck length to the (Fig. 1), a possible harbinger of Planckian dissi-
speed of light, ‘P/c. Such a Planckian analysis pation. Recall that in a Fermi liquid, the con-
can be extended to many-body systems in conductivity can be well described by a Drude
tact with a heat bath. All that is necessary is to formula,
include the temperature T. A similar dimen-
ne e2
sional analysis then leads to s¼ ttr ð3Þ
m
ħ
tP ¼ ð2Þ where ne is the charge carrier density, e is the
kB T
charge of an electron, m is its mass, and ttr is
as the shortest time for heat loss in a many- the transport lifetime, defined as
body system obeying quantum mechanics,
where kB is Boltzmann’s constant. Because ħEF EF
ttr ¼ 2 ¼ k T tP ð4Þ
ðkB T Þ B Fig. 1. T-linear resistivity in strange metals.
1
Shown is the in-plane resistivity of La2–xSrxCuO4
Department of Physics and Institute for Condensed Matter which contains the Fermi energy EF of the (x = 0.21). The dotted points are extrapolated from
Theory, University of Illinois, Urbana, IL 61801, USA. 2H. H.
Wills Physics Laboratory, University of Bristol, Bristol BS8 quasiparticles. No such energy scale appears high-field magnetoresistance data (8). The shaded
1TL, UK. 3High Field Magnet Laboratory (HFML-EMFL) and in Eq. 2. If the scattering rate in cuprates is area shows the Mott-Ioffe-Regel (MIR) boundary
Institute for Molecules and Materials, Radboud University, directly proportional to the resistivity, as it is defined here as when the mean free path becomes
6525 ED Nijmegen, Netherlands. 4Department of Physics,
University of Illinois, Urbana, IL 61801, USA. in simple metals, T-linear resistivity is equivalent comparable to the Fermi wavelength lF. [Adapted
*Corresponding author. Email: dimer@illinois.edu to scale-invariant Planckian dissipation only from (8, 18)]
Phillips et al., Science 377, eabh4273 (2022) 8 July 2022 1 of 10

RES EARCH | R E V I E W
effective propagating degrees of freedom. carry the current, and if not, does explicating Sr2RuO4 (36), exhibit T-linear resistivity at high
Evidently, in strongly correlated electron mat- the propagating degrees of freedom in the T with a magnitude that clearly violates the
ter, the current-carrying degrees of freedom in strange metal require a theory as novel as MIR limit, but as the system cools down, con-
the infrared (IR) need not have a particle quantum gravity? ventional Fermi-liquid behavior is restored
interpretation. (37, 63). Hence, although they are bona fide
Over time, the label “strange metal” has Is strange metallicity ubiquitous? bad metals—exhibiting metallic resistivity be-
seemingly become ubiquitous, used to describe A survey of the DC transport properties of yond the MIR limit—they do not classify as
any metallic system whose transport properties several strange metal candidates is presented strange (16, 64).
display behavior that is irreconcilable with in Table 1 (4, 6, 8, 10, 20–61). In addressing Another subset, identified here as quantum
conventional Fermi liquid or Boltzmann trans- the above question, we must first acknowl- critical metals, exhibit T-linear resistivity down
port theory. This catch-all phraseology, how- edge the many definitions of strange metallic to the lowest temperatures studied, but only at
ever, is unhelpful, as it fails to differentiate behavior that exist, the simplest being a ma- a singular quantum critical point (QCP) in their
between the various types of non–Fermi liquid terial hosting a metallic-like resistivity in the T versus tuning parameter g phase diagram at
behavior observed, some of which deserve spe- absence of quasiparticles. A more precise, if which a continuous quantum phase transition
cial deliberation on their own. We attempt empirical, definition centers on the T-linear to a symmetry-broken phase is suppressed to
to bring strange metal phenomenology into resistivity, specifically one that is distinguish- T = 0. Here, g can be pressure, magnetic field,
sharper focus by addressing a number of per- able from the resistivity manifested by simple doping level, or even strain. In most cases, the
tinent questions. Does the term refer to the metals (which is attributed to electron-phonon phase transition in question is associated with
resistive behavior of correlated electron sys- scattering). For a metal to be classified as finite-momentum antiferromagnetism [as in
tems at high or low temperatures, or both? strange, the T-linearity must extend far beyond pure YbRh2Si2 (49), CeCoIn5 (55), and BaFe2
Does it describe any T-linear resistivity asso- the typical bounds associated with phonon- (As1–xPx)2 (45)], although similar behavior has
ciated with the Planckian time scale, or some- mediated resistivity. At low T, this is typically recently been reported in systems exhibit-
thing unique? Does it describe the physics of one-third of the Debye temperature, whereas ing zero-momentum order, such as nematic
a doped Mott insulator or the physics associated at high T, it is the temperature at which the FeSe1–xSx (41) or ferromagnetic CeRh6Ge4
with quantum criticality (whose underlying magnitude of the resistivity is roughly half the (57). Away from the QCP, the low-T resistivity
origins may or may not include Mottness as value corresponding to the MIR limit. A subset recovers the canonical T2 Fermi liquid form,
a key ingredient)? Finally, does anything local of correlated metals, such as SrRuO3 (62) and albeit with a coefficient that is enhanced as the
Table 1. Summary of the DC transport properties of various strange finite region of the phase diagram. Fifth column: T2 dependence of the
metal candidates. First column: Candidate compound or family of inverse Hall angle cot QH in the same temperature range where r(T)
compounds. For the hole-doped cuprates, underdoped (UD), optimally is T-linear. Sixth column: Compounds satisfying the “Modified Kohler’s”
doped (OP) and overdoped (OD) compounds are treated separately; the label have a low-field magnetoresistance (MR), defined as [r(H, T) –
transport properties of individual compounds within each subset are r(0, T)]/r(0, T), that exhibits a T dependence similar to that of tan2 QH.
generic. For the electron-doped cuprates, only La2–xCexCuO4 is listed; Seventh and eighth columns: High-field MR behavior of strange metal
Pr- and Nd-based compounds show similar behavior. Second column: Bad candidates. The observation of an H-linear MR at high fields does not
metallic behavior; a check mark indicates T-linear resistivity beyond the imply that the MR exhibits quadrature scaling over all fields and
Mott-Ioffe-Regel (MIR) limit. A cross indicates either a tendency toward temperatures. *FeSe1–xSx H-linear/quadrature MR seen in this family
saturation or a marked reduction in slope near the MIR limit. Third coexists with a more conventional MR contribution, indicating the
column: A check mark identifies systems that at any point in their presence of both strange metal and Fermi liquid–like components in the
respective phase diagram(s) exhibit T-linear resistivity down to the lowest DC transport. **In YbBAl4, although T-linear resistivity is observed over a
temperatures studied. Fourth column: “Extended criticality” refers to wide pressure range, its limiting low-T dependence is T1.5. A dash
systems where a predominant T-linear resistivity at low T extends over a indicates no reports confirming or disproving the considered behavior.
r º T as r º T as Extended cot QH º T2 Modified Kohler’s H-linear MR Quadrature

T→∞ T→0 criticality (at low H) (at low H) (at high H) MR
UD p-cuprates √ (6) × (20) × (21) √ (22) √ (23) — —
.............................................................................................................................................................................................................................................................................................................................................
OP p-cuprates √ (4) — — √ (24) √ (25) √ (26) × (27)
.............................................................................................................................................................................................................................................................................................................................................
OD p-cuprates √ (6) √ (8) √ (8) √ (28) × (29) √ (29) √ (29)
.............................................................................................................................................................................................................................................................................................................................................
La 2–x Ce x CuO 4 × (30) √ (31, 32) √ (31, 32) × (33) × (34) √ (35) × (35)
.............................................................................................................................................................................................................................................................................................................................................
Sr 2RuO4 √ (36) × (37) × (38) × (39) × (37) × (37) × (37)
.............................................................................................................................................................................................................................................................................................................................................
Sr 3 Ru 2 O 7 √ (10) √ (10) × (10) × — — —
.............................................................................................................................................................................................................................................................................................................................................
FeSe1–xSx × (40) √ (41) × (41) √ (42) √ (42) √* (43) √* (43)
.............................................................................................................................................................................................................................................................................................................................................
BaFe 2(As1–xPx)2 × (44) √ (45) × (45) — √ (46) √ (47) √ (47)
.............................................................................................................................................................................................................................................................................................................................................
Ba(Fe 1/3 Co 1/3 Ni 1/3 ) 2 As 2 — √ (48) × (48) — — √ (48) √ (48)
.............................................................................................................................................................................................................................................................................................................................................
YbRh2Si2 × (49) √ (50) √ (51) √ (52) — — —
.............................................................................................................................................................................................................................................................................................................................................
YbBAl 4 × (53) √** (53) √** (53) — — — —
.............................................................................................................................................................................................................................................................................................................................................
CeCoIn5 × (54) √ (55, 56) × (55, 56) √ (54) √ (54) — —
.............................................................................................................................................................................................................................................................................................................................................
CeRh 6Ge4 × (57) √ (57) × (57) — — — —
.............................................................................................................................................................................................................................................................................................................................................
(TMTSF) 2PF6 — √ (58) √ (58) — — — —
.............................................................................................................................................................................................................................................................................................................................................
MATBG √ (59) √ (60) √ (60) √ (61) — — —
.............................................................................................................................................................................................................................................................................................................................................

QCP is approached and the order parameter scaling MR coexists with a more conventional (MATBG) (60)], pressure [for YbBAl4 (53) and
fluctuations soften. orbital MR (43). Both sets of behavior highlight the organics (58)], or magnetic field [for Ge-
By contrast, in overdoped cuprates [both once again the possible coexistence of two re- doped YbRh2Si2 (51)]. If quantum criticality is
hole- (8, 9) and electron-doped (31, 32)], Ge- laxation times or two distinct charge-carrying the cause, then it is difficult to fathom how a
doped YbRh2Si2 (51), YbBAl4 (53), and the sectors in real materials. Curiously, in single- thermodynamic quantity can be fashioned to
organic Bechgaard salts (58), r(T) is predom- band materials, quadrature scaling breaks diverge over an entire phase.
inantly T-linear down to low T not at a singular down inside the pseudogap regime (26, 27), Despite these difficulties, it is worth explor-
point in these materials’ respective phase dia- whereas modified Kohler’s scaling is recov- ing the connection T-linear resistivity has with
grams, but over an extended range of the ered (23, 25), suggesting that the two pheno- continuous quantum critical phenomena, which
relevant tuning parameter. At first sight, this mena may be mutually exclusive in such for the sake of argument we presume to be tied
“extended criticality” is difficult to reconcile materials. In multiband materials such as to a singular point in the phase diagram. Re-
with current theories of quantum criticality, FeSe1–xSx, on the other hand, these differ- gardless of the origin of the QCP, universality
which predict a crossover to a purely T2 resis- ent manifestations of strange metallic trans- allows us to answer a simple question: What
tivity and thus a recovery of Fermi liquid be- port appear side by side (42, 43). Irrespective constraints does quantum criticality place on
havior at low T everywhere except at the of these caveats and complexities, what is the T dependence of the resistivity? The answer
(singular) QCP. Arguably, it is this feature— striking about the quadrature-scaling MR is to this question should be governed only by the
incompatibility with both standard Fermi liquid that it occurs in systems with varied Fermi fundamental length scale for the correlations.
and quantum critical scenarios—that distin- surface topologies, dominant interactions, and The simplest formulation of quantum critical-
guishes a genuine strange metal. Intriguingly, energy scales, hinting at some universal but as ity is single-parameter scaling in which the
in many of these systems, a1 is found to scale yet unidentified organizing principle. spatial and temporal correlations are governed
with the superconducting transition temper- Restricting the strange metal moniker, as by the same diverging length (see Fig. 2).
ature Tc. Moreover, for La2–xCexCuO4 (31, 32) done here, to materials that exhibit low-T Making the additional assumption that the
and bis-(tetramethyltetraselenafulvalene) T-linear resistivity over an extended region of relevant charge carriers are formed from the
hexafluorophosphate [(TMTSF)2PF 6] (58), phase space likewise restricts strange metal- quantum critical fluctuations, we find that a
extended criticality emerges once the spin licity to a select “club.” The following sections simple scaling analysis on the singular part of
density wave transition has been fully sup- explore various possible attributes that they the free energy results in the scaling law
pressed, suggesting an intimate link between have in common.
q2 kB T ðd2Þ=z
the strange metal transport, superconductivity, sðw ¼ 0; T Þ ¼ f ðw ¼ 0Þ ð5Þ
and the presence of critical or long-wavelength Is it quantum criticality? ħ ħc
spin fluctuations. In hole-doped cuprates, how- Scale-free T-linear resistivity is highly sugges- (73) for the T dependence of the conductivity,
ever, the strange metal regime looks different, tive of some form of underlying quantum crit- where f(w = 0) is a nonzero constant, q is the
in the sense that the extended criticality em- icality in which the only relevant scale is the
erges beyond the end of the pseudogap regime temperature governing collisions between
that does not coincide with a magnetic quan- excitations of the order parameter (66). In
tum phase transition (65). Furthermore, al- fact, following the advent of marginal Fermi
though the pseudogap plays host to a multitude liquid (MFL) phenomenology and its associated
of broken-symmetry states, the jury is still out (T, w)-linear self energies (67), the common
as to whether any of these are responsible for interpretation of such T-linear resistivity was
pseudogap formation or are merely instabil- and still remains the nucleus of ideas centered
ities of it. on quantum criticality. The strict definition of
Besides T-linear resistivity, strange metals quantum criticality requires the divergence of
also exhibit anomalous behavior in their mag- a thermodynamic quantity. In heavy fermion
netotransport, including (i) a quadratic tem- metals, the electronic heat capacity ratio Cel/T
perature dependence of the inverse Hall angle indeed grows as ln(1/T) as the antiferromag-
cot QH = sxy/sxx (24), (ii) a transverse magnetic correlations diverge (49, 55, 68). In cer-
netoresistance (MR) that at low field exhibits tain hole-doped cuprates, Cel/T also scales as
modified Kohler’s scaling [Dr/r(0) º tan2 ln(1/T) at doping levels close to the end of
QH º (1/T2)2 or 1/(A + BT2)2 (25)], and/or (iii) the pseudogap regime (69), although here,
an H-linear MR at high fields that may or evidence for a divergent length scale of an Fig. 2. Single-parameter scaling hypothesis.
may notqfollow quadrature
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi scaling [whereby associated order parameter is currently lacking Depicted here is the collective scaling of a physical
Dr/T º 1 þ gðH=T Þ2] (43, 47). The combination (70). Moreover, photoemission suggests that at system near a critical point. The essential idea is
of a modified Kohler’s rule and T2 Hall angle a T-independent critical doping pc ≈ 0.19, all that the correlations within each of the blocks
has been interpreted to indicate the presence signatures of incoherent spectral features that shown are independent of one another. That is,
of distinct relaxation times, either for differ- define the strange metal cease, giving way to a spatial correlations in a volume smaller than the
ent loci in momentum space (22) or for relax- more conventional coherent response (71). The correlation volume, xd, and temporal correlations on
ation processes normal and tangential to the abruptness of the transition suggests that it is a time scale shorter than xt are small, and space-
underlying Fermi surface (24). The H-linear first-order, posing a challenge to interpreta- time regions of size xdxt behave as independent
MR, on the other hand, is inextricably tied to tions based solely on criticality. blocks. At the critical point, the correlation length
the T-linear zero-field resistivity via its H/T As mentioned above, another major hurdle diverges. The single-scaling parameter hypothesis
scaling relation, a relation that can also extend for the standard criticality scenario is that the assumes that temporal correlations diverge
over a broad range of the relevant tuning T-linear resistivity persists over a wide range also as a simple power of the spatial correlation
parameter (29). In some cases, this link can be of the relevant tuneable parameter, whether length, namely xt º xz, where z is known as the
obscured, either because r(T) itself is not strict- doping [as is the case for cuprates (8, 9, 31, 32, 72) dynamical critical exponent and by causality
ly T-linear (29) or because the quadrature- and magic-angle twisted bilayer graphene must exceed unity.

charge, and z is the dynamical exponent, which criticality beyond single-parameter scaling, such S—the imaginary part of the self energy—can
from causality must obey the inequality z ≥ 1. as a noncritical form of the entropy suggested be used to define a lifetime through
Absent from this expression is any dependence recently (81), is at work here (see below). ħ
ħvk Dk ¼ Im Sðk; wÞ ¼ 2 ð8Þ
on an ancillary energy scale—for example, EF or Another critical feature of the conductivity is t
the plasma frequency wp—as the only assump- its behavior at finite wave vector k, which may be where vk is the group velocity for momentum
tion is scale-invariant transport irrespective of quantified by the dynamic charge susceptibility, state k. Extracting the slope from the data in
the details of the system. The analogous ex- k2 Fig. 2 of (92) and using the experimentally
pression for the optical conductivity is c″ðk; wÞ ¼ 2 ℜsðk; wÞ ð7Þ reported Fermi velocity vF = 1.1 eV/Å, we find
we
that the single-particle scattering rate ħ/t ~
sðw; T ¼ 0Þºwðd2Þ=z ð6Þ 1.7kBT (i.e., on the order of the Planckian limit).
determined from electron energy-loss spec-
(74). In pure YbRh2Si2, for example, s–1(w) fol- troscopy (EELS). A restriction on EELS is that Similar results were obtained in subsequent
lows an w-linear dependence at low frequencies it measures the longitudinal charge response, ARPES studies (92–94) with a key extension
in the same region of the (T, H) phase diagram— whereas optics yields the transverse. At van- introduced in (88) whereby the width of nodal
the quantum critical “fan”—where r(T) is also ishing momentum they are expected to be equal. states was observed to obey a quadrature form
linear, consistent with this notion of single- Because optics has no momentum resolution, described by the expression [(ħw)2 + (bkBT)2]l,
parameter scaling (75). In cuprates, on the comparison with EELS can only be made as where l is a doping-dependent exponent equal
other hand, the situation is more nuanced. At k → 0. The primary charge excitation in strange to ½ at optimal doping.
intermediate frequencies—sometimes referred metals is a 1-eV plasmon that was long believed This extraction of the scattering rate from
to as the mid-IR response—s(w) exhibits a to exhibit the same behavior as in a normal ARPES, however, is not entirely problem-free,
ubiquitous w–2/3 dependence (7). Although this Fermi liquid (82, 83). Recent high-resolution as vF is hard to define in ARPES experiments
feature in s(w) has been interpreted in terms of momentum-resolved EELS (M-EELS) measure- at energies close to the Fermi level and where,
quantum critical scaling (7), it is inconsistent ments have called this belief into question, for the most part, the width of the state ex-
with the single-parameter scaling described showing that the large-k response is domi- ceeds its energy. Indeed, the integral of the
above. At any doping level, s(w) in the cuprates nated by a continuum that remains flat to density of states using as input the vF extracted
exhibits a minimum at roughly the charge high energies, roughly 2 eV (84–86). Such from APRES measurements is found to account
transfer scale of 1 eV. This is traditionally behavior is reminiscent of the MFL (67) sce- for only half of the as-measured electronic spe-
(76, 77) used as the energy scale demarcating nario except in that picture, the continuum cific heat coefficient (95). Furthermore, this
the separation between intraband and inter- persists up to a cutoff scale determined by reliance on Fermiology (fermion phenome-
band transitions and hence serves to separate the temperature, and not the Mott scale of nology), present also in (10), leaves open the
the low-energy from the high-energy continua. 2 eV. In addition, the continuum exhibits precise meaning of figure 2 of (10), in which
It has long been debated whether the broad scale-invariant features but with a dynamical the magnitude of the T-linear transport
sub-eV s(w) response in cuprates is best an- critical exponent, z ~ ∞, not possible from a scattering rate extracted from the DC
alyzed in terms of one or two components simple QCP. resistivity is found to scale as 1/vF for a
(77, 78). In the former case, the w–2/3 tail is We conclude, then, that no form of tradi- series of materials that violate the MIR limit
simply a consequence of the strong w-linear tional quantum criticality can account easily at intermediate to high temperatures. Despite
dependence in 1/ttr(w)—as in MFL—whereas for the power laws seen in strange metallic this, a similar extraction in (9), again using
in the latter, it forms part of an incoherent transport (although we recognize that T-linear Fermiology but focusing on the low-T resistiv-
response that is distinct from the coherent resistivity is observed above what appear to be ity, also found a transport scattering rate close
Drude weight centered at w = 0, which itself is genuine singular QCPs). The photoemission to the Planckian bound. This consistency be-
described with either a constant or w-dependent experiments (71) indicating a first-order transi- tween the two analyses reflects the curious
scattering rate. tion pose an additional problem exacerbated by fact that the T-linear slope of the DC resistivity
Returning to the DC resistivity, we find that the possibility that the criticality might be rele- does not vary markedly as the MIR threshold
in cuprates, where d = 3, an exponent z = –1 is vant to a whole region (8, 9, 51, 53, 58, 60, 65, 87, 88) is crossed. It does not, however, necessarily jus-
required to account for the T-linear depen- rather than a point. Such criticality over an tify either approach in validating T-linear scat-
dence; this value is strictly forbidden by cau- extended region is reminiscent of critical tering at the Planckian limit. Finally, although
sality (73). For d = 2, as in the case of MATBG, charged matter (89, 90) arising from dilatonic T-linearity and Planckian dissipation appear
the T dependence vanishes. This can be fixed models in gauge-gravity duality. These ideas synonymous in the cuprates, this is not uni-
with the replacement of d → d* = 1 for both have been the most successful thus far in versally the case. In YbRh2Si2 (75), for example,
materials. Although d* can be construed as the reproducing the various characteristics of the T-linear scattering rate is found to deviate
number of dimensions (79) transverse to the strange metal physics (see below and Table 2). strongly from the Planckian limit with ttr ~ 0.1tP
Fermi surface, it is difficult to justify such a (52), and in the electron-doped cuprates, the
procedure here, as the persistence of T-linearity Is it Planckian dissipation? notion of a Planckian limit to the scattering
with no change in slope above and below the Whereas the electrical resistivity in metals can rate has recently been challenged (96). This cer-
MIR requires a theory that does not rely on FL be measured directly, the scattering rate is tainly adds to the intrigue regarding quantum
concepts such as a Fermi velocity or energy. entirely an inferred quantity. Herein lies the criticality as the underlying cause of Planckian
Furthermore, it is well known that introducing catch with Planckian dissipation. Angle-resolved dissipation, for which several alternative pro-
d* yields a power law for the heat capacity, C º photoemission (ARPES) experiments on cup- posals have recently emerged (97, 98).
T3/2, which is not seen experimentally (80). On rates as early as 1999 reported that the width In principle, the optical conductivity per-
dimensional grounds, the z = –1 result in the of the momentum distribution curves (MDCs) mits an extraction of t without recourse to
context of the Drude formula is a consequence at optimal doping along the nodal direction Fermiology. Within a Drude model, the optical
of compensating the square power of the plasma [(0, 0) to (p, p)] scale as a linear function of conductivity
frequency with powers of T so that the scaling temperature and a0 + 0.75w for frequencies
form of Eq. 5 is maintained. A distinct possibility that exceed 2.5kBT (91). The momentum line- 1 w2p ttr
sðwÞ ¼ ð9Þ
is that perhaps some other form of quantum width, which in photoemission enters as Im 4p 1 þ iwttr

Table 2. Snapshot of current theoretical modeling of the strange metal ED, exact diagonalization; CA, cold atoms; DMFT/EDMFT, dynamical mean-
regime. Indicated are consistency with T-linear resistivity, w–2/3 scaling of field theory/embedded dynamical mean-field theory; A-B, Aharonov-Bohm
the mid-IR optical conductivity, quadrature-scaling magnetoresistance, effect; ECFL, extremely correlated Fermi liquid; QCP, quantum critical point.
extended quantum criticality, and what predictions are made in terms of *T-linear resistivity is an input. **A slope change occurs through the MIR.
experimental observables. Scenarios: MFL, marginal Fermi liquid; EFL, ersatz ***Quadrature scaling obtained only for a bivalued random resistor model
Fermi liquid; SYK, Sachdev-Ye-Kitaev; AdS/CFT, anti–de Sitter space/ (121) with equal weights (27). ****Although this scaling was thought to arise
conformal field theory conjecture; AD/EMD, anomalous dimensions/ in pure AdS with an inhomogeneous charge density (123), later studies
Einstein-Maxwell-dilaton; HM, Hubbard model; QMC, quantum Monte Carlo; (124, 125) found otherwise.
r º T as T → 0 r º T as T → ∞ s º w–2/3 Quadrature MR Extended criticality Experimental prediction

Phenomenological
............................................................................................................................................................................................................................................................................................................................................
MFL √ (67) × (67) × × × Loop currents (107)
............................................................................................................................................................................................................................................................................................................................................
EFL — * — — × × Loop currents (108)
............................................................................................................................................................................................................................................................................................................................................
Numerical
............................................................................................................................................................................................................................................................................................................................................
ECFL × (109) — — × ×
............................................................................................................................................................................................................................................................................................................................................
HM (QMC/ED/CA) — (110) √ (110–114) × — — —
............................................................................................................................................................................................................................................................................................................................................
DMFT/EDMFT √ (115) √ (116, 117) × — √ (117) —
............................................................................................................................................................................................................................................................................................................................................
QCP (118) — — — × —
............................................................................................................................................................................................................................................................................................................................................
Gravity-based
............................................................................................................................................................................................................................................................................................................................................
SYK √ (119, 120) √** (120) × √*** (121) — ×
............................................................................................................................................................................................................................................................................................................................................
AdS/CFT √ (122) √ (122) √**** (90, 126) × × ×
............................................................................................................................................................................................................................................................................................................................................
AD/EMD √ (127–129) √ (90, 126, 127, 129, 130) √ (90, 126, 130) × √ (126) Fractional A-B (129)
............................................................................................................................................................................................................................................................................................................................................
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
contains only ttr and wp = 4pne e2 =m. At zero spectral weight transfer from the UV to the IR Mottness (99–102), as it is ubiquitous in Mott
frequency, the Drude formula naturally yields is absent. On the other hand, hole or electron systems and strictly absent in weakly correlated
the DC conductivity sDC and an estimate for doping of the parent cuprate induces a marked metals. Even in many of the strange or quantum
the relaxation rate can be extracted from the transfer of spectral weight of roughly 1 to 2 eV. critical metals described in Table 1, there is little
width at half maximum of the full Drude re- As a result, the low-energy spectral weight or no evidence that Mottness is playing a notable
sponse. However, there is an important caveat: grows (77, 99–102) at the expense of the degrees role. Such a distinction may thus offer a hint
ttr is frequency-dependent in the cuprates, a of freedom at high energy, a trend that persists to the source of the uniqueness of the cuprate
condition that is consistent with various phys- (76) even inside the superconducting state. strange metal. In bad metals, on the other hand,
ical models including both the Fermi liquid This is an intrinsic feature of Mott systems, a gradual transfer of low-frequency spectral
and MFL scenarios as well as the large body namely that the number of low-energy degrees weight out to energies on the order of the Mott
(88, 91) of MDC analysis performed on the of freedom is derived from the high-energy scale is almost universally observed with in-
cuprates. Although this prevents a clean sep- spectral weight. As this physics is distinct creasing temperature (16), suggesting that
aration of the conductivity into coherent and from that of a Fermi liquid and intrinsic to Mottness is one of the key components of
incoherent parts, it was shown in (7) that in Mott physics, it is termed “Mottness” (102). bad metallic transport.
the low-frequency limit, w < 1.5kBT/ħ, ttr ~ 0.8tP, Notably, the mid-IR response with its charac-
in agreement with the DC analysis of (9). teristic w–2/3 scaling is absent from the parent
A second key issue remains: How can such Mott insulating state. Hence, it must reflect the
Drude analysis be justified for those strange doping-induced spectral weight transfer across
metals in which the MIR limit is violated and the Mott gap. It is perhaps not a surprise, then,
the Drude peak shifts to finite frequencies (16)? that no low-Tc material exhibits such a mid-
Indeed, in the high-T limit, “bad metallicity” IR feature. In fact, some theories of cuprate
can be ascribed to a transfer of spectral weight superconductivity (103) credit its origin to
from low to high w, rather than to an ever- the mid-IR scaling. We can quantify the total
increasing scattering rate (that within a Drude number of low-energy degrees of freedom that
picture results in a continuous broadening of arise from the UV-IR mixing across the Mott
the Lorentzian fixed at zero frequency). Given gap by integrating the optical conductivity,
the marked crossover in the form of s(w) at low
2mVcell W
frequencies, it is indeed mysterious that the Neff ðWÞ ¼
pe2 0
∫ sðwÞdw ð10Þ
slope of the T-linear resistivity continues un-
abated with no discernible change. up to the optical gap W ≈ 1.2 eV, where Vcell is Fig. 3. Integrated optical conductivity Neff
the unit cell volume. The energy scale of 1.2 eV exceeds dopant concentration. Shown is Neff for
Is it Mottness? corresponds to the minimum of the optical con- electron-doped Pr2–xCexCuO4–d (triangles) and hole-
Table 1 encompasses a series of ground states ductivity, as mentioned above. In a rigid-band doped La2–xSrxCuO4–d (circles) The dashed line
from which T-linear resistivity emerges. In some semiconductor model in which such spectral indicates what is expected for doping a semi-
of these materials, such as the heavy fermions, weight transfer is absent, Neff = x, where x is the conductor. The expectation is that each Ce or Sr
the high- and low-energy features of the spec- number of holes. In the cuprates, however, Neff atom contributes just a single charge carrier.
trum are relatively distinct in the sense that exceeds x (Fig. 3). This is the defining feature of [Reprinted with permission from (77)]

The optical response in cuprates tells us that interactions might seem initially unphysical, are at least conformally invariant in d dimen-
there are degrees of freedom that couple to SYK models are nonetheless natural candidates sions, are dual to a theory of gravity in a d +
electromagnetism that have no interpretation to destroy Fermi liquids, which, by their nature, 1 spacetime that is asymptotically AdS. The
in terms of doped holes. That is, they are not permit a purely local description in momen- radial direction represents the energy with the
local entities, as they arise from the mixing tum space. As a result, Fermi liquids are im- quantum theory residing at the UV boundary,
of UV and IR degrees of freedom. It is such pervious to generic repulsive local-in-space whereas the IR limit is deep in the interior at
mixing that could account for the lack of any interactions (134). Coupling a Fermi liquid to the black hole horizon. Hence, intrinsic to
distinctive energy scale (102)—that is, scale an array of disordered SYK islands, however, this construction is a separation between bulk
invariance—underlying the strange metal. leads (120, 121) to a nontrivial change in the (gravitational) and boundary (quantum me-
Additionally, optical conductivity studies showed electron Green function across the MIR, and chanical) degrees of freedom. That the boundary
(104) that throughout the underdoped regime of hence a change in slope of the resistivity is of a gravitational object has features distinct
the cuprate phase diagram, the effective mass unavoidable (121), although it can be mini- from the bulk dates back to the observations
remains constant. As a result, the Mott transition mized through fine tuning (120). of Beckenstein (141) and Hawking (142, 143)
proceeds by a vanishing of the carrier number An added feature of these disordered mod- that the information content of a black hole
rather than the mass divergence of the Brinkman- els is that in certain limits, they have a gravity scales with the area, not the volume. The re-
Rice scenario (105). [Note that although quan- dual (132, 133, 135, 136). This state of affairs quirement that the boundary theory be strongly
tum oscillation experiments on underdoped arises because the basic propagator (132, 133, 135) coupled then arises by maintaining that the
cuprates show evidence for mass enhancement in the SYK model in imaginary time describes AdS radius of curvature exceeds the Planck
(106), this is thought to be tied to the charge the motion of fermions, with appropriate bound- length ‘P. More explicitly, because the AdS ra-
order around 1/8 doping.] Such dynamical ary conditions, between two points of the as- dius and the coupling constant of the bound-
mixing between the UV and IR scales in Mott ymptotic boundary of a hyperbolic plane. In ary theory are proportional, the requirement
systems is well known to give rise to spectral real time, simply replacing the hyperbolic plane R >> ‘P translates into a boundary theory that
weight in the lower Hubbard band (100–102) with the spacetime equivalent, namely two- is strongly coupled.
that exceeds the number of electrons, strictly dimensional anti–de Sitter (AdS) space (a max- The first incarnation (122, 144, 145) of this
1 + x, that the band can hold. Consequently, imally symmetric Lorentzian manifold with duality in the context of fermion correlators
part of the electromagnetic response of strange constant negative curvature), accurately de- involved modeling fermions at finite density
metals at low energies has no interpretation scribes all the correlators. It is from this re- in 2 + 1 dimensions. From the duality, the con-
in terms of electron quasiparticles, as it arises alization that the dual description between a formally invariant vacuum of such a system
strictly from UV-IR mixing. Precisely how such random spin model and gravity in AdS2 lies corresponds to gravity in AdS4, the extra di-
mixing leads to scale-invariant T-linear resis- (132, 133, 137). Hence, although the origins of mension representing the radial direction
tivity remains unresolved. SYK were independent of gravity, its correla- along which identical copies of the boundary
tors can be deduced from the asymptotics of CFT lie (albeit with differing energy scales).
Is it about gravity? the corresponding spacetime. At the asymptote, Surprisingly, what was shown (122) is that
To frame the theoretical modeling of strange only the time coordinate survives and hence, the low-energy (IR) properties of such a sys-
metallicity, we group the work into three prin- ultimately, SYK dynamics is ultralocal in space tem in the presence of a charge density are
cipal categories: phenomenological, numerical, with correlations diverging only in time, an determined by an emergent AdS2 × R2 (with
and gravity-related. Table 2 lists some repre- instantiation of local quantum criticality. R2 representing a plane) spacetime at the black
sentative contributions (67, 90, 107–130); be- Such local quantum criticality is not a new hole horizon. The actual symmetry includes
cause of space limitations, it is not possible to concept in condensed matter systems and scale invariance and is denoted by SL(2, R) (a
cite all relevant work. Both phenomenological indeed lies at the heart of MFL phenome- special Lie group of real 2 × 2 matrices with a
models considered here require (such as the nology (67) as well as dynamical mean field unit determinant). Once again, the criticality
ersatz Fermi liquid or EFL) or predict (MFL) theory (DMFT) (116), and is consistent with of boundary fermions is determined entirely
loop currents, but they do so for fundamentally the momentum-independent continuum found by the fluctuations in time, that is, local quan-
different reasons. For EFL (108), such current in the M-EELS data discussed earlier (85). The tum criticality as seen in SYK. The temperature
order is needed to obtain a finite resistivity in deeper question is why gravity has anything and frequency dependence of the conductivity
the absence of momentum relaxation (certainly to do with a spin problem with nonlocal in- are then determined by the same exponent
not a natural choice given the Drude fit to teractions. The issue comes down to critical- (122) as expected from Eqs. 5 and 6 and, as a
the optical conductivity discussed previously), ity and to the structure of general relativity. result, a simultaneous description of T-linearity
whereas in MFL, loop currents (107) are thought The second equivalence principle on which and w–2/3 dependence is not possible, as noted
to underpin the local fluctuation spectrum (67). general relativity is based states that no local in Table 2.
The extremely correlated Fermi liquid theory measurement can detect a uniform gravita- This particular hurdle is overcome by the
(ECFL) (109) predicts a resistivity that interpo- tional field. A global measurement is required. anomalousdimensions/Einstein-Maxwell-dilaton
lates between Fermi liquid–like T2 at low T to The same is true for a critical system because (AD/EMD) theories (89, 90, 126, 146–149), which,
T-linear for T >> TFL. Quantum Monte Carlo no local measurement can discern criticality. as indicated in Table 2, have been the most
(QMC) simulations (110, 111, 113, 114) as well as Observables tied to the diverging correlation successful to date in describing the range of
cold atom experiments (112) on the Hubbard length are required. Hence, at least concep- physics observed in strange metals. What is
model have established that at high temper- tually, it is not unreasonable to expect a link new here is the introduction of extra fields,
atures, the resistivity is indeed T-linear. The between critical matter and gravity. The mod- dilatons for example, that permit hyperscal-
fermion-sign problem, however, prevents any ern mathematical machinery that makes it ing violation (79) and anomalous dimensions
definitive statement about the low-T behavior possible to relate the two is the gauge-gravity (89, 90, 126, 146–149) for all operators. Con-
in the presence of Mott physics. Non–Fermi duality or the AdS/CFT (anti–de Sitter/conformal sequently, under a scale change of the coordi-
liquid transport in Sachdev-Ye-Kitaev (SYK)– field theory) conjecture. The key claim of this nates, the metric is no longer unscathed. That
based models (131–133) is achieved by an duality (138–140) is that some strongly inter- is, the manifold is not fixed and it is the mat-
all-to-all random interaction. Although such acting quantum theories, namely ones that ter fields that determine the geometry. Such

systems have a covariance, rather than scale ing. To illustrate her first theorem, consider anomalous dimension for the underlying gauge
invariance indicative of pure AdS metrics. A Maxwellian electromagnetism, which is invar- field, offers a viable solution, but the price is
consequence of this covariance is that even iant under the transformation Am → Am + @ mL. abandoning locality (164) of the action.
the currents acquire anomalous dimensions. This theorem states that there must be a
But how is this possible given that a tenet of conservation law with the same number of Is it important?
field theory is that no amount of renormal- derivatives as in the gauge principle. Hence, Given the immense difficulty in constructing a
ization can change the dimension of the current the conservation law only involves a single theory of strange metals, one might ask why
(150) from d – 1? What makes this possible is derivative, namely @ mJm = 0. This is Nöther’s bother? To gauge the importance of strange
that in EMD theories, the extra radial di- First Theorem (160) in practice. metals, look no further than Fig. 4. This figure
mension allows physics beyond Maxwellian What Nöther (160) spent the second half shows that the coefficient a1 of the T-linear
electromagnetism. For example, the standard of her famous paper trying to rectify is that resistivity component in the strange metal
Maxwell action, S = ∫dVdF 2 where F = dA, re- the form of the gauge transformation is not regime of overdoped hole-doped cuprates tracks
quires that the dimension of the gauge field be unique; hence, the conservation law is arbi- the doping dependence of the T = 0 superfluid
fixed to unity, [A] = 1 (151). EMD theories use trary. It is for this reason that in the second density ns(0). As mentioned earlier, a similar
instead an action of the form S = ∫dVddyyaF 2 half (160) of her foundational paper, she re- correlation exists between a1 and Tc in electron-
where y is the radial coordinate of the d + tained all possible higher-order integer deriv- doped cuprates (31, 32), the Bechgaard salts
1 AdS spacetime. Comparing these two actions atives in the gauge principle. These higher- (58), and the iron pnictides (58), establishing a
leads immediately to the conclusion that the order derivatives both add constraints to and fundamental link between high-temperature
dimension of A now acquires the value [A] = change the dimension of the current. Stated superconductivity and strange metals.
1 – (a/2). Hence, even in the bulk of the ge- succinctly, NST (160) dictates that the full For a long time, the drop in ns(0) with doping
ometry, the dimension of the gauge field is family of generators of U(1) invariance deter- in cuprates was attributed to pair breaking, a
not unity. Depending on the value of a, a < 0 mines the dimension of the current. It is easy symptom of the demise of the dominant pair-
at the UV conformal boundary or a > 0 at the to see how this works. Suppose we can find a ing interaction within a disordered lattice.
IR at the black hole horizon, the equations of quantity Y^ that commutes with @ m. That is, Recent mutual inductance measurements,
motion are nonstandard and obey fractional @m Y^ ¼ Y^ @m . If this is so, then we can insert however, have challenged this view, arguing
electromagnetism (128, 152) consistent with a this into the conservation law with impu- that the limiting low-T behavior of ns(T) was
nontraditional dimension for the gauge field. nity. What this does is redefine the current: incompatible with conventional pair-breaking
@m Y^ J m ¼ @m J~ . The new current J~ acquires
m m
In EMD theories, it is precisely the anomalous scenarios (165). Certainly, the correlation be-
dimension (89, 90, 126, 146–148) for conserved whatever dimensions Y has, such that J~m ¼
^ tween a1 and ns(0) is unforeseen in such mod-
quantities that gives rise to the added freedom d 1 dY . But because of the first theorem, Y^ els. Moreover, if the strange metal regime is
for extended quantum criticality to occur, en- must have come from the gauge transforma- indeed populated with non-quasiparticle states,
abling the simultaneous fitting (130) of T- tion and hence must ultimately be a differen- then Fig. 4 indicates a pivotal role for these
linearity and w–2/3 of the optical conductivity, tial operator itself. That is, there is an equally states in the pairing condensate (166, 167). On
and establishing the basis for a proposal for valid class of electromagnetisms with gauge more general grounds, this result informs us
the strange metal based on [A] = 5/3 (127). transformations of the form Am →Am þ @m Y^ L. that the door to unlocking cuprate supercon-
Within these holographic systems, a Drude- For EMD theories (102, 128, 152), Y^ is given by ductivity is through the strange metal regime,
like peak in the optical conductivity can emerge the fractional Laplacian, D(g–1)/2 with [Am] = g and any theory that divorces superconductiv-
both from the coherent (quasiparticle-like) [with g = 1 – (a/2) to make contact with the ity from strange metals is unlikely to be a prom-
sector (153) as well as from the incoherent [“un- EMD theories introduced earlier]. For most ising avenue. To conclude, solving the strange
particle” (154)] sector (155–158). Application of matter as we know it, g = 1. The success of metal regime kills two birds with one stone.
EMD theory has also provided fresh insights EMD theories raises the possibility that the Perhaps there is some justice here. After all, we
into the phenomenon of “lifetime separa- strangeness of the strange metal hinges on know from Pippard’s (168) work, which can be
tion” seen in the DC and Hall conductivities the fact that g ≠ 1. This can be tested exper- reformulated (128, 152) in terms of fractional
of hole-doped cuprates (22, 24, 28) as well as imentally using the standard Aharonov-Bohm
in other candidate strange metals (52, 61). geometry (128, 129) in which a hole of radius r
For a system with broken translational invar- is punched into a cuprate strange metal. Be-
iance, the finite density conductivity com- cause [A] is no longer unity, the integral of
prises two distinct components (159). The DC A·d‘ is no longer the dimensionless flux. For
resistivity is dominated by the particle-hole physically realizable gauges, this ultimately
symmetric term—whose Hall conductivity is provides an obstruction to charge quantiza-
consequently zero—whereas the T dependence tion. As a result, deviations (128, 129) from
of the Hall angle is set by the more conven- the standard pr2 × B dependence for the flux
tional (umklapp) momentum relaxation that would be the key experimental feature that a
governs the response of the coherent charge nonlocal gauge principle is operative in the
density. strange metal. An alternative would be, as
The success of EMD theories in the context Anderson (161) advocated, the use of fractional
of strange metal physics raises a philosophi- or unparticle propagators with the standard
cal question: Is all of this just a game? That gauge principle. However, in the end, it all
is, is the construction of bulk theories with comes down to gauge invariance. The stan-
funky electromagnetism fundamental? The dard gauge-invariant condition prevents the Fig. 4. Correlation between the superfluid den-
answer lies in Nöther’s Second Theorem (NST) power laws in unparticle stuff from influenc- sity ns(0) and the coefficient a1 of the T-linear
(102, 128, 152), a theorem far less known than ing the algebraic fall-off of the optical conduc- resistivity. The data shown are for Tl2Ba2CuO6+d
her ubiquitous first theorem but ultimately of tivity (130, 162), as they offer just a prefactor to (Tl2201) across the strange metal regime. [Adapted
more importance as it identifies a shortcom- the polarizations (163). The escape route, an from (166, 167)]

Laplacians, that explaining superconductivity 22. A. Carrington, A. P. Mackenzie, C. T. Lin, J. R. Cooper, 45. J. G. Analytis et al., Transport near a quantum critical point in
even in elemental metals necessitates a non- Temperature dependence of the Hall angle in single-crystal BaFe2(As1–xPx)2. Nat. Phys. 10, 194–197 (2014). doi: 10.1038/
YBa2(Cu1-xCox)3O7-d. Phys. Rev. Lett. 69, 2855–2858 (1992). nphys2869
local relationship between the current and the doi: 10.1103/PhysRevLett.69.2855; pmid: 10046606 46. S. Kasahara et al., Evolution from non-Fermi- to Fermi-liquid
gauge field. What seems to be potentially new 23. M. K. Chan et al., In-plane magnetoresistance obeys Kohler’s transport via isovalent doping in BaFe2(As1–xPx)2
about the cuprates is that now the normal rule in the pseudogap phase of cuprate superconductors. superconductors. Phys. Rev. B 81, 184519 (2010).
Phys. Rev. Lett. 113, 177005 (2014). doi: 10.1103/ doi: 10.1103/PhysRevB.81.184519
state, as a result of the strange metallicity, also PhysRevLett.113.177005; pmid: 25379934 47. I. M. Hayes et al., Scaling between magnetic field and
requires nonlocality. 24. T. R. Chien, Z. Z. Wang, N. P. Ong, Effect of Zn impurities on temperature in the high-temperature superconductor BaFe2
the normal-state Hall angle in single-crystal YBa2Cu3-xZnxO7-d. (As1–xPx)2. Nat. Phys. 12, 916–919 (2016). doi: 10.1038/
Phys. Rev. Lett. 67, 2088–2091 (1991). doi: 10.1103/ nphys3773
RE FE RENCES AND N OT ES
PhysRevLett.67.2088; pmid: 10044332 48. Y. Nakajima et al., Quantum-critical scale invariance in a
1. T. Matsubara, A New Approach to Quantum-Statistical 25. J. M. Harris et al., Violation of Kohler’s rule in the normal- transition metal alloy. Commun. Phys. 3, 181 (2020).
Mechanics. Prog. Theor. Phys. 14, 351–378 (1955). state magnetoresistance of YBa2Cu3O7-d and La2SrxCuO4. doi: 10.1038/s42005-020-00448-5
doi: 10.1143/PTP.14.351 Phys. Rev. Lett. 75, 1391–1394 (1995). doi: 10.1103/ 49. O. Trovarelli et al., YbRh2Si2: Pronounced non-Fermi-liquid
2. S. Chakravarty, B. I. Halperin, D. R. Nelson, Two-dimensional PhysRevLett.75.1391; pmid: 10060281 effects above a low-lying magnetic phase transition.
quantum Heisenberg antiferromagnet at low temperatures. 26. P. Giraldo-Gallo et al., Scale-invariant magnetoresistance in a Phys. Rev. Lett. 85, 626–629 (2000). doi: 10.1103/
Phys. Rev. B 39, 2344–2371 (1989). doi: 10.1103/ cuprate superconductor. Science 361, 479–481 (2018). PhysRevLett.85.626; pmid: 10991356
PhysRevB.39.2344; pmid: 9948474 doi: 10.1126/science.aan3178; pmid: 30072535 50. J. Custers et al., The break-up of heavy electrons at a
3. J. Zaanen, Superconductivity: Why the temperature is high. 27. C. Boyd, P. W. Phillips, Single-parameter scaling in the quantum critical point. Nature 424, 524–527 (2003).
Nature 430, 512–513 (2004). doi: 10.1038/430512a; magnetoresistance of optimally doped La2-xSrxCuO4. doi: 10.1038/nature01774; pmid: 12891349
pmid: 15282588 Phys. Rev. B 100, 155139 (2019). doi: 10.1103/ 51. J. Custers et al., Evidence for a non-Fermi-liquid phase in Ge-
4. M. Gurvitch, A. T. Fiory, Resistivity of La1.825Sr0.175CuO4 and PhysRevB.100.155139 substituted YbRh2Si2. Phys. Rev. Lett. 104, 186402 (2010).
YBa2Cu3O to 1100 K: Absence of saturation and its 28. T. Manako, Y. Kubo, Y. Shimakawa, Transport and structural doi: 10.1103/PhysRevLett.104.186402; pmid: 20482193
implications. Phys. Rev. Lett. 59, 1337–1340 (1987). study of Tl2Ba2CuO6+d single crystals prepared by the KCl 52. S. Paschen et al., Hall-effect evolution across a heavy-fermion
doi: 10.1103/PhysRevLett.59.1337; pmid: 10035206 flux method. Phys. Rev. B 46, 11019–11024 (1992). quantum critical point. Nature 432, 881–885 (2004).
5. S. Martin, A. T. Fiory, R. M. Fleming, L. F. Schneemeyer, doi: 10.1103/PhysRevB.46.11019; pmid: 10002965 doi: 10.1038/nature03129; pmid: 15602556
J. V. Waszczak, Normal-state transport properties of Bi2+xSr2-y 29. J. Ayres et al., Incoherent transport across the strange- 53. T. Tomita, K. Kuga, Y. Uwatoko, P. Coleman, S. Nakatsuji, Strange
CuO6+d crystals. Phys. Rev. B 41, 846–849 (1990). metal regime of overdoped cuprates. Nature 595, metal without magnetic criticality. Science 349, 506–509
doi: 10.1103/PhysRevB.41.846; pmid: 9992839 661–666 (2021). doi: 10.1038/s41586-021-03622-z; (2015). doi: 10.1126/science.1262054; pmid: 26228141
6. H. Takagi et al., Systematic evolution of temperature- pmid: 34321672 54. Y. Nakajima et al., Non-Fermi Liquid Behavior in the
dependent resistivity in La2-xSrxCuO4. Phys. Rev. Lett. 69, 30. N. R. Poniatowski, T. Sarkar, S. Das Sarma, R. L. Greene, Magnetotransport of CeMIn5 (M: Co and Rh): Striking
2975–2978 (1992). doi: 10.1103/PhysRevLett.69.2975; Resistivity saturation in an electron-doped cuprate. Phys. Rev. Similarity between Quasi Two-Dimensional Heavy Fermion
pmid: 10046689 B 103, L020501 (2021). doi: 10.1103/PhysRevB.103.L020501 and High-Tc Cuprates. J. Phys. Soc. Jpn. 76, 024703 (2007).
7. D. van der Marel et al., Quantum critical behaviour in a 31. K. Jin, N. P. Butch, K. Kirshenbaum, J. Paglione, R. L. Greene, doi: 10.1143/JPSJ.76.024703
high-Tc superconductor. Nature 425, 271–274 (2003). Link between spin fluctuations and electron pairing in 55. A. Bianchi, R. Movshovich, I. Vekhter, P. G. Pagliuso,
doi: 10.1038/nature01978; pmid: 13679910 copper oxide superconductors. Nature 476, 73–75 (2011). J. L. Sarrao, Avoided antiferromagnetic order and quantum
8. R. A. Cooper et al., Anomalous criticality in the electrical doi: 10.1038/nature10308; pmid: 21814279 critical point in CeCoIn5. Phys. Rev. Lett. 91, 257001 (2003).
resistivity of La2-xSrxCuO4. Science 323, 603–607 (2009). 32. J. Yuan et al., Scaling of the strange-metal scattering in doi: 10.1103/PhysRevLett.91.257001; pmid: 14754138
doi: 10.1126/science.1165015; pmid: 19074310 unconventional superconductors. Nature 602, 431–436 56. J. Paglione et al., Field-induced quantum critical point in
9. A. Legros et al., Universal T-linear resistivity and Planckian (2022). doi: 10.1038/s41586-021-04305-5; pmid: 35173341 CeCoIn5. Phys. Rev. Lett. 91, 246405 (2003). doi: 10.1103/
dissipation in overdoped cuprates. Nat. Phys. 15, 142–147 33. P. Li, F. F. Balakirev, R. L. Greene, High-field Hall resistivity PhysRevLett.91.246405; pmid: 14683139
(2019). doi: 10.1038/s41567-018-0334-2 and magnetoresistance of electron-doped Pr2-xCexCuO4-d. 57. B. Shen et al., Strange-metal behaviour in a pure
10. J. A. N. Bruin, H. Sakai, R. S. Perry, A. P. Mackenzie, Phys. Rev. Lett. 99, 047003 (2007). doi: 10.1103/ ferromagnetic Kondo lattice. Nature 579, 51–55 (2020).
Similarity of scattering rates in metals showing T-linear PhysRevLett.99.047003; pmid: 17678392 doi: 10.1038/s41586-020-2052-z; pmid: 32132691
resistivity. Science 339, 804–807 (2013). doi: 10.1126/ 34. N. R. Poniatowski, T. Sarkar, R. L. Greene, Anomalous 58. N. Doiron-Leyraud et al., Correlation between linear resistivity
science.1227612; pmid: 23413351 normal-state magnetotransport in an electron-doped cuprate. and Tc in the Bechgaard salts and the pnictide
11. P. Fournier et al., Insulator-Metal Crossover near Optimal Phys. Rev. B 103, 125102 (2021). doi: 10.1103/ superconductor Ba(Fe1–xCox)2As2. Phys. Rev. B 80, 214531
Doping in Pr2–xCexCuO4: Anomalous Normal-State Low PhysRevB.103.125102 (2009). doi: 10.1103/PhysRevB.80.214531
Temperature Resistivity. Phys. Rev. Lett. 81, 4720–4723 35. T. Sarkar, P. R. Mandal, N. R. Poniatowski, M. K. Chan, 59. H. Polshyn et al., Large linear-in-temperature resistivity in
(1998). doi: 10.1103/PhysRevLett.81.4720 R. L. Greene, Correlation between scale-invariant normal- twisted bilayer graphene. Nat. Phys. 15, 1011–1016 (2019).
12. A. P. Mackenzie, S. R. Julian, D. C. Sinclair, C. T. Lin, Normal- state resistivity and superconductivity in an electron-doped doi: 10.1038/s41567-019-0596-3
state magnetotransport in superconducting Tl2Ba2CuO6+d to cuprate. Sci. Adv. 5, eaav6753 (2019). doi: 10.1126/ 60. Y. Cao et al., Strange Metal in Magic-Angle Graphene with
millikelvin temperatures. Phys. Rev. B 53, 5848–5855 (1996). sciadv.aav6753; pmid: 31114800 near Planckian Dissipation. Phys. Rev. Lett. 124, 076801
doi: 10.1103/PhysRevB.53.5848; pmid: 9984193 36. A. W. Tyler, A. P. Mackenzie, S. NishiZaki, Y. Maeno, High- (2020). doi: 10.1103/PhysRevLett.124.076801
13. N. Nagaosa, P. A. Lee, Ginzburg-Landau theory of the spin- temperature resistivity of Sr2RuO4: Bad metallic transport in 61. R. Lyu et al., Strange metal behavior of the Hall angle in
charge-separated system. Phys. Rev. B 45, 960 (1992). a good metal. Phys. Rev. B 58, 10107 (1998). doi: 10.1103/ twisted bilayer graphene. Phys. Rev. B 103, 245424 (2021).
doi: 10.1103/PhysRevB.45.966 PhysRevB.58.R10107 doi: 10.1103/PhysRevB.103.245424
14. A. F. Ioffe, A. R. Regel, Non-crystalline, amorphous, and liquid 37. N. E. Hussey et al., Normal-state magnetoresistance of 62. P. B. Allen et al., Transport properties, thermodynamic
electronic semiconductors. Prog. Semicond. 4, 237 (1960). Sr2RuO4. Phys. Rev. B 57, 5505–5511 (1998). doi: 10.1103/ properties, and electronic structure of SrRuO3. Phys. Rev. B
15. M. Gurvitch, Ioffe-Regel criterion and resistivity of metals. PhysRevB.57.5505 53, 4393–4398 (1996). doi: 10.1103/PhysRevB.53.4393;
Phys. Rev. B 24, 7404–7407 (1981). doi: 10.1103/ 38. M. E. Barber, A. S. Gibbs, Y. Maeno, A. P. Mackenzie, C. W. Hicks, pmid: 9983992
PhysRevB.24.7404 Resistivity in the Vicinity of a van Hove Singularity: Sr2RuO4 63. A. P. Mackenzie et al., Observation of quantum oscillations in
16. N. E. Hussey, K. Takenaka, H. Takagi, Universality of the under Uniaxial Pressure. Phys. Rev. Lett. 120, 076602 (2018). the electrical resistivity of SrRuO3. Phys. Rev. B 58, 13318
Mott-Ioffe-Regel limit in metals. Philos. Mag. 84, 2847–2864 doi: 10.1103/PhysRevLett.120.076602; pmid: 29542933 (1998). doi: 10.1103/PhysRevB.58.R13318
(2004). doi: 10.1080/14786430410001716944 39. A. P. Mackenzie et al., Hall effect in the two-dimensional 64. O. Gunnarsson, M. Calandra, J. E. Han, Saturation of
17. N. F. Mott, Conduction in non-crystalline systems IX. the metal Sr2RuO4. Phys. Rev. B 54, 7425–7429 (1996). electrical resistivity. Rev. Mod. Phys. 75, 1085–1099 (2003).
minimum metallic conductivity. Philos. Mag. 26, 1015–1026 doi: 10.1103/PhysRevB.54.7425; pmid: 9984367 doi: 10.1103/RevModPhys.75.1085
(1972). doi: 10.1080/14786437208226973 40. S. Kasahara et al., Field-induced superconducting phase of 65. N. E. Hussey, J. Buhot, S. Licciardello, A tale of two metals:
18. N. E. Hussey et al., Dichotomy in the T-linear resistivity in FeSe in the BCS-BEC cross-over. Proc. Natl. Acad. Sci. U.S.A. Contrasting criticalities in the pnictides and hole-doped
hole-doped cuprates. Philos. Trans. R. Soc. A 369, 1626–1639 111, 16309–16313 (2014). doi: 10.1073/pnas.1413477111; cuprates. Rep. Prog. Phys. 81, 052501 (2018). doi: 10.1088/
(2011). doi: 10.1098/rsta.2010.0196; pmid: 21422018 pmid: 25378706 1361-6633/aaa97c; pmid: 29353812
19. V. J. Emery, S. A. Kivelson, Superconductivity in bad metals. 41. S. Licciardello et al., Electrical resistivity across a nematic 66. K. Damle, S. Sachdev, Nonzero-temperature transport near
Phys. Rev. Lett. 74, 3253–3256 (1995). doi: 10.1103/ quantum critical point. Nature 567, 213–217 (2019). quantum critical points. Phys. Rev. B 56, 8714–8733 (1997).
PhysRevLett.74.3253; pmid: 10058150 doi: 10.1038/s41586-019-0923-y; pmid: 30760921 doi: 10.1103/PhysRevB.56.8714
20. C. Proust, B. Vignolle, J. Levallois, S. Adachi, N. E. Hussey, 42. W. K. Huang et al., Non-Fermi liquid transport in the vicinity 67. C. M. Varma, P. B. Littlewood, S. Schmitt-Rink, E. Abrahams,
Fermi liquid behavior of the in-plane resistivity in the of the nematic quantum critical point of superconducting A. E. Ruckenstein, Phenomenology of the normal state
pseudogap state of YBa2Cu4O8. Proc. Natl. Acad. Sci. U.S.A. FeSe1–xSx. Phys. Rev. Res. 2, 033367 (2020). doi: 10.1103/ of Cu-O high-temperature superconductors. Phys. Rev. Lett.
113, 13654–13659 (2016). doi: 10.1073/pnas.1602709113; PhysRevResearch.2.033367 63, 1996–1999 (1989). doi: 10.1103/PhysRevLett.63.1996;
pmid: 27856753 43. S. Licciardello et al., Coexistence of orbital and quantum pmid: 10040734
21. N. Barišic et al., Universal sheet resistance and revised phase critical magnetoresistance in FeSe1–xSx. Phys. Rev. Res. 1, 68. Hv. Löhneysen et al., Non-Fermi-liquid behavior in a heavy-
diagram of the cuprate high-temperature superconductors. 023011 (2019). doi: 10.1103/PhysRevResearch.1.023011 fermion alloy at a magnetic instability. Phys. Rev. Lett. 72,
Proc. Natl. Acad. Sci. U.S.A. 110, 12235–12240 (2013). 44. D. Hu et al., w/T scaling and magnetic quantum criticality in 3262–3265 (1994). doi: 10.1103/PhysRevLett.72.3262;
doi: 10.1073/pnas.1301989110; pmid: 23836669 BaFe2(As0.7P0.3)2. arXiv 1812.11902 (2018). pmid: 10056148

69. B. Michon et al., Thermodynamic signatures of quantum Bi2Sr2CaCu2O8+d from high-resolution laser angle-resolved 086401 (2013). doi: 10.1103/PhysRevLett.110.086401;
criticality in cuprate superconductors. Nature 567, 218–222 photoemission. Phys. Rev. B 81, 174516 (2010). doi: 10.1103/ pmid: 23473178
(2019). doi: 10.1038/s41586-019-0932-x; pmid: 30760922 PhysRevB.81.174516 117. W. Wú, X. Wang, A. M. S. Tremblay, Non-Fermi liquid phase
70. J. G. Storey, J. L. Tallon, G. V. M. Williams, Pseudogap ground 94. A. Damascelli, Z. Hussain, Z.-X. Shen, Angle-resolved and linear-in-temperature scattering rate in overdoped
state in high-temperature superconductors. Phys. Rev. B 78, photoemission studies of the cuprate superconductors. two-dimensional Hubbard model. Proc. Natl. Acad. Sci. U.S.A.
140506 (2008). doi: 10.1103/PhysRevB.78.140506 Rev. Mod. Phys. 75, 473–541 (2003). doi: 10.1103/ 119, e2115819119 (2022). doi: 10.1073/pnas.2115819119;
71. S.-D. Chen et al., Incoherent strange metal sharply bounded RevModPhys.75.473 pmid: 35320041
by a critical doping in Bi2212. Science 366, 1099–1102 95. T. Yoshida et al., Low-energy electronic structure of the high- 118. S. Lederer, Y. Schattner, E. Berg, S. A. Kivelson,
(2019). doi: 10.1126/science.aaw8850; pmid: 31780552 Tc cuprates La2–xSrxCuO4 studied by angle-resolved Superconductivity and non-Fermi liquid behavior near a
72. N. E. Hussey, H. Gordon-Moys, J. Kokalj, R. H. McKenzie, photoemission spectroscopy. J. Phys. 19, 125209 (2007). nematic quantum critical point. Proc. Natl. Acad. Sci. U.S.A.
Generic strange-metal behaviour of overdoped cuprates. doi: 10.1088/0953-8984/19/12/125209 114, 4905–4910 (2017). doi: 10.1073/pnas.1620651114;
J. Phys. Conf. Ser. 449, 012004 (2013). doi: 10.1088/ 96. N. Poniatowski, T. Sarkar, R. Lobo, S. Das Sarma, pmid: 28439023
1742-6596/449/1/012004 R. L. Greene, Counterexample to the conjectured Planckian 119. A. A. Patel, S. Sachdev, Theory of a Planckian Metal.
73. P. Phillips, C. Chamon, Breakdown of one-parameter scaling bound on transport. Phys. Rev. B 104, 235138 (2021). Phys. Rev. Lett. 123, 066601 (2019). doi: 10.1103/
in quantum critical scenarios for high-temperature copper-oxide doi: 10.1103/PhysRevB.104.235138 PhysRevLett.123.066601; pmid: 31491164
superconductors. Phys. Rev. Lett. 95, 107002 (2005). 97. S. Hartnoll, A. P. Mackenzie, Planckian Dissipation in Metals. 120. P. Cha, A. A. Patel, E. Gull, E.-A. Kim, Slope invariant T-linear
doi: 10.1103/PhysRevLett.95.107002; pmid: 16196953 arXiv 2107.07802 (2021). resistivity from local self-energy. Phys. Rev. Res. 2, 033434
74. X.-G. Wen, Scaling theory of conserved current and universal 98. Z. Nussinov, S. Chakrabarty, Exact Universal Chaos, Speed (2020). doi: 10.1103/PhysRevResearch.2.033434
amplitudes at anisotropic critical points. Phys. Rev. B 46, Limit, Acceleration, Planckian Transport Coefficient, 121. A. A. Patel, J. McGreevy, D. P. Arovas, S. Sachdev,
2655–2662 (1992). doi: 10.1103/PhysRevB.46.2655; “Collapse” to equilibrium, and Other Bounds in Thermal Magnetotransport in a Model of a Disordered Strange
pmid: 10003951 Quantum Systems. Ann. Phys. 10.1016/j.aop.2022.168970 Metal. Phys. Rev. X 8, 021049 (2018). doi: 10.1103/
75. L. Prochaska et al., Singular charge fluctuations at a (2022). doi: 10.1016/j.aop.2022.168970 PhysRevX.8.021049
magnetic quantum critical point. Science 367, 285–288 99. C. T. Chen et al., Electronic states in La2-xSrxCuO4+d probed 122. T. Faulkner, N. Iqbal, H. Liu, J. McGreevy, D. Vegh, Strange
(2020). doi: 10.1126/science.aag1595; pmid: 31949077 by soft-x-ray absorption. Phys. Rev. Lett. 66, 104–107 (1991). metal transport realized by gauge/gravity duality. Science
76. H. J. A. Molegraaf, C. Presura, D. van Der Marel, P. H. Kes, doi: 10.1103/PhysRevLett.66.104; pmid: 10043153 329, 1043–1047 (2010). doi: 10.1126/science.1189134;
M. Li, Superconductivity-induced transfer of in-plane 100. M. B. J. Meinders, H. Eskes, G. A. Sawatzky, Spectral-weight pmid: 20688983
spectral weight in Bi2Sr2CaCu2O8+d. Science 295, 2239–2241 transfer: Breakdown of low-energy-scale sum rules in 123. G. T. Horowitz, J. E. Santos, D. Tong, Optical conductivity
(2002). doi: 10.1126/science.1069947; pmid: 11910103 correlated systems. Phys. Rev. B 48, 3916–3926 (1993). with holographic lattices. J. High Energy Phys. 2012, 168
77. S. L. Cooper et al., Growth of the optical conductivity in the doi: 10.1103/PhysRevB.48.3916; pmid: 10008840 (2012). doi: 10.1007/JHEP07(2012)168
Cu-O planes. Phys. Rev. B 41, 11605–11608 (1990). 101. H. Eskes, A. M. Oleś, M. B. Meinders, W. Stephan, Spectral 124. B. W. Langley, G. Vanacore, P. W. Phillips, Absence of
doi: 10.1103/PhysRevB.41.11605; pmid: 9993591 properties of the Hubbard bands. Phys. Rev. B 50, power-law mid-infrared conductivity in gravitational crystals.
78. M. A. Quijada et al., Anisotropy in the ab-plane optical 17980–18002 (1994). doi: 10.1103/PhysRevB.50.17980; J. High Energy Phys. 2015, 163 (2015). doi: 10.1007/
properties of Bi2Sr2CaCu2O8 single-domain crystals. pmid: 9976228 JHEP10(2015)163
Phys. Rev. B 60, 14917–14934 (1999). doi: 10.1103/ 102. P. W. Phillips, Identifying the propagating charge modes in 125. A. Donos, J. P. Gauntlett, Holographic Q-lattices. J. High Energy
PhysRevB.60.14917 doped Mott insulators. Rev. Mod. Phys. 82, 1719–1742 Phys. 2014, 40 (2014). doi: 10.1007/JHEP04(2014)040
79. S. A. Hartnoll, A. Lucas, S. Sachdev, Holographic Quantum (2010). doi: 10.1103/RevModPhys.82.1719 126. E. Kiritsis, Y. Matsuo, Hyperscaling-violating Lifshitz
Matter (MIT Press, 2018). 103. A. J. Leggett, A “midinfrared” scenario for cuprate hydrodynamics from black-holes: Part II. J. High Energy Phys.
80. J. Loram, K. Mirza, J. Wade, J. Cooper, W. Liang, The superconductivity. Proc. Natl. Acad. Sci. U.S.A. 96, 8365–8372 2017, 41 (2017). doi: 10.1007/JHEP03(2017)041
electronic specific heat of cuprate superconductors. (1999). doi: 10.1073/pnas.96.15.8365; pmid: 10411881 127. S. A. Hartnoll, A. Karch, Scaling theory of the cuprate strange
Physica 235C–240C, 134–137 (1994). doi: 10.1016/0921- 104. Y. S. Lee et al., Electrodynamics of the nodal metal state in metals. Phys. Rev. B 91, 155126 (2015). doi: 10.1103/
4534(94)91331-5 weakly doped high- Tc cuprates. Phys. Rev. B 72, 054529 PhysRevB.91.155126
81. J. Zaanen, Planckian dissipation, minimal viscosity and the (2005). doi: 10.1103/PhysRevB.72.054529 128. G. La Nave, K. Limtragool, P. W. Phillips, Fractional
transport in cuprate strange metals. SciPost Phys. 6, 61 105. W. F. Brinkman, T. M. Rice, Application of Gutzwiller’s electromagnetism in quantum matter and high-energy
(2019). doi: 10.21468/SciPostPhys.6.5.061 Variational Method to the Metal-Insulator Transition. Phys. physics. Rev. Mod. Phys. 91, 021003 (2019). doi: 10.1103/
82. N. Nücker et al., Plasmons and interband transitions in Rev. B 2, 4302–4304 (1970). doi: 10.1103/PhysRevB.2.4302 RevModPhys.91.021003
Bi2Sr2CaCu2O8. Phys. Rev. B 39, 12379–12382 (1989). 106. B. J. Ramshaw et al., Quasiparticle mass enhancement 129. K. Limtragool, P. W. Phillips, Anomalous dimension of the
doi: 10.1103/PhysRevB.39.12379; pmid: 9948098 approaching optimal doping in a high-Tc superconductor. electrical current in strange metals from the fractional
83. N. Nücker, U. Eckern, J. Fink, P. Müller, Long-wavelength Science 348, 317–320 (2015). doi: 10.1126/science.aaa4990; Aharonov-Bohm effect. Europhys. Lett. 121, 27003 (2018).
collective excitations of charge carriers in high- pmid: 25814065 doi: 10.1209/0295-5075/121/27003
Tc superconductors. Phys. Rev. B 44, 7155(R) (1991). 107. M. E. Simon, C. M. Varma, Detection and implications of a 130. A. Karch, K. Limtragool, P. W. Phillips, Unparticles and
84. S. Vig et al., Measurement of the dynamic charge response of time-reversal breaking state in underdoped cuprates. anomalous dimensions in the cuprates. J. High Energy Phys.
materials using low-energy, momentum-resolved electron Phys. Rev. Lett. 89, 247003 (2002). doi: 10.1103/ 2016, 175 (2016). doi: 10.1007/JHEP03(2016)175
energy-loss spectroscopy (M-EELS). SciPost Phys. 3, 026 (2017). PhysRevLett.89.247003; pmid: 12484975 131. S. Sachdev, J. Ye, Gapless spin-fluid ground state in a
85. M. Mitrano et al., Anomalous density fluctuations in a strange 108. D. V. Else, T. Senthil, Strange Metals as Ersatz Fermi Liquids. random quantum Heisenberg magnet. Phys. Rev. Lett. 70,
metal. Proc. Natl. Acad. Sci. U.S.A. 115, 5392–5396 (2018). Phys. Rev. Lett. 127, 086601 (2021). doi: 10.1103/ 3339–3342 (1993). doi: 10.1103/PhysRevLett.70.3339;
doi: 10.1073/pnas.1721495115; pmid: 29735712 PhysRevLett.127.086601; pmid: 34477402 pmid: 10053843
86. A. A. Husain et al., Crossover of Charge Fluctuations across 109. B. S. Shastry, P. Mai, Aspects of the normal state resistivity 132. A. Kitaev, “A Simple Model of Quantum Holography” [video]
the Strange Metal Phase Diagram. Phys. Rev. X 9, 041062 of cuprate superconductors. Phys. Rev. B 101, 115121 (2020). (2015); https://online.kitp.ucsb.edu/online/entangled15/
(2019). doi: 10.1103/PhysRevX.9.041062 doi: 10.1103/PhysRevB.101.115121 kitaev/.
87. R. L. Greene, P. R. Mandal, N. R. Poniatowski, T. Sarkar, 110. E. W. Huang, R. Sheppard, B. Moritz, T. P. Devereaux, 133. A. Kitaev, S. J. Suh, The soft mode in the Sachdev-Ye-Kitaev
The Strange Metal State of the Electron-Doped Cuprates. Strange metallicity in the doped Hubbard model. Science model and its gravity dual. J. High Energy Phys. 2018, 183
Annu. Rev. Phys. 11, 213–229 (2020). doi: 10.1146/ 366, 987–990 (2019). doi: 10.1126/science.aau7063; (2018). doi: 10.1007/JHEP05(2018)183
annurev-conmatphys-031119-050558 pmid: 31753997 134. J. Polchinski, Effective Field Theory and the Fermi Surface
88. T. J. Reber et al., A unified form of low-energy nodal 111. J. Kokalj, Bad-metallic behavior of doped Mott insulators. [lectures]. arXiv 9210046v2 (1992).
electronic interactions in hole-doped cuprate Phys. Rev. B 95, 041110 (2017). doi: 10.1103/ 135. S. Sachdev, Holographic metals and the fractionalized Fermi
superconductors. Nat. Commun. 10, 5737 (2019). PhysRevB.95.041110 liquid. Phys. Rev. Lett. 105, 151602 (2010). doi: 10.1103/
doi: 10.1038/s41467-019-13497-4; pmid: 31844065 112. P. T. Brown et al., Bad metallic transport in a cold atom PhysRevLett.105.151602; pmid: 21230891
89. C. Charmousis, B. Goutéraux, B. Soo Kim, E. Kiritsis, Fermi-Hubbard system. Science 363, 379–382 (2019). 136. J. Maldacena, D. Stanford, Remarks on the Sachdev-Ye-
R. Meyer, Effective holographic theories for low-temperature doi: 10.1126/science.aat4134; pmid: 30523078 Kitaev model. Phys. Rev. D 94, 106002 (2016). doi: 10.1103/
condensed matter systems. J. High Energy Phys. 2010, 151 113. A. Vranić et al., Charge transport in the Hubbard model at PhysRevD.94.106002
(2010). doi: 10.1007/JHEP11(2010)151 high temperatures: Triangular versus square lattice. Phys. 137. S. Sachdev, Quantum Phase Transitions (Cambridge Univ.
90. B. Goutéraux, E. Kiritsis, Quantum critical lines in holographic Rev. B 102, 115142 (2020). doi: 10.1103/PhysRevB.102.115142 Press, ed. 2, 2011).
phases with (un)broken symmetry. J. High Energy Phys. 114. J. Vučičević et al., Conductivity in the Square Lattice Hubbard 138. J. Maldacena, The Large-N Limit of Superconformal Field
2013, 53 (2013). doi: 10.1007/JHEP04(2013)053 Model at High Temperatures: Importance of Vertex Theories and Supergravity. Int. J. Theor. Phys. 38, 1113–1133
91. T. Valla et al., Evidence for quantum critical behavior in the Corrections. Phys. Rev. Lett. 123, 036601 (2019). (1999). doi: 10.1023/A:1026654312961
optimally doped cuprate Bi2Sr2CaCu2O8+d. Science 285, doi: 10.1103/PhysRevLett.123.036601; pmid: 31386456 139. E. Witten, Anti de Sitter space and holography. Adv. Theor.
2110–2113 (1999). doi: 10.1126/science.285.5436.2110; 115. P. Cha, N. Wentzell, O. Parcollet, A. Georges, E.-A. Kim, Math. Phys. 2, 253–291 (1998). doi: 10.4310/
pmid: 10497125 Linear resistivity and Sachdev-Ye-Kitaev (SYK) spin liquid ATMP.1998.v2.n2.a2
92. A. Kaminski et al., Momentum anisotropy of the scattering behavior in a quantum critical metal with spin-1/2 140. S. Gubser, I. Klebanov, A. Polyakov, Gauge theory correlators
rate in cuprate superconductors. Phys. Rev. B 71, 014517 fermions. Proc. Natl. Acad. Sci. U.S.A. 117, 18341–18346 from non-critical string theory. Phys. Lett. B 428, 105–114
(2005). doi: 10.1103/PhysRevB.71.014517 (2020). doi: 10.1073/pnas.2003179117; pmid: 32699148 (1998). doi: 10.1016/S0370-2693(98)00377-3
93. J. M. Bok et al., Momentum dependence of the single-particle 116. X. Deng et al., How bad metals turn good: Spectroscopic 141. J. D. Bekenstein, Black Holes and Entropy. Phys. Rev. D 7,
self-energy and fluctuation spectrum of slightly underdoped signatures of resilient quasiparticles. Phys. Rev. Lett. 110, 2333–2346 (1973). doi: 10.1103/PhysRevD.7.2333

142. S. W. Hawking, Breakdown of predictability in gravitational 154. P. W. Phillips, B. W. Langley, J. A. Hutasoit, Un-Fermi entropy. Phys. Rev. D 101, 106006 (2020). doi: 10.1103/
collapse. Phys. Rev. D 14, 2460–2473 (1976). doi: 10.1103/ liquids: Unparticles in strongly correlated electron matter. PhysRevD.101.106006
PhysRevD.14.2460 Phys. Rev. B 88, 115129 (2013). doi: 10.1103/ 165. M. Čulo et al., Possible superconductivity from incoherent
143. S. W. Hawking, Particle creation by black holes. Commun. PhysRevB.88.115129 carriers in overdoped cuprates. SciPost Phys. 11, 012 (2021).
Math. Phys. 43, 199–220 (1975). doi: 10.1007/BF02345020 155. S. A. Hartnoll, J. Polchinski, E. Silverstein, D. Tong, Towards doi: doi:1468/SciPostPhys.11.1.012
144. S.-S. Lee, Non-Fermi liquid from a charged black hole: strange metallic holography. J. High Energy Phys. 2010, 120 166. C. Putzke et al., Reduced Hall carrier density in the overdoped
A critical Fermi ball. Phys. Rev. D 79, 086006 (2009). (2010). doi: 10.1007/JHEP04(2010)120 strange metal regime of cuprate superconductors. Nat. Phys. 17,
doi: 10.1103/PhysRevD.79.086006 156. E. Kiritsis, F. Peña Benitez, Scaling of the holographic 826–831 (2021). doi: 10.1038/s41567-021-01197-0
145. M. Cubrović, J. Zaanen, K. Schalm, String theory, quantum phase AC conductivity for non-Fermi liquids at criticality. 167. I. Božović, X. He, J. Wu, A. T. Bollinger, Dependence of the
transitions, and the emergent Fermi liquid. Science 325, J. High Energy Phys. 2015, 177 (2015). doi: 10.1007/ critical temperature in overdoped copper oxides on
439–444 (2009). doi: 10.1126/science.1174962; pmid: 19556462 JHEP11(2015)177 superfluid density. Nature 536, 309–311 (2016).
146. B. Goutéraux, Charge transport in holography with 157. C.-F. Chen, A. Lucas, Origin of the Drude peak and of zero doi: 10.1038/nature19061; pmid: 27535534
momentum dissipation. J. High Energy Phys. 2014, 181 sound in probe brane holography. Phys. Lett. B 774, 569–574 168. B. Pippard, An experimental and theoretical study of the
(2014). doi: 10.1007/JHEP04(2014)181 (2017). doi: 10.1016/j.physletb.2017.10.023 relation between magnetic field and current in a
147. A. Karch, Conductivities for hyperscaling violating 158. R. A. Davison, S. A. Gentle, B. Goutéraux, Slow Relaxation and superconductor. Proc. R. Soc. London Ser. A 216, 547–568
geometries. J. High Energy Phys. 2014, 140 (2014). Diffusion in Holographic Quantum Critical Phases. Phys. Rev. (1953). doi: 10.1098/rspa.1953.0040
doi: 10.1007/JHEP06(2014)140 Lett. 123, 141601 (2019). doi: 10.1103/PhysRevLett.123.141601;
148. S. Cremonini, A. Hoover, L. Li, Backreacted DBI magnetotransport pmid: 31702178 AC KNOWL ED GME NTS
with momentum dissipation. J. High Energy Phys. 2017, 133 159. M. Blake, A. Donos, Quantum critical transport and the Funding: Supported by the Center for Emergent Superconductivity,
(2017). doi: 10.1007/JHEP10(2017)133 Hall angle in holographic models. Phys. Rev. Lett. 114, a DOE Energy Frontier Research Center, grant DE-SC0021238
149. E. Blauvelt, S. Cremonini, A. Hoover, L. Li, S. Waskie, 021601 (2015). doi: 10.1103/PhysRevLett.114.021601; (P.A. and P.W.P.) and by Netherlands Organisation for
Holographic model for the anomalous scalings of the pmid: 25635540 Scientific Research (NWO) (“Strange Metals” grant 16METL01),
cuprates. Phys. Rev. D 97, 061901 (2018). doi: 10.1103/ 160. E. Noether, Invariante Variationsprobleme. Nachr. Ges. Wiss. and the European Research Council (ERC) under the European
PhysRevD.97.061901 Gottingen Math. Phys. Kl. 1918, 235–257 (1918). Union’s Horizon 2020 research and innovation programme (grant
150. D. J. Gross, in Les Houches 1975, Sesssion XXVIII: Methods in 161. P. W. Anderson, Infrared conductivity of cuprate metals: 835279-Catch-22) (N.E.H.). The work on anomalous dimensions for
Field Theory (North-Holland, 1976), pp. 181–250. Detailed fit using Luttinger-liquid theory. Phys. Rev. B 55, conserved currents was funded through NSF DMR-2111379.
151. What is really required is that [qA] = 1, where q is the charge. 11785–11788 (1997). doi: 10.1103/PhysRevB.55.11785 Competing interests: The authors declare no competing interests.
In insisting that [A] = 1, we are setting q = 1, but still all 162. K. Limtragool, P. W. Phillips, Power-law optical License information: Copyright © 2022 the authors, some rights
of our statements refer to the product qA. conductivity from unparticles: Application to the reserved; exclusive licensee American Association for the
152. G. L. Nave, P. W. Phillips, Anomalous Dimensions for Boundary cuprates. Phys. Rev. B 92, 155128 (2015). doi: 10.1103/ Advancement of Science. No claim to original US government
Conserved Currents in Holography via the Caffarelli–Silvestre PhysRevB.92.155128 works. https://www.science.org/about/science-licenses-journal-
Mechanism for p-forms. Commun. Math. Phys. 366, 119–137 163. Y. Liao, Effects of unparticles on running of gauge couplings. article-reuse
(2019). doi: 10.1007/s00220-019-03292-z Eur. Phys. J. C 55, 483–488 (2008). doi: 10.1140/epjc/
153. R. A. Davison, B. Goutéraux, Dissecting holographic s10052-008-0601-y
conductivities. J. High Energy Phys. 2015, 90 (2015). 164. B. Basa, G. La Nave, P. W. Phillips, Classification of Submitted 15 May 2021; accepted 25 May 2022
doi: 10.1007/JHEP09(2015)090 nonlocal actions: Area versus volume entanglement 10.1126/science.abh4273

RES EARCH
◥ stability and to avoid the high-pressure drop

RE P OR T in industrial fixed-bed reactors, zeolite cata-
lysts need to be shaped into composites—for
CATALYSIS example, as extrudates with the use of a binder
such as clay (32). The location of Pt sites in
Maximizing noble metal utilization in solid catalysts these composite catalysts can be roughly clas-
sified into three types: (i) on the outer surface
by control of nanoparticle location of zeolite crystals; (ii) inside the zeolite crystals;
and (iii) on the binder (fig. S3). We controlled
Kang Cheng1,2†, Luc C. J. Smulders1†, Lars I. van der Wal1, Jogchum Oenema1, the Pt location on the composite catalysts
Johannes D. Meeldijk1,3, Nienke L. Visser1, Glenn Sunley4, Tegan Roberts4, Zhuoran Xu5, through either ion exchange (IE) by using
Eric Doskocil5, Hideto Yoshida1,6, Yanping Zheng2, Jovana Zečević1, Pt(NH3)4(NO3)2 as Pt precursor or ion ad-
Petra E. de Jongh1, Krijn P. de Jong1* sorption (IA) by using H2PtCl6 (fig. S4). The
actual weight loading of Pt was determined
Maximizing the utilization of noble metals is crucial for applications such as catalysis. We found that by means of inductively coupled plasma (ICP)
the minimum loading of platinum for optimal performance in the hydroconversion of n-alkanes for spectrometry (tables S3 and S4). Because all
industrially relevant bifunctional catalysts could be reduced by a factor of 10 or more through the raw materials are based on commercial pro-
rational arranging of functional sites at the nanoscale. Intentionally depositing traces of platinum ducts and the IE and IA methods are com-
nanoparticles on the alumina binder or the outer surface of zeolite crystals, instead of inside the zeolite monly applied in industry, no major barriers
crystals, enhanced isomer selectivity without compromising activity. Separation between platinum and are foreseen for the catalyst preparation on
zeolite acid sites preserved the metal and acid functions by limiting micropore blockage by metal a large scale.
clusters and enhancing access to metal sites. Reduced platinum nanoparticles were more active than High-angle annular dark-field scanning trans-
platinum single atoms strongly bonded to the alumina binder. mission electron microscopy (HAADF-STEM)
imaging combined with ultramicrotomy can
N
expose cross sections of zeolite crystals or
oble metals (NMs) are widely used in a alkanes to enhance the quality of liquid fuels shaped catalysts and enables the visualization
variety of commercial and emerging tech- (fig. S1), and its performance is influenced by of the spatial distribution of Al2O3 binder,
nologies, including catalytic converters in the Pt nanoparticle (NP) size and distribution, zeolite crystals, and Pt sites (29). The zeolite-
automobiles; electrocatalysts in hydrogen metal-support interactions, and acidic prop- Al2O3 composite, by use of acetic acid as a
fuel cells; and catalysts for petroleum, erties (20–23). To describe the required Pt peptizing agent in this work, exhibited a
biomass, and waste conversion (1–5). Increasing loading sufficient to maintain the bifunctional uniform distribution of zeolite crystals, which
demand for NMs in these applications is driving metal-acid balance in hydroconversion cata- were coated and stabilized by the Al2O3 binder
approaches that make more efficient use of lysts, Guisnet and coworkers proposed a widely (Fig. 1A). By contrast, large agglomerates of
NMs (6–9), including so-called single-atom applied parameter of the surface Pt sites to zeolites and Al2O3 were observed over zeolite+
catalysts (SACs) in which isolated single metal Brønsted acid sites ratio (npt/na) (24). To meet Al2O3 composite without acetic acid (fig. S5).
atoms or ions are stabilized by the support this criterion, the typical Pt loading for bi- Because the kinetic diameter of the Pt(NH3)42+
(10–13). However, strong metal-support inter- functional catalysts is in the range of 0.3 to (0.48 nm) is comparable with the HZSM-22
actions often lead to limited reducibility and 3 wt % (25–27). For integrating two functional pore size (0.45 by 0.55 nm) and HMOR pore size
in some cases low reactivity. Confinement of NMs components, Weisz’s intimacy criterion, which (0.67 by 0.70 nm), the diffusion of Pt(NH3)42+
inside zeolite channels or cages can benefit the is often interpreted as “the closer the better,” ions into the micropores during the IE process
adsorption of reagents and stabilization of has been applied to spatial organization (28). could be limited. The STEM images of Pt-
reaction intermediates to improve catalytic ac- Recently, we have shown at Pt loadings around HZSM-22 slices confirmed that after reduc-
tivity, product selectivity, or both (14–16). Suc- 0.5 wt % that separation of Pt and acid sites at tion, most Pt NPs were present on the external
cessful examples of confinement effects have the nanoscale enhanced the isomer selectivity crystal surfaces, regardless of the Pt loading,
been demonstrated for the conversion of small in hydroisomerization of linear alkanes, whereas and a small fraction was located inside the
molecules, including carbon monoxide (CO) placing Pt inside zeolite crystals with closest HZSM-22 crystals (Fig. 1B). The average sizes
oxidation, methane (CH4) oxidation, and the proximity promoted acid-catalyzed cracking of Pt NPs for 0.2Pt-HZSM-22 (where 0.2 is Pt
water-gas shift reaction (17–19). or overcracking reactions because of the high loading in weight percent) and 1.0Pt-HZSM-22
In the petrochemical industry, platinum (Pt) concentrations of intracrystalline carbenium were 1.5 and 1.7 nm, respectively. For 0.2Pt-
is frequently used in combination with acidic ions (29–31). With this concept of nanoscale HMOR and 1.0Pt-HMOR, the average diame-
zeolites for the hydroconversion of linear intimacy, we found that lowering the Pt loading ters of Pt NPs were 1.2 and 1.6 nm, respectively,
holds potential (30), which inspired us to and the Pt NPs were exclusively deposited in-
study the lower limit of NMs for catalysis as side HMOR crystals (Fig. 1C). The Pt NPs were
1
Materials Chemistry and Catalysis, Debye Institute for a function of the Pt NP location. larger than the micropores of the zeolite and
Nanomaterials Science, Utrecht University, 3584 CG Utrecht, We controlled the Pt location and loading were likely accommodated by the local destruc-
Netherlands. 2State Key Laboratory of Physical Chemistry of
Solid Surfaces, College of Chemistry and Chemical
on industrially relevant platinum-zeolite- tion of micropores during NP growth, an effect
Engineering, Xiamen University, Xiamen 361005, China. alumina (Al2O3) composite catalysts to de- confirmed in previous studies (33).
3
Electron Microscopy Centre, Utrecht University, 3584 CG crease the Pt loading for hydroisomerization The zeolite powders containing Pt NPs were
Utrecht, Netherlands. 4Applied Sciences, bp Innovation and
while maintaining optimal performance. We then mixed with boehmite and acetic acid to
Engineering, BP plc, Saltend, Hull HU12 8DS, UK. 5Applied
Sciences, bp Innovation and Engineering, BP plc, Naperville, used n-heptane as a model molecule relevant prepare bifunctional catalysts with a zeolite-
IL 60563, USA. 6The Institute of Scientific and Industrial to gasoline upgrading. We used the one- to-Al2O3 weight ratio of 1/1. The TEM imaging
Research, Osaka University, 8-1 Mihogaoka, Ibaraki, dimensional (1D) zeolites HZSM-22 and HMOR revealed that for IE-prepared Pt-HZSM-22/
Osaka 567-0047, Japan.
*Corresponding author. Email: k.p.dejong@uu.nl to construct bifunctional catalysts (tables S1 Al2O3, most Pt NPs remained on the HZSM-22
†These authors contributed equally to this work. and S2 and fig. S2). To provide mechanical crystal surfaces with only a few NPs inside the
Cheng et al., Science 377, 204–208 (2022) 8 July 2022 1 of 4

RES EARCH | REPOR T
Fig. 1. HAADF-STEM and STEM–energy-dispersive x-ray (EDX) images of (C) (C1) 0.2Pt-HMOR, (C2) 1.0Pt-HMOR, and (C3) scheme for Pt distribution over
ultramicrotomed 70-nm-thick sections of catalysts. (A) HZSM-22/Al2O3 HMOR. (D and E) 0.01Pt-HZSM-22/Al2O3 and 0.01Pt-Al2O3/HZSM-22, respec-
composite by using acetic acid as a peptizing agent. The images (A1) to (A3) tively. (F and G) 0.5Pt-HZSM-22/Al2O3 and 0.5Pt-Al2O3/HZSM-22, respectively.
show the location of silicon (green) and aluminum (red), which is indicative of the (H and I) 0.5Pt-HMOR/Al2O3 and 0.5Pt-Al2O3/HMOR, respectively. In the
presence of the HZSM-22 and Al2O3 components. (B) (B1) 0.2Pt-HZSM-22, schematic representations, the red dots and blue hexagons indicate Pt NPs and
(B2) 1.0Pt-HZSM-22, and (B3) scheme for Pt distribution over HZSM-22. zeolite crystals, respectively.
HZSM-22 crystals, and those NPs were near cient to maintain the n-C7 conversion and Over HMOR-based catalysts, both the n-C7
NPs present on HZSM-22 surfaces; no NPs selectivity of C7 isomers (Fig. 2A). Such a conversion and the selectivity of C7 isomers
were observed on the mesoporous Al 2O 3 low Pt loading is usually insufficient for increased with the Pt loading, and overall,
binder (Fig. 1, D and F, and fig. S6A). Con- catalyzing this reaction. However, 0.01Pt- the performance of catalysts with Pt inside
versely, almost all of the Pt NPs were located Al2O3/HZSM-22 exhibited inferior activity HMOR crystals was more strongly influenced
on the Al2O3 binder for Pt-Al2O3/HZSM-22 and selectivity (Fig. 2B). The C7 isomers by the Pt loading (Fig. 2, C and D, and fig. S8).
prepared with the IA method (Fig. 1, E and G, consisted of monobranched and traces of At the same Pt loading, the specific n-C7 con-
and fig. S6B). Correspondingly, the Pt NPs dibranched C7 isomers (mono-C7 and di-C7, version over Pt-HMOR/Al2O3 was lower than
were still exclusively present inside the HMOR respectively), and the main side products the corresponding Pt-Al2O3/HMOR, probably
crystals for IE-prepared Pt-HMOR/Al2O3 (Fig. were propane, n-butane, and iso-butane from because of pore blockage or poor accessibility
1H), and all of the Pt NPs were deposited on acid-catalyzed hydrocracking (30). of Pt NPs (34). With a 10-fold increase of Pt
the Al2O3 binder for IA-prepared Pt-Al2O3/ At similar conversion levels, 0.01Pt-HZSM-22/ loading from 0.05 to 0.5%, the n-C7 conversion
HMOR (Fig. 1I). With this methodology, we Al2O3 also provided higher selectivity of C7 increased from 58 to 71% with a selectivity of
prepared a series of composite catalysts with isomers than did 0.01Pt-Al2O3/HZSM-22 (fig. C7 isomers ~80% over Pt-Al2O3/HMOR, whereas
different Pt loadings and locations (table S4). S9). Increasing the Pt loading >0.01 wt % de- the n-C7 conversion increased from 22 to 67%
The Pt loading in all cases had a minor effect creased the selectivity of C7 isomers, which and the selectivity of C7 isomers increased
on the catalyst acidity (fig. S7). was likely the result of more Pt NPs sited in from 28 to 72% over Pt-HMOR/Al2O3. Regard-
The catalytic performance of HZSM-22– the pore mouth region of HZSM-22, which less of the Pt loading and location over the
based catalysts in the hydroisomerization of would promote cracking reactions (30). For composite catalysts, after more than 100 hours,
n-C7 was strongly affected by the Pt loading Pt-Al2O3/HZSM-22 with Pt-on-Al2O3, both the we observed virtually no catalyst deactivation
and distribution, especially at loading of 0.005 n-C7 conversion and selectivity of C7 isomers for both HZSM-22– and HMOR-based catalysts
to 0.05 wt % (Fig. 2 and fig. S8). With an in- increased steadily and then reached a plateau (figs. S10 and S11).
crease of Pt loading, the n-C7 conversion com- as Pt loading increased to 0.05 wt % (Fig. 2B). The dependence of the maximum yield of C7
pared at identical temperature and weight With Pt-on-binder, therefore, a minimum load- isomers, an important performance indicator
hourly space velocity (WHSV) increased first ing of 0.05 wt % was required to sustain the in alkane hydroisomerization (26), on the plat-
and then reached a plateau. Unexpectedly, n-heptane conversion and the selectivity of inum loading is summarized in Fig. 3. The
with the Pt-on–HZSM-22 configuration, an C7 isomers, to ensure the equilibrium of the acid density of HMOR is much higher than
ultralow Pt loading of 0.01 wt % was suffi- dehydrogenation-hydrogenation reactions. that of HZSM-22. Therefore, HMOR-based

RES EARCH | REPOR T
catalysts required more Pt sites to obtain the crease of Pt loading above 0.05 wt % maximized yields of C7 isomers (Fig. 3B). Disadvantageous
same nPt/na ratio (fig. S12). A composite cat- the yield of C7 isomers toward 68%. Likewise, for the yield of C7 isomers was to confine the Pt
alyst with Pt-on–HZSM-22 exhibited an opti- the maximum yield of C7 isomers for Pt-Al2O3/ inside HMOR crystals, which promoted the
mum loading of only 0.01 wt % for maximizing HMOR went up with Pt loading, and rela- cracking reaction and prevented the conver-
the yield of C7 isomers (≈64%) (Fig. 3A). For tive to HZSM-22, a much higher loading of Pt sion of n-C7. The maximum yield of C7 isomers
Pt-Al2O3/HZSM-22 with Pt-on-binder, the in- (≈0.1 wt %) was needed to provide maximum over 0.05Pt-Al2O3/HMOR was near that over
0.5Pt-HMOR/Al2O3, suggesting that a rational
placement of Pt could lead to a 10-fold re-
duction in its usage.
In the low loading range (≤0.01 wt %), Pt
placed on the Al2O3 binder was less effective
than on HZSM-22 crystals (Figs. 2, A and B,
and 3A). Highly dispersed Pt sites stabilized
by the alumina (Pt–O–Al) could be difficult to
reduce and not be sufficiently active for de-
hydrogenation of n-heptane. To investigate
this, we intentionally prepared a SAC-Pt-Al2O3/
HMOR catalyst with single-atom Pt dispersed
on Al2O3, which was almost inactive in n-heptane
hydroconversion (fig. S13). The x-ray absorp-
tion spectroscopy (XAS) shows the k2-weighted
(where k is photo-electron wave number), phase-
corrected extended x-ray absorption fine struc-
ture (EXAFS) spectra of reduced Pt catalysts
together with Pt standards (Fig. 4A). The
scattering from the first shell of Pt–O led to
the peak below 2.0 Å, whereas the scattering
Pt–Pt in the Pt foil resulted in peaks at 2.4
and 2.9 Å (7). The scattering peak at 2.2 Å
observed with a bond distance shorter than
that of Pt–Pt may indicate the formation of Pt
dimers (35).
Strong Pt–Pt scattering for 0.1Pt-HMOR and
Fig. 2. Hydroconversion of n-heptane over bifunctional catalysts with different Pt loadings and 0.1Pt-HZSM-22 with 0.1 wt % loadings could
locations. (A) Pt-HZSM-22/Al2O3 with Pt on the HZSM-22 crystals and (B) Pt-Al2O3/HZSM-22 with Pt on be observed (Fig. 4A), which is indicative of
the Al2O3 binder. Data are partly from (30). (C) Pt-HMOR/Al2O3 with Pt inside the HMOR crystals and the formation of metallic Pt NPs or clusters. A
(D) Pt-Al2O3/HMOR with Pt on the Al2O3 binder. Data are partly from (30). Reaction conditions were pressure Pt–O scattering peak was observed over 0.01Pt-
(P) = 10 bar, H2/n-C7 = 10/1, WHSV = 2.1 gn-C7 gcat–1 hour–1, and temperature (T) = 320°C for HZSM-22– HZSM-22/Al2O3 and 0.01Pt-Al2O3/HZSM-22,
based catalysts and T = 270°C for HMOR-based catalysts. indicating that single-atom Pt (ions) was
present within the two samples together with
Pt NPs observed with electron microscopy
(Fig. 1, D and E). More than half of Pt was
metallic over 0.01Pt-HZSM-22/Al2O3, which
exhibited higher reactivity in n-C7 conversion
than that of 0.01Pt-Al2O3/HZSM-22 that con-
tained mainly oxidized Pt (fig. S14 and table
S5). As the Pt loading on the Al2O3 binder in-
creased from 0.01 to 0.05 wt %, the Pt–O scat-
tering became negligible, and Pt–Pt scattering
appeared, which suggested the formation of
larger numbers of small Pt NPs or clusters in
line with electron microscopy results (Fig. 1).
The in situ x-ray photoelectron spectroscopy
(Fig. 4B) further suggested reduced 0.1Pt-
HZSM-22/Al2O3 and 0.1Pt-Al2O3/HZSM-22 to
mainly contain metallic Pt (313.7 eV), whereas
the atomically dispersed Pt on Al2O3 (SAC-
0.2Pt-Al2O3) was in a partially oxidized state
(314.6 eV) (36). The weaker Pt signals of 0.1Pt-
Fig. 3. The effect of Pt loading on the maximum yield of C7 isomers. (A) Pt–HZSM-22–Al2O3 composite HZSM-22/Al2O3 than those of 0.1Pt-Al2O3/
catalysts, with Pt NPs placed either on Al2O3 binder or on HZSM-22 crystals, and (B) Pt-HMOR-Al2O3, HZSM-22 imply that a fraction of Pt NPs might
with Pt NPs placed either on Al2O3 binder or inside HMOR crystals. Reaction conditions were P = 10 bar, be confined in the micropores of HZSM-22,
H2/n-C7 = 10/1, and WHSV = 2.1 gn-C7 gcat–1 hour–1. The red dots and blue hexagons indicate Pt NPs and causing cracking reactions at high (≥0.1 wt %)
zeolite crystals, respectively. Pt loadings (Fig. 2A).

RES EARCH | REPOR T
25. A. C. M. van den Broek, J. van Grondelle, R. A. van Santen,

J. Catal. 167, 417–424 (1997).
26. J. A. Martens et al., ChemSusChem 6, 421–425 (2013).
27. Z. Guo et al., Angew. Chem. Int. Ed. 59, 1548–1551
(2020).
28. P. B. Weisz, E. W. Swegler, Science 126, 31–32 (1957).
29. J. Zečević, G. Vanbutsele, K. P. de Jong, J. A. Martens, Nature
528, 245–248 (2015).
30. K. Cheng et al., Angew. Chem. Int. Ed. 59, 3592–3600
(2020).
31. J. Oenema et al., ACS Catal. 10, 14245–14257 (2020).
32. N. L. Michels, S. Mitchell, J. Pérez-Ramírez, ACS Catal. 4,
2409–2417 (2014).
33. J. Zečević, A. M. J. van der Eerden, H. Friedrich, P. E. de Jongh,
K. P. de Jong, ACS Nano 7, 3698–3705 (2013).
34. M. Guisnet, F. Alvarez, G. Giannetto, G. Perot, Catal. Today 1,
415–433 (1987).
35. U. Müller, K. Sattler, J. Xhie, N. Venkateswaran, G. Raina, Z.
Phys. D At. Mol. Clust. 19, 319–321 (1991).
36. T. Huizinga, H. F. J. van ’t Blik, J. C. Vis, R. Prins, Surf. Sci. 135,
Fig. 4. Chemical state of Pt. (A) The k2-weighted Pt L3 edge EXAFS spectra in R space of bifunctional 580–596 (1983).
catalysts. Catalysts were reduced under 3.5 vol % H2 flow at 350°C for 1 hour. PtO2 and Pt foil were used as AC KNOWL ED GME NTS
Pt standards. (B) In situ XPS spectra of reduced catalysts. N. Nikolopoulos, G. Totarella, and S. Zanoni are acknowledged
for measuring and analyzing the N2 physisorption data. Funding:
This work has been supported by BP plc and Topsector Chemie
To increase the Pt utilization over bifunc- RE FERENCES AND NOTES consortia for Knowledge and Innovation (TKIs). K.P.d.J.
acknowledges funding from the European Research Council (ERC), EU
tional Pt-zeolite catalysts, we have varied the 1. A. Wang, J. Li, T. Zhang, Nat. Rev. Chem. 2, 65–81
FP7 ERC Advanced Grant 338846. K.C. thanks the support from
proximity between sites by controlling the spa- (2018).
the National Key Research and Development Program of Ministry
tial distribution of Pt NPs. Because of the low 2. L. Chong et al., Science 362, 1276–1281 (2018). of Science and Technology 2020YFB0606401. H.Y. thanks the
3. R. T. Hannagan et al., Science 372, 1444–1447 Program for Advancing Strategic International Networks to
diffusivity of alkanes and alkenes, Pt sites (2021). Accelerate the Circulation of Talented Researchers by JSPS.
should be placed in an accessible location 4. F. Zhang et al., Science 370, 437–441 (2020). Author contributions: K.C. conducted the catalyst preparation;
in solid catalysts rather than in the micro- 5. Z. Huang et al., Nat. Catal. 3, 170–178 (2020). catalytic evaluation; sample characterizations by NH3-TPD, SEM,
pores of a zeolite, such as on a mesoporous 6. L. DeRita et al., Nat. Mater. 18, 746–751 (2019). and XRF; and drafted the manuscript. L.C.J.S. assessed the
7. H. Jeong et al., Nat. Catal. 3, 368–375 (2020). catalyst stability, performed additional characterizations, and
alumina binder or the outer surface of the helped revise the manuscript. L.I.v.d.W., J.D.M., N.L.V., and H.Y.
8. K. D. Rasmussen, H. Wenzel, C. Bangs, E. Petavratzi, G. Liu,
zeolite crystals. Additionally, maintaining a Environ. Sci. Technol. 53, 11541–11551 (2019). performed STEM imaging. J.O. guided the catalyst preparation and
separation between Pt and zeolite acid sites maintained the reaction unit. G.S., T.R., P.E.d.J., and J.Z.
9. A. E. Hughes, N. Haque, S. A. Northey, S. Giddey, Resources 10,
contributed to the discussion of all the results. Z.X. and E.D.
can preserve the metal and acid functions to 93 (2021).
performed the XAS measurements and analyzed the data. Y.Z.
the greatest extent through three main mech- 10. B. Qiao et al., Nat. Chem. 3, 634–641 (2011).
performed XPS analysis. K.P.d.J. initiated and supervised the
11. S. Mitchell, E. Vorobyeva, J. Pérez-Ramírez, Angew. Chem.
anisms: (i) limiting micropore blockage by Int. Ed. 57, 15316–15329 (2018).
project and manuscript writing. All authors approved the final
version of the manuscript. Competing interests: The authors
metal clusters, (ii) enhancing accessibility of 12. Z. Li et al., Chem. Rev. 120, 623–682 (2020). declare no competing interests. Data and materials availability:
metal sites, and (iii) favoring pore mouth 13. H. Xiong, A. K. Datye, Y. Wang, Adv. Mater. 33, e2004319 All data and images are available in the manuscript or supplementary
catalysis to reduce secondary cracking reac- (2021). materials. License information: Copyright © 2022 the
tions (29). Furthermore, reducible Pt ensem- 14. J. D. Kistler et al., Angew. Chem. Int. Ed. 53, 8904–8907 authors, some rights reserved; exclusive licensee American
(2014). Association for the Advancement of Science. No claim to
ble sites were confirmed to be more active as 15. N. Kosinov, C. Liu, E. J. M. Hensen, E. A. Pidko, Chem. Mater. original US government works. https://www.science.org/about/
the dehydrogenation-hydrogenation function 30, 3177–3198 (2018). science-licenses-journal-article-reuse
than Pt SAC, which was strongly bonded by 16. L. Liu et al., Nat. Mater. 18, 866–873 (2019).
the alumina binder, especially at low platinum 17. S. M. Wu, X. Y. Yang, C. Janiak, Angew. Chem. Int. Ed. 58,
12340–12354 (2019).
loadings. Our findings generally agree with 18. J. Zhang et al., Nat. Catal. 1, 540–546 (2018).
science.org/doi/10.1126/science.abn8289
the classical criterion that a minimum amount Materials and Methods
19. Z. Jin et al., Science 367, 193–197 (2020).
Supplementary Text
of NMs is necessary to maintain the metal- 20. L. Liu, A. Corma, Chem. Rev. 118, 4981–5079 (2018). Figs. S1 to S14
acid balance and ensure a quasi-equilibration 21. B. Zhang et al., J. Am. Chem. Soc. 141, 8185–8197 (2019). Tables S1 to S5
of the dehydrogenation-hydrogenation reac- 22. T. W. van Deelen, C. Hernández Mejía, K. P. de Jong, Nat. Catal. References (37–40)
2, 955–970 (2019).
tions, but this study confirms that the required 23. L. I. van der Wal et al., ACS Catal. 11, 3842–3855 (2021). Submitted 11 January 2022; resubmitted 24 March 2022
loading for NMs is profoundly affected by how 24. G. E. Giannetto, G. R. Perot, M. R. Guisnet, Ind. Eng. Chem. Accepted 18 May 2022
functional sites are arranged at the nanoscale. Prod. Res. Dev. 25, 481–490 (1986). 10.1126/science.abn8289

RES EARCH
THERMOELECTRICS improved the conversion efficiencies of the

fabricated single and segmented TE modules
High figure-of-merit and power generation in to 10.5 and 13.3% at a temperature difference
of 506 K (Fig. 1B), respectively. These experi-
high-entropy GeTe-based thermoelectrics mental values were among the highest in the
entire TE community (4, 34–45), indicating
Binbin Jiang1,2†, Wu Wang1†, Shixuan Liu1, Yan Wang1, Chaofan Wang1, Yani Chen1, Lin Xie1, an important step toward TE applications.
Mingyuan Huang1, Jiaqing He1,3* Solubility in alloys can be largely increased
by increasing the elemental species based on
The high-entropy concept provides extended, optimized space of a composition, resulting in unusual entropy-driven structural stabilization, which
transport phenomena and excellent thermoelectric performance. By tuning electron and phonon has been summarized as high-entropy mate-
localization, we enhanced the figure-of-merit value to 2.7 at 750 kelvin in germanium telluride–based rials (4, 16). Because of the well-known dis-
high-entropy materials and realized a high experimental conversion efficiency of 13.3% at a temperature advantage of high lattice thermal conductivity
difference of 506 kelvin with the fabricated segmented module. By increasing the entropy, the increased in GeTe, this high-entropy phenomenon can
crystal symmetry delocalized the distribution of electrons in the distorted rhombohedral structure, be used to lower its lattice thermal conductivity
resulting in band convergence and improved electrical properties. By contrast, the localized phonons by increasing the phonon scattering. We used
from the entropy-induced disorder dampened the propagation of transverse phonons, which was the Ag, Sb, and Pb as the alloying elements be-
origin of the increased anharmonicity and largely depressed lattice thermal conductivity. We provide a cause of their high limited solubility. We found
paradigm for tuning electron and phonon localization by entropy manipulation, but we have also obvious impurity peaks in the x-ray powder
demonstrated a route for improving the performance of high-entropy thermoelectric materials. diffraction (XRD) patterns for Ge0.88Pb0.12Te,
Ge0.89Ag0.11Te and Ge0.87Sb0.13Te samples
E
when the number of components was three,
nergy waste during the conversion and due to the entropy-driven structural stabiliza- and these peaks were indexed to PbTe, Ag2Te,
operation processes is a serious prob- tion also may strengthen phonon propagation. and Sb2Te3, respectively (Fig. 2A). When the
lem globally and has attracted increasing Therefore, the challenge is to avoid possible number of components increased to five,
attention (1–4). Thermoelectric (TE) tech- electron localization while achieving the lo- the second phases were eliminated in all of the
nology provides the prospect for harvest- calized distribution of phonons in high-entropy samples, indicating the stabilization phenom-
ing waste heat from scattered heat sources, materials. enon of the phase structure dominated by the
which suffer from low conversion efficiency GeTe belongs to the IV-VI chalcogenides and increasing entropy. The single phase was stable
(5–8). Conversion efficiency is related primar- stabilizes as a rhombohedral structure at room even when we further introduced Cd, Mn, and
ily to the dimensionless figure-of-merit [zT = temperature by distorting high-temperature Sn, which formed a more complicated com-
S2sT/(ke + kL)] of thermoelectric materials, cubic symmetry (20–22). IV-VI chalcogenides position with six components. These results
where S, s, T, ke, and kL denote the Seebeck with high TE performance typically exhibit proved that the high-entropy phenomenon
coefficient, electrical conductivity, absolute octahedral-like coordination, described as a could be used to obtain single-phase compo-
temperature, carrier, and lattice thermal con- cationic atom surrounded by six anionic atoms. sitions with stable structures, offering the
ductivities, respectively (9–12). To pursue high In these materials, however, each atom con- possibility of increasing the solubility limit.
TE performance, the electrical and thermal tains only three valence electrons on average The ternary chalcogenide Ge-Sb-Te is a typ-
transport properties should be simultaneously because of the small sp-hybridization result- ical phase-change material with an amorphous
decoupled and optimized. Generally, success- ing from the large energy difference between structure (23–25), which typically exhibits sub-
ful strategies for improving TE performance the s-orbitals and the p-orbitals (23, 24). The stantially deteriorated electrical transport prop-
require high electrical transport properties sharing of three valence electrons to form erties because of the damaged periodicity of
from the ordered band structure and low lat- six bonds results in delocalized electrons, units atomic arrangement. To further differentiate
tice thermal conductivity from the disordered like localized electron distribution in covalent the crystalline high-entropy composition from
atomic arrangement, following the phonon- bonding, which has been called metavalent or the amorphous form and to confirm the sites
glass electron-crystal concept (13–15). The re- resonant bonding (25, 26). We demonstrated of the alloyed elements, we conducted energy-
cently proposed high-entropy strategy provides that delocalized electrons from increased dispersive spectroscopy (EDS) analysis at an
an extended composition range, in which a crystal symmetry and localized phonons from atomic scale using STEM. We performed EDS
stabilized crystalline structure and disordered entropy-induced disorder could coexist in high- mapping of Ge0.61Ag0.11Sb0.13Pb0.12Bi0.01Te and
atomic arrangement can be simultaneously entropy GeTe-based materials, resulting in a zT other samples (Fig. 2, B to G, and fig. S1). The
realized (4, 16–19). Possible Anderson localiza- value of 2.7 at 750 K (Fig. 1A) (11, 20–22, 27–33). atomic arrangements of all five elements
tion of the electrons in the disordered system, The atomic electric field and charge density showed good periodicity, indicating the intu-
however, may deteriorate the electrical trans- of the GeTe-based materials, which were deter- itive appearance of the crystalline high-entropy
port properties (6). Furthermore, the increased mined by differential phase contrast (DPC) composition. We found that the Ge, Ag, Sb,
crystal symmetry of low-symmetry materials imaging in scanning transmission electron and Pb atoms were located in the same site,
microscopy (STEM), clearly showed the in- whereas the Te atom was located solely in
creased delocalization of electrons from de- another site, demonstrating that possible anti-
1
Shenzhen Key Laboratory of Thermoelectric Materials, creasing rhombohedral distortion, which was site defects were absent even if Sb and Te
Department of Physics, Southern University of Science and responsible for improved band convergence had a similar nature (46). The homogeneous
Technology, Shenzhen 518055, China. 2School of Materials and power factor. Moreover, the localized distribution of all five elements further con-
and Energy, University of Electronic Science and Technology
of China, Chengdu 610054, China. 3Guangdong-Hong Kong- phonons from entropy-induced disorder largely firmed the single phase (fig. S2), illustrating
Macao Joint Laboratory for Photonic-Thermal-Electrical damped the propagation of transverse phonons that high entropy should provide the driving
Energy Materials and Devices, Southern University of and increased the anharmonicity, resulting in force for stabilizing different elements, with
Science and Technology, Shenzhen 518055, China.
*Corresponding author. Email: hejq@sustech.edu.cn ultralow lattice thermal conductivity. The high content that exceeded the solubility limit of
†These authors contributed equally to this work. performance of the GeTe-based materials one structure.
Jiang et al., Science 377, 208–213 (2022) 8 July 2022 1 of 6

RES EARCH | REPOR T
range, indicating a temperature-independent

trend (Fig. 3B). This phenomenon demonstrated
that point defect scattering and the Umklapp
scattering for the phonons were very strong
in this material system (47). Because of the
largely improved PF and depressed kL values
at a low temperature range, the zT values
improved to 0.6 at room temperature and to 2.7
at 750 K for the Ge0.61Ag0.11Sb0.13Pb0.12Bi0.01Te
sample (Fig. 3C). We calculated the aver-
age zT in the 300 to 750 K range, which
showed a highly comparable value >1.7 for
the Ge0.61 Ag0.11 Sb0.13Pb0.12Bi0.01 Te sample
because of the largely improved TE perform-
Fig. 1. High physical properties of the GeTe-based TE materials and modules. (A) Temperature
ance at the low temperature range (Fig. 3D).
dependence of zT values for the high-entropy Ge0.61Ag0.11Sb0.13Pb0.12Bi0.01Te sample in this work, for which
The high-performance high-entropy compo-
the reported results in the literature (11, 20–22, 27–33) are also included for comparison. (B) Maximum
sition also shows a good reproductivity and
conversion efficiency (hmax) as a function of DT for the fabricated GeTe-based TE modules in this work. The
thermal stability (figs. S6 and S7).
reported results of the GeTe-based, Pb(Te,Se,S)/BT–based, SKD/BT–based, and HH/BT–based materials
Because of the largely improved zT values
from the literature (4, 34–45) are also included for comparison. BT, SKD, and HH represent Bi2Te3,
in the high-entropy GeTe-based materials, we
Skutterudite and half-Heusler materials, respectively.
fabricated a TE module to realize a high ex-
perimental conversion efficiency. We fabri-
cated four TE modules, which were labeled
as 11, 7:3, 8:3, and 9:3, respectively (Fig. 3F).
The module labeled as 11 was a single module
with a leg length of 11 mm, which consisted
of p-type Ge0.61Ag0.11Sb0.13Pb0.12Bi0.01Te and
n-type Pb0.997In0.003Te0.996I0.004. The module
labeled as 7:3 was a segmented module with a
high-temperature leg length of 7 mm and a
low-temperature leg length of 3 mm. The
high-temperature legs consisted of p-type
Ge 0.61 Ag0.11 Sb0.13Pb0.12Bi 0.01 Te and n-type
Pb 0.997 In 0.003 Te 0.996 I 0.004 , and the low-
temperature legs consisted of commercial
p-type Bi0.5Sb1.5Te3 and n-type Bi2Te2.7Se0.3.
The construction of the modules labeled 8:3
and 9:3 was understood to be analogous to
the ones labeled 7:3. We also measured the
output properties by the commercial Power
Generation Efficiency Characteristics Evalu-
ation System (PEM-2, Advance Riko, Japan)
(Fig. 3, E and F, and figs. S8 to S12).
Fig. 2. Formation of the high-entropy GeTe-based materials. (A) XRD patterns of GeTe-based materials We found that the open-circuit voltage was
with different numbers of components. Blue arrows indicate the impurity peaks of the second phases. largely improved by using the Bi2Te3-based
The nominal compositions of the GeTe-based materials labeled with S1 to S13 are shown by their chemical materials as low-temperature legs (Fig. 3E),
formulas. (B to G) Overall EDS mapping image and individual Ge (C), Ag (D), Sb (E), Pb (F), and Te (G) which exhibited a positive effect on the seg-
EDS mapping images of the high-entropy Ge0.61Ag0.11Sb0.13Pb0.12Bi0.01Te sample. mented module. Because of the improved
open-circuit voltage, which was related to
the Seebeck coefficient, the output power (P)
In our previous work, we proved that a high- conductivity decreased with increasing ele- calculated by P = IU was also improved (Fig.
entropy composition should have ultralow lat- mental species, whereas the Seebeck coeffi- 3E). The output power reached a maximum
tice thermal conductivity due to the increased cient followed the opposite trend, the obvious value of 3.5 W for the 7:3 module with a hot-
lattice strains while maintaining the electrical result of the decreased carrier concentration. side temperature of 800 K. Subsequently, the
transport path by stabilizing the crystal struc- The calculated power factor (PF = sS2) at a maximum value of the output power decreased
ture (4). This was also the original motive for low temperature range for the high-entropy with increasing leg length because of the in-
introducing different elements into GeTe to samples was much larger than that for the low- creased internal resistance of the module. The
form the high-entropy materials. We measured entropy samples with multiple phases and pure increased internal resistance of the module
the TE properties of a series of samples that GeTe (Fig. 3A). This was unusual and could was illustrated by the increased slope of the
were formed by increasing the elemental spe- not be completely explained by the eliminated current-dependent voltage curve (Fig. 3E).
cies, the nominal compositions of which are electron scattering around the phase boundary. Although the increased leg length exhibited
listed in detail in Fig. 2A (labeled as S1 to S13), As expected, the lattice thermal conductivity decreased output power, the heat flow (Q) also
and determined the electrical conductivity kL was largely depressed to an ultralow value decreased. Then the conversion efficiency (h)
and Seebeck coefficient (fig. S4). The electrical (~0.3 Wm−1K−1) throughout the temperature would have an optimized value with increasing

RES EARCH | REPOR T
Fig. 3. TE performance of the GeTe-based materials and modules. (A to C) Temperature dependences of the power factor PF (A), lattice thermal conductivity kL
(B), and zT (C) values of all samples in this work. (D) Comparison of the average zT value in this work with the reported results for GeTe, SnSe, PbTe, FeNbSb,
Ce0.85Fe3CoSb12, Cu2Se, and SnTe in the literature (5, 11, 13, 17, 20–22, 29–35, 41, 61, 62). (E and F) Current dependences of the output voltage and output power (E)
and conversion efficiencies (F) for the fabricated 11, 7:3, 8:3, and 9:3 modules.
leg length due to the relation of h = P/(P + Q). tion. The atomic structures of the GeTe and tially caused redistribution and changes to
The conversion efficiency reached a maximum Ge0.61Ag0.11Sb0.13Pb0.12Bi0.01Te samples along electron localization. We further applied DPC-
value of 13.3% for the segmented 8:3 module the [110] zone axis were revealed by high-angle STEM imaging to obtain the atomic electric
at a hot-side temperature of 800 K, which was annular dark-field (HAADF) imaging in STEM field and charge density of the GeTe-based
much higher than 10.5% for the single 11 mod- (Fig. 4, A and B). With the Z-contrast nature materials (49). The electron probe deflected
ule (Fig. 3F). Compared with the high conver- of HAADF imaging (where Z indicates the by the electric field of the specimen caused a
sion efficiency in the literature (tables S2 and atomic number), the brighter spots in the shift in the center of mass of the diffraction
S3), our GeTe-based module realized a higher HAADF images of both samples indicated Te pattern, which was determined by the seg-
value than the others (PbTe, GeTe, Skutterudite, sites and thelower-intensity spots indicated mented detector in DPC-STEM (fig. S15) and
and half-Heusler). The origin of this high con- Ge sites. The Ge sites incorporated with alloy- used to calculate the projected electric field
version efficiency should be from the optimized ing elements in high-entropy sample exhibited with Ehrenfest’s theorem (50, 51). We observed
zT over the whole temperature range. higher intensity compared with the same pris- coupling between the electric field of the Ge
Strong alloying scattering and Anderson tine sites. We also found that the Ge atoms and Te sites in the pristine sample (Fig. 4D),
localization will always occur for electrons deviated from the geometric center of the although the coupling effect of the electric
in disordered materials, typically resulting in rhombus formed by Te atoms in the GeTe field in these two sites was largely reduced in
the deteriorated electrical transport properties sample, resulting in a polarized electric field the high-entropy sample (Fig. 4E). The pro-
(6). In our experiment, we introduced Ag, Sb, and the origin of ferroelectricity (48). The jected electric field images of the GeTe-based
and Pb atoms into the Ge sites to maintain the deviation of Ge atoms from the geometric cen- samples were further converted to projected
ordered Te sublattice. Because the contribu- ter is defined as a variable value of d (Fig. 4C). charge density maps according to Gauss’s law,
tion of Te-5p orbitals to the valence maximum The d value decreased with increasing ele- as shown in Fig. 4, G and H. In the octahedron
(S, h, Z, and L bands) is much higher than that mental species (Fig. 4F), as calculated from formed by one Ge atom and six Te atoms, we
of Ge-4p orbitals (fig. S13), Anderson localiza- atomic-resolution HAADF images of the four observed three shorter bonds and three longer
tion from disordered Ge sites would have less GeTe-based samples (fig. S14). This was con- bonds (Fig. 4J). The electrons were highly
effect on the electrical transport properties of sistent with the XRD results (Fig. 2A) where localized and formed shorter bonds because
p-type GeTe. The weakened scattering around the (2-10) and (104) peaks in the rhombohedral of symmetry breaking in the rhombohedral
the phase boundary and Anderson localiza- structure gradually converged into one (220) structure with a distorted lattice. The fair
tion of the electrons were the possible reasons peak in the cubic structure. The increased features in the gaps of the coupled Ge and Te
for the well-maintained PF in the high-entropy crystal symmetry from the rhombohedral atoms (shorter bonds) in the charge density
materials; however, the improved PF with in- to the cubic structure by the formation of map of the GeTe sample (Fig. 4G) provided
creasing entropy (Fig. 3A) was surprising. The high-entropy materials was the origin of the an indicator for the existence of localized
electrical transport properties were closely weakened ferroelectricity (48). electrons. By forming high-entropy mate-
related to the localization of electrons, and The tuned distortions quantified by the rials with more element species, the charge
were visualized by charge density distribu- variable d in the GeTe-based materials poten- density in the gaps between the Ge sites and

RES EARCH | REPOR T
of the improved periodicity in undistorted

atomic arrangements (52). For example, the
lattice thermal conductivity of rhombohedral
GeTe decreased with increasing tempera-
ture due to strengthened Umklapp scattering,
whereas cubic GeTe exhibited an increased
value after the phase transition (Fig. 3B, black
line) (20, 47). The average long-range crystal
symmetry was increased with increasing en-
tropy, as shown by the decreased deviation
value of d (Fig. 4F). The disordered distri-
bution of different elemental species in one
atomic site caused fluctuations in the atomic
environment and the breaking translational
symmetry at an atomic scale. This indicated
that many different octahedrons were com-
posed of the six Te atoms, which were cen-
tered on the Ge, Ag, Sb, or Pb atoms. Because
of fluctuations in the atomic environment,
the lattice also showed changed parameters
in different regions, resulting in large lattice
strains. We used geometric phase analysis
based on the high-resolution STEM-HAADF
images (fig. S18, A and B) to measure the lat-
tice strains of the high-entropy GeTe-based
materials (Fig. 5, A and B, and fig. S18C). The
spatial distribution of lattice strains in the
high-entropy Ge0.61Ag0.11Sb0.13Pb0.12Bi0.01Te
sample showed typical gathered regions on
the nanoscale, whereas the low-entropy GeTe
sample showed small fluctuations from normal
background noise. We counted the statistical
Fig. 4. Delocalizing the distribution of electrons. (A and B) STEM-HAADF images of the GeTe (A) and distribution of lattice strains to quantify the
high-entropy Ge0.61Ag0.11Sb0.13Pb0.12Bi0.01Te (B) samples. The orange rhombus indicates the Te atoms and effect of high entropy (main plots in Fig. 5,
the red arrow represents the deviation direction of the Ge atom. (C) Sketch map of the deviation value A and B, and fig. S18C), which indicated a
(d) from the geometry center. (D and E) Electrical field distributions of the the GeTe (D) and high-entropy broadened statistical distribution and larger
Ge0.61Ag0.11Sb0.13Pb0.12Bi0.01Te (E) samples. (F) Changed deviation values (d) with increasing entropy. root mean square s in the high-entropy sam-
(G and H) Charge distribution of the GeTe (G) and high-entropy Ge0.61Ag0.11Sb0.13Pb0.12Bi0.01Te (H) samples. ple compared with the low-entropy sample.
(I) Distance dependence of the charge density of the white-boxed region in (G) and (H). (J and K) Schematic The large lattice strains demonstrated that
diagrams of the localized (J) and delocalized (K) distribution of electrons calculated by density functional the symmetry of the crystal structure broke
theory calculations, The red solid line indicate short bonds and the red dotted lines indicate long bonds. down, which changed the distribution and
(L) Schematic diagram of band convergence by tuning the deviation value d. propagation of the phonons. We used the
Raman spectra to measure the changes in op-
tical phonons (Fig. 5C), which were the direct
surrounding Te sites was quite similar (Fig. 4H), verge, which should be the origin of the high expression of the changed atomic environment
demonstrating a delocalized distribution of PF over a wide temperature range. The de- and force constant (53). The formation of high-
electrons. With the incorporation of alloying localized electrons decrease the difference of entropy materials was expected to break the
elements (Fig. 4K), localized electrons could p-orbital bonding in different orientations, center of inversion symmetry in GeTe, result-
transfer from the shorter bonds to the longer resulting in a small band difference between ing in the development of a net dipole moment
bonds, resulting in delocalized electron distri- the light and heavy bands (23, 25). We cal- and the broadening of Raman peaks or new
bution and similar bond features. This phe- culated the band structures of low-entropy active Raman modes (54). The two intrinsic
nomenon was also observed by comparing the GeTe and high-entropy Ge5AgSbPbTe8 (fig. Raman peaks largely broadened and a new
charge densities between the Ge and Te sites S17), which illustrated the increased equiv- low-frequency mode at 60 cm−1 appeared when
(Fig. 4I). We also calculated the electron lo- alent degenerated valleys of the band structure the composition changed from GeTe to high-
calization function to show the distribution (Nv) by delocalizing electrons. The increased entropy Ge0.61Ag0.11Sb0.13Pb0.12Bi0.01Te (Fig.
of electrons (fig. S16), which demonstrated Nv will increase the effective mass of carriers 5C) because of the loss of center of inversion
the decreased value (delocalization) by form- by the term of m* = (Nv)2/3mb* and contrib- symmetry with different features of the Ge,
ing high-entropy composition. Because of the ute to a high Seebeck coefficient (20). The Ag, Sb, and Pb atoms (55). The vibrating
localized electrons around the shorter bonds, change in the band structure was also ex- atoms experienced distribution of chemical
there was a large difference in electron distri- pressed by our schematic diagram (Fig. 4L) environments, leading to fluctuations in the
butions between the shorter and longer bonds, and explained as slight symmetry breaking vibration frequencies. These Raman results
resulting in a split valence maximum. The well- by Pei et al. (20). suggested that the phonon mode split with the
distributed six bonds with similar features Overall, the increased crystal symmetry will breakdown of the center of inversion symmetry
enable the split valence maximums to con- increase lattice thermal conductivity because in the octahedron. Thus, the propagation of

RES EARCH | REPOR T
REFERENCES AND NOTES

1. J. Mao et al., Science 365, 495–498 (2019).
2. G. Tan, L. D. Zhao, M. G. Kanatzidis, Chem. Rev. 116,
12123–12149 (2016).
3. J. He, T. M. Tritt, Science 357, eaak9997 (2017).
4. B. Jiang et al., Science 371, 830–834 (2021).
5. B. Qin et al., Science 373, 556–561 (2021).
6. S. Roychowdhury et al., Science 371, 722–727 (2021).
7. G. J. Snyder et al., Adv. Mater. 32, e2001537 (2020).
8. K. Xia et al., Energy Environ. Sci. 12, 1568–1574 (2019).
9. H. Bai et al., Nat. Commun. 12, 7207 (2021).
10. T. R. Wei et al., Science 369, 542–545 (2020).
11. J. Dong et al., Energy Environ. Sci. 12, 1396–1403 (2019).
12. Z. Dong et al., Nat. Commun. 13, 35 (2022).
13. H. Liu et al., Nat. Mater. 11, 422–425 (2012).
14. X. Zhou et al., Mater. Today 21, 974–988 (2018).
15. X. L. Shi, J. Zou, Z. G. Chen, Chem. Rev. 120, 7399–7515 (2020).
16. N. Dragoe, D. Bérardan, Science 366, 573–574 (2019).
17. L. Hu et al., Adv. Energy Mater. 8, 1802116 (2018).
18. R. Liu et al., Adv. Mater. 29, 1702712 (2017).
19. B. Jiang et al., Mater. Today Phys. 5, 20–28 (2018).
20. J. Li et al., Joule 2, 976–987 (2018).
21. M. Hong et al., Adv. Mater. 30, 1705942 (2018).
22. Z. Zheng et al., J. Am. Chem. Soc. 140, 2673–2686 (2018).
23. M. Cagnoni, D. Führen, M. Wuttig, Adv. Mater. 30, e1801787
(2018).
24. M. K. Brod, M. Y. Toriyama, G. J. Snyder, Chem. Mater. 32,
9771–9779 (2020).
25. M. Wuttig, V. L. Deringer, X. Gonze, C. Bichara, J. Y. Raty,
Adv. Mater. 30, e1803777 (2018).
Fig. 5. Localized distribution of phonons. (A and B) Distribution of normal (A) and shear strains (B) in the 26. S. Lee et al., Nat. Commun. 5, 3525 (2014).
GeTe and high-entropy Ge0.61Ag0.11Sb0.13Pb0.12Bi0.01Te samples. Left images and green lines are from GeTe 27. Z. Liu et al., Proc. Natl. Acad. Sci. U.S.A. 115, 5332–5337 (2018).
28. S. Perumal et al., Joule 3, 2565–2580 (2019).
samples, and right images and pink lines are from the Ge0.61Ag0.11Sb0.13Pb0.12Bi0.01Te sample. The value of s is 29. D. Wu et al., J. Am. Chem. Soc. 136, 11412–11419 (2014).
the root mean square. (C) Raman intensity and full width at half maximum (FWHM) of the GeTe (left) and 30. M. Hong et al., Joule 4, 2030–2043 (2020).
high-entropy Ge0.61Ag0.11Sb0.13Pb0.12Bi0.01Te (middle) samples. (D) Measured sound velocities of the transverse 31. J. Shuai, Y. Sun, X. Tan, T. Mori, Small 16, e1906921 (2020).
32. X. Xu, L. Xie, Q. Lou, D. Wu, J. He, Adv. Sci. (Weinh.) 5, 1801514
(nt) and longitudinal (nl) waves and the calculated Gruneisen parameter (g) and lattice thermal conductivity (kL) (2018).
values at room temperature, where n0 is the measured sound velocity of the GeTe sample. 33. T. Xing et al., Adv. Mater. 33, e2008773 (2021).
34. T. Xing et al., Energy Environ. Sci. 14, 995–1003 (2021).
35. Z. Bu et al., Energy Environ. Sci. 14, 6506–6513 (2021).
heat between the split phonon modes was more quickly than the longitudinal sound 36. G. Bai et al., Adv. Energy Mater. 11, 2102012 (2021).
scattered because of the different vibrational velocity nl with an increasing degree of dis- 37. P. Jood, M. Ohta, A. Yamamoto, M. G. Kanatzidis, Joule 2,
1339–1355 (2018).
frequencies. This phenomenon of broaden- order (entropy), illustrating more softened 38. X. Hu et al., Energy Environ. Sci. 9, 517–529 (2016).
ing and excessive phonon modes could be transverse phonons compared with the lon- 39. B. Jiang et al., Energy Environ. Sci. 13, 579–591 (2020).
explained by the (quasi)localized phonons, gitudinal system. Because the Gruneisen pa- 40. Q. Zhang et al., Energy Environ. Sci. 10, 956–963 (2017).
41. C. Fu et al., Nat. Commun. 6, 8144 (2015).
which arose from phonon damping related to rameter (g) was closely proportional to the 42. P. Qiu et al., Joule 3, 1538–1548 (2019).
structural disorder (56, 57). Unlike selective ratio of nl/nt, the decrease in nt from the soft- 43. Y. Xing et al., Joule 4, 2475–2483 (2020).
Anderson electron localization, the phonon ened transverse phonons would increase g 44. Y. Xing et al., Energy Environ. Sci. 12, 3390–3399 (2019).
45. B. Jiang et al., Nat. Commun. 12, 3234 (2021).
localization was nonselective in the disordered (60). Thus, the anharmonicity demonstrat- 46. T. Zhu, L. Hu, X. Zhao, J. He, Adv. Sci. (Weinh.) 3, 1600004
system. The holes were transported mainly in ing the interactions among the different (2016).
the Te sublattice in the p-type GeTe, whereas branches of phonons would be largely strength- 47. D. T. Morelli, V. Jovovic, J. P. Heremans, Phys. Rev. Lett. 101,
035901 (2008).
the phonons were definitely transported in ened (Fig. 5D), resulting in depressed and
48. D. Di Sante, P. Barone, R. Bertacco, S. Picozzi, Adv. Mater. 25,
both the Ge and Te sublattices. temperature-independent lattice thermal 509–513 (2013).
Phonon damping typically would strengthen conductivity (47). 49. W. Gao et al., Nature 575, 480–484 (2019).
scattering for phonons and decrease the mean In conclusion, we promoted the TE per- 50. N. Shibata et al., Nat. Phys. 8, 611–615 (2012).
51. K. Muller et al., Nat. Commun. 5, 1–8 (2014).
free path. In a high-entropy system with large formance of GeTe-based materials and mod- 52. G. J. Snyder, E. S. Toberer, Nat. Mater. 7, 105–114 (2008).
disorder, phonon damping should be suffi- ules by forming a high-entropy composition. 53. A. Pawbake et al., Phys. Rev. Lett. 122, 145701 (2019).
ciently strong, resulting in a small mean free The distribution of electrons was highly de- 54. S. Acharya, J. Pandey, A. Soni, Appl. Phys. Lett. 109, 133904
(2016).
path comparable to the wavelength and ap- localized because of the increased crystal
55. K. S. Andrikopoulos, S. N. Yannopoulos, A. V. Kolobov, P. Fons,
proaching the Loffe-Regel limit (58). Phonons symmetry driven by increasing entropy in the J. Tominaga, J. Phys. Chem. Solids 68, 1074–1078 (2007).
with frequencies above the Loffe-Regel limit disordered system, which maintained good 56. H. Shintani, H. Tanaka, Nat. Mater. 7, 870–877 (2008).
should be substantially damped and no longer electrical properties. The localized phonons 57. S. Ren et al., Nat. Commun. 12, 5755 (2021).
58. J. E. Graebner, B. Golding, L. C. Allen, Phys. Rev. B 34,
propagate (56). A disordered structure should from the disorder resulted in a strong damping
5696–5701 (1986).
dampen transverse phonons, but not longitu- effect on the propagation of transverse pho- 59. X. Y. Li et al., Phys. Rev. Lett. 124, 225902 (2020).
dinal phonons, because the transverse rather nons, which was responsible for the strength- 60. T. T. Jia, G. Chen, Y. S. Zhang, Phys. Rev. B 95, 155206 (2017).
than the longitudinal phonon mode would ened anharmonicity and depressed lattice 61. Y. Wu et al., Joule 3, 1276–1288 (2019).
62. P. Zong et al., Energy Environ. Sci. 10, 183–191 (2017).
be more sensitive to disorder (59). As a re- thermal conductivity. The demonstrated
sult, disordered systems typically show softer strategy of delocalized electrons and local- AC KNOWL ED GME NTS
transverse phonon modes compared with lon- ized phonons in the high-entropy materials Funding: We thank the SUSTech Core Research Facilities for
gitudinal systems. In this work, we measured provided the possibility of simultaneously use of the STEM instrument. This work was supported by
the Guangdong Province Program (grant no. 00201517), the
the sound velocities of all samples (Fig. 5D). optimizing the electrical and thermal trans- Guangdong-Hong Kong-Macao Joint Laboratory (grant no.
The transverse sound velocity nt decreased port properties. 2019B121205001), the National Natural Science Foundation of

RES EARCH | REPOR T
China (grant nos. 52002167, 11874194, and 11934007), the the Raman measurements. B.J., W.W., and J.H. wrote the SUPPLEMENTARY MATERIALS
Science and Technology Innovation Committee Foundation manuscript, which was edited by all authors. Competing science.org/doi/10.1126/science.abq5815
of Shenzhen (grant nos. KQTD2016022619565991 and interests: The authors declare no competing interests. Data and Materials and Methods
ZDSYS20141118160434515), and special funds (grant no. materials availability: All data are available in the manuscript Figs. S1 to S19
G02206302). Author contributions: B.J. and J.H. designed or the supplementary materials. License information: Copyright Tables S1 to S3
this work. B.J. and S.L. synthesized the samples, fabricated the © 2022 the authors, some rights reserved; exclusive licensee References (63–87)
modules, and performed the transport property measurements. American Association for the Advancement of Science. No claim to
W.W. performed the STEM observations and analysis. Y.W. and original US government works. https://www.science.org/about/ Submitted 18 April 2022; accepted 1 June 2022
L.X. performed the calculations. C.W., Y.C., and M.H. performed science-licenses-journal-article-reuse 10.1126/science.abq5815

RES EARCH
CHEMISTRY merization in water. It is combined with the

development of two integrated RPTs (gel-sol-
Dilution-induced gel-sol-gel-sol transitions gel-sol) upon dilution with the supramolecular
units shown in Fig. 1.
by competitive supramolecular pathways in water To elucidate the pathways that direct the
sol-gel transition upon dilution, we investi-
Lu Su1,2†, Jesús Mosquera1,3†, Mathijs F. J. Mabesoone1, Sandra M. C. Schoenmakers1, gated the interaction between BTA-EG4 and
Cyprien Muller1, Marle E. J. Vleugels1, Shikha Dhiman1, Stefan Wijker1, a library of surfactants at micromolar concen-
Anja R. A. Palmans1, E. W. Meijer1,4* trations (figs. S10 to S16). We illustrate our
detailed explorations using a combination of
Fascinating properties are displayed by synthetic multicomponent supramolecular systems that spectroscopic, cryo-TEM, light scattering, iso-
comprise a manifold of competitive interactions, thereby mimicking natural processes. We present thermal titration calorimetry, and theoretical
the integration of two reentrant phase transitions based on an unexpected dilution-induced assembly techniques (Fig. 2 and figs. S1 to S5, S17 to S21)
process using supramolecular polymers and surfactants. The co-assembly of the water-soluble by taking the cationic surfactant cetrimonium
benzene-1,3,5-tricarboxamide (BTA-EG4) and a surfactant at a specific ratio yielded small-sized bromide (CTAB, CMC ≈ 1 mM) as an example.
aggregates. These interactions were modeled using the competition between self-sorting and Nuclear magnetic resonance (NMR) and cryo-
co-assembly of both components. The small-sized aggregates were transformed into supramolecular TEM measurements of BTA-EG4 (250 mM)
polymer networks by a twofold dilution in water without changing their ratio. Kinetic experiments were performed in the presence of 0, 1, 2, and
show the in situ growth of micrometer-long fibers in the dilution process. We were able to create 3 eq. of CTAB (Fig. 2, A and B). As a result of
systems that undergo fully reversible hydrogel-solution-hydrogel-solution transitions upon dilution by the formation of micrometer-sized polymers
introducing another orthogonal interaction. in D2O, no resonances from BTA-EG4 were
observed in the NMR, whereas the sample
T
containing 250 mM CTAB showed sharp and
he design of interactive and adaptive soft transitions in molecular systems with many well-split peaks. CTAB peaks disappeared when
materials requires a complex composi- components under thermodynamic equilibrium mixed in equimolar ratio with BTA-EG4 in
tion of chemical components for which has been presented by Jacobs and Frenkel (18). NMR, and cryo-TEM confirmed the presence of
nature is often a source of inspiration (1). Their statement—that only minor adjustments micrometer-long aggregates (Fig. 2B). Together,
Multicomponent systems formed by to the strengths of intermolecular interactions these results indicate that all the surfactants
multiple associative and dissociative inter- are required to regulate the formation of dif- are integrated in the polymers. However, when
actions are key signatures of biological mat- ferent domains with specific compositions—is more equivalents of CTAB were applied, we
ter, but the competitive interplay of all these relevant for our current study. observed a gradual increase in intensity and
interactions makes the understanding of cer- Two frequently used units to construct resolution of the peaks associated with BTA-
tain natural processes difficult (2, 3). Synthetic aqueous synthetic supramolecular systems EG4. At 2 eq. of CTAB, the ethylene glycol peak
systems, because of their simplicity, can un- are surfactants and supramolecular mono- of BTA-EG4 appeared; 1H diffusion-ordered
ravel some of these challenges while yielding mers (6, 19–24). Monomer BTA-EG4 under- spectroscopy (DOSY) NMR (fig. S17) revealed
new functionalities (4–7). Although this field goes supramolecular polymerization in water that all the surfactants were still embedded,
is still in its infancy, several examples have to afford one-dimensional aggregates at micro- but in much smaller assemblies. These smaller
highlighted its potential for applications in molar concentration driven by hydrogen bonding aggregates were confirmed by cryo-TEM im-
catalysis, electronics, and medicine (8–11), with and hydrophobic effects, and is made water- ages showing fibers less than 500 nm in length.
the development of the lipid nanoparticle– soluble by hydrophilic tetraethylene glycol At 3 eq. of CTAB, both signals corresponding to
encapsulated mRNA vaccine as an obvious groups (Fig. 1A and fig. S6) (24). Dynamic the short BTA-EG4/CTAB fibers and BTA-
successful highlight in this respect (12). hydrogels are formed at concentrations above EG4/CTAB micelles were observed in DOSY
Liquid-liquid phase separations are often 8 mM (figs. S7 and S8) (25). The cationic surfactant NMR; cryo-TEM showed small spherical aggre-
accompanied by the occurrence of reentrant octyltrimethylammonium bromide (OTAB) gates and occasionally short fibers. Ultraviolet-
phase transitions (RPTs), which cause fasci- forms spherical micelles with a hydrodynamic visible (UV-vis) spectroscopy confirmed fiber
nating phenomena in these multicomponent diameter ~2 nm above its critical micelle con- disruption and the formation of small BTA
systems. Here, changes in temperature and/or centration (CMC = ~297 mM at 20°C) (26). An aggregates (Fig. 2C). This transition was de-
addition of one of the components give rise to unexpected finding sparked our attention pendent on the overall concentration as well
a recurrence of the original phase (11, 13, 14). when studying orthogonality by combining as the ratio of both components (Fig. 2D), with
This phenomenon was recently found to play these two units. We observed that the addition more surfactants needed at lower BTA con-
a critical role in the formation of intracellular of OTAB provoked the full disruption of the centrations. The saturation of the polymer
membraneless organelles, which are mainly BTA-EG4 hydrogel. Surprisingly, the result- with surfactants destabilized the structure
systems based on liquid-liquid phase sepa- ing solution was transformed into a trans- and destroyed the fibers at high concentra-
ration (15–17). A broader perspective on phase parent hydrogel upon dilution, followed by tions of surfactant, most probably as a result of
reentering the solution phase with further electrostatic and/or geometric constraints in
dilution, giving rise to a supramolecular sol- the structures formed (Fig. 2E).
1
Institute for Complex Molecular Systems, Laboratory of gel-sol RPT in water. The first two states were Studies with other surfactants revealed three
Macromolecular and Organic Chemistry, Eindhoven University visualized by cryogenic transmission electron relevant conclusions: (i) The affinity between
of Technology, 5600 MB Eindhoven, Netherlands. 2Leiden
microscopy (cryo-TEM) (Fig. 1B) and small- the surfactant and BTA-EG4 is determined
Academic Centre for Drug Research, Leiden University,
2333 CC Leiden, Netherlands. 3Centro de Investigacións angle x-ray scattering (SAXS) (fig. S9). This by the length of the aliphatic chain of the sur-
Científicas Avanzadas, Universidade da Coruña (CICA), A process was fully reversible by increasing the factant, (ii) BTA-EG4 fibers are unstable be-
Coruña, Spain. 4School of Chemistry and RNA Institute, concentration and led us to investigate the low the CMC of the surfactants, and (iii) the
University of New South Wales, Sydney, NSW 2052, Australia.
*Corresponding author. Email: e.w.meijer@tue.nl underlying mechanism of this hitherto unde- transition is dependent on both the overall
†These authors contributed equally to this work. scribed dilution-induced supramolecular poly- concentration and the ratio of the components.
Su et al., Science 377, 213–218 (2022) 8 July 2022 1 of 5

RES EARCH | REPOR T
A uses the intrinsic property of a polymeric sur-

OH O factant to lead to further aggregation at higher
O O
O concentration, whereas the second strategy
=
=
HN O
OH
takes advantage of an orthogonal supramo-
=
O
O
H
N
O lecular double network with a different gela-
NH O O
O
CTAB tion concentration window.
O
UPy-EG11 The polymeric surfactant, octyl polyethylene
O O BTA-EG4 OTAB OPEG
O glycol (OPEG, 2 kDa; Fig. 1A and figs. S22 and
HO S23) was synthesized with a C8-spacer. The
interaction of OPEG with BTA-EG4 at micro-
molar concentration was evaluated by UV-vis,
NMR, and cryo-TEM, showing a behavior sim-
ilar to that observed for CTAB (figs. S24 and
S25). Interestingly, an optimized BTA-EG4:
OPEG molar ratio of 2:3 ([BTA-EG4] = 80 mM)
B in Milli-Q (MQ)–H2O gave rise to an opaque
Gel 1 with a mesh size of ~10 to 20 nm as
indicated by cryo-TEM (Fig. 3A) and SAXS
(fig. S36). The densely packed network was
composed of wormlike fibers of fused BTA-
EG4/OPEG micelles. Rheological experiments
5 min
showed a storage modulus, G′, of up to 2 kPa
Dilution with a complex viscosity, h*, above 2 kPa·s, as
well as a fast and near-complete self-healing
behavior (Fig. 3, C to E). When diluting this gel
with water to [BTA-EG4] = 35 mM, a solution
was obtained after equilibrium, owing to the
50 nm 50 nm dissociation of the fused micelles as depicted
by cryo-TEM and SAXS. In line with the mor-
phology transition, h* dropped to ~1 Pa·s. When
C
Dilution this solution was further diluted to [BTA-EG4] =
27 mM, which is below the CMC of OPEG
Orthogonal (~42 mM; fig. S26), a transparent and dynamic
Strategy 1 Gel Sol Gel Sol Driving Forces Gel 2 was formed with G′ of ~5 Pa and h* of
~10 Pa·s. This gelation was triggered by the
Strategy 2 dissociation/release of OPEG from the BTA-
Sol Gel Sol Orthogonal EG4/OPEG micellar complex, which enabled
Networks the elongation of the BTA fibers and led to the
Gel Sol
formation of the typical, entangled BTA-EG4
fiber network. With further dilution to [BTA-
Fig. 1. Dilution-induced gelation in water. (A) Supramolecular components and their individual assemblies
EG4] = 0.8 mM and still the same BTA-EG4:
in water. (B) Cryo-TEM images of dilution-induced gelation in a BTA-EG4/OTAB system. Left: A solution
OPEG molar ratio of 2:3, the hydrogel changed
of BTA-EG4 (4 wt %, 31 mM) and OTAB (6.5 eq., 202 mM) shows small aggregates with a diameter of 5 to
into a viscous solution showing h* of ~0.4 Pa·s
8 nm, suggesting the presence of micellar BTA-EG4/OTAB complexes. Right: A transparent hydrogel
and micrometer-long individual fibers. A de-
of BTA-EG4 (2 wt %) and OTAB (101 mM), generated upon twofold dilution with MQ-H2O, shows an entangled
tailed characterization of this gel-sol-gel-sol sys-
BTA-EG4 fiber network. Scale bar, 50 nm. (C) Two strategies for dilution-induced gel-sol-gel-sol transitions
tem (cryo-TEM, SAXS, and rheology) is presented
by competitive supramolecular pathways in water.
in figs. S27 to S36.
The good biocompatibility of BTA-EG4 (27)
From these results, we hypothesize that the sol- characteristic peaks (211 and 227 nm) in UV- provides further opportunity for application
gel transition upon dilution is driven by the vis spectra related to fiber structure appeared of this multicomponent system in more com-
concentration-dependent complexation of sur- (Fig. 2H), as corroborated by cryo-TEM (fig. S21). plex media. Phosphate-buffered saline and
factants to BTA-EG4 polymers. To corroborate We hypothesize that CTAB is released when DMEM/F12 basal medium were used for the
this hypothesis, we constructed a thermody- the overall concentration decreases, there- whole process, showing RPT behavior iden-
namic mass-balance model that describes com- by enabling the elongation of the BTA fiber to tical to that observed in MQ-H2O (fig. S37). Ad-
petition among supramolecular polymerization, eventually achieve micrometer-long structures. ditionally, when physiological temperature
micellization, complexation of CTAB to the The dilution-induced supramolecular polym- was applied, G′ of Gel 1 became reduced
polymers, and polymer collapse (Fig. 2, E and erization became faster upon adding more by a factor of 8 (G′ ~ 250 Pa), probably owing
F, figs. S1 to S5, and supplementary text). We water (fig. S20). to a fastened dissociation of BTA-EG4/OPEG
subsequently studied the dilution-induced With an understanding of the mechanisms micelles (fig. S38).
effect of a solution of BTA-EG4 with 3 eq. of underlying the dilution-induced supramolec- The second strategy relies on an orthogonal
CTAB, in which only small aggregates exist. ular polymerization in water, an integration of supramolecular network formed by water-soluble
Upon dilution, peaks corresponding to both two RPTs was developed by introducing an supramolecular polymers based on the ureido-
BTA-EG4 and CTAB gradually broadened additional orthogonal interaction through two pyrimidinone unit (UPy-EG11, Fig. 1A), which
and weakened in NMR (Fig. 2G) while the different strategies (Fig. 1C). The first strategy is less dynamic than BTA-EG4 polymers (28).

RES EARCH | REPOR T
A B 0 eq. 1 eq. 2 eq. 3 eq .

CTAB
3 eq. CTAB
BTA
2 eq. CTAB +
CTAB
1 eq. CTAB
BTA
100 nm 100 nm 100 nm 100 nm
3.5 3.0 2.5 2.0 1.5 1.0
Chemical shift (ppm)
C 1.0 D E M/S P/S

[BTA-EG4] = 250 M 1.0
500 M KMM
0 eq. CTAB 400 M
1 eq. CTAB 250 M
2 eq. CTAB 100 M
Abs
3 eq. CTAB
P/S
0.5 0.5
KP/S
KMic Mic Ke
0.0
0.0
200 225 250 275 300 325 350 0 1 2 3 4 5 6 7
Mic S M P
Wavelength (nm) Surfactant equivalents
F 1.0 G H 1.0
Model 1 Model 2 No dilution

0.8 P/S P/S 2-fold dilution
Normalized Abs
M/S M/S 5-fold dilution 5-fold dilution
BTA
Fraction
0.6
2-fold dilution
+
3 eq. 0.5
0.4 CTAB
No dilution
0.2
CTAB
0.0 0.0
0 1 2 3 4 5 6 3.5 3.0 2.5 2.0 1.5 1.0 200 225 250 275 300 325 350
Surfactant equivalents Chemical shift (ppm) Wavelength (nm)
Fig. 2. Interaction and dilution-induced assembly of BTA-EG4 and CTAB in monomer, BTA-EG4 polymer, free surfactant, and surfactant micelle, respectively,
water. (A to C) 1H NMR spectra (400 MHz, D2O; peaks corresponding to BTA and and Ki and si the equilibrium constant and cooperativity parameter of process i.
CTAB are in blue and yellow zones, respectively) (A), cryo-TEM images (scale bar, (F) Simulated fraction curves of P/S and M/S with 1 eq. (model 1) and 2 eq.
100 nm) (B), and UV-vis spectra (C) of BTA-EG4 (250 mM) in the presence of (model 2) of [CTAB]/[BTA-EG4] accommodations, in which the concentration of
0, 1, 2, and 3 eq. of CTAB, respectively. (D) Experimental fraction curves of BTA-EG4 was held constant (250 mM) while the surfactant content was varied.
BTA-EG4/CTAB polymer obtained by normalizing the UV absorbance at 227 nm. The (G and H) 1H NMR spectra (400 MHz, D2O; peaks corresponding to BTA and
concentration of BTA-EG4 was held constant (500, 400, 250, or 100 mM, CTAB are in blue and yellow zones, respectively) (G) and UV-vis spectra
respectively) while CTAB content was varied. (E) Thermodynamic model for the (H) of [BTA-EG4]/[CTAB] (1/3) upon dilution, with [BTA-EG4]initial = 250 mM.
cooperative, competitive pathways, with M, P, S, and Mic representing BTA-EG4 CTAB absorption in all UV-vis spectra was removed for clarity.
In water, UPy-EG11 gives rise to micrometer- consisting of different gelation windows. As factant, a transparent solution with a low h*
long bundled fibers with a width of ~5 to 14 nm. depicted in Fig. 3B, F to H, an opaque hydrogel of ~1 Pa·s was accomplished. Further dilution
UPy-EG11/CTAB interactions were confirmed Gel 1′ was fabricated with a BTA-EG4:UPy- yielded the sol-gel-sol transition as in the first
by 1H NMR (fig. S39). However, UPy-EG11 and EG11:OTAB molar ratio of 3.2:1:21.2 at [BTA- strategy. All results of the detailed character-
UPy-EG11/CTAB mixtures (up to 5 eq. of CTAB) EG4] of 93 mM in MQ-H2O. Gel 1′ is mainly ization are presented in figs. S40 to S43.
showed almost identical UV spectra and cryo- stabilized by the UPy-EG11 network, whereas After exploring the different states at specific
TEM images, indicating that CTAB binds with BTA-EG4 exists as BTA-EG4/OTAB micelles. concentrations, we tested methods for tracking
UPy-EG11 bundles yet does not markedly dis- Upon dilution to [BTA-EG4] = 46.7 mM, the the formation of these polymers upon dilu-
rupt the supramolecular polymer. When mixed, UPy-EG11 network was not dense enough to tion. We used total internal reflection fluores-
BTA-EG4 and UPy-EG11 form self-sorted construct a three-dimensional network that cence microscopy (TIRFM) to trace the in situ
homopolymers. Hence, BTA-EG4, UPy-EG11, tightly held all the water. As the length of the formation of the supramolecular polymeric
and a surfactant make an ideal combination to BTA-EG4 fibers was not sufficient to form a fibers after the addition of droplets of con-
construct an interpenetrating double network network because of interaction with the sur- centrated solution to an aqueous film (Fig. 4).

RES EARCH | REPOR T
Fig. 3. Dilution-induced gel-sol-gel-sol

transitions in water by using a two-
component (BTA-EG4/OPEG) or three- Gel Sol Gel Sol
component (BTA-EG4/UPy-EG11/
OTAB) system, respectively. (A) Cryo-
A
TEM images and schematic illustrations
of gel-sol-gel-sol transitions of a BTA-
EG4/OPEG system with their macro-
scopic images and molecular-level illus-
trations as insets above and below.
(B) Schematic illustrations of gel-sol-gel-
sol transitions of a BTA-EG4/UPy-EG11/ 100 nm 100 nm 100 nm 100 nm
OTAB system, displayed as in (A).
(C and F) Complex viscosity points (w =
1 rad/s, 1% strain, 20°C) corresponding
to the gel-sol-gel-sol transitions in
the BTA-EG4/OPEG system (C) and the
BTA-EG4/UPy-EG11/OTAB system (F),
with the dashed line fitted through the
cubic spline curve. (D and G) Frequency-
dependent oscillatory rheology (1% Gel 1 Sol 1 Gel 2 Sol 2
strain, 20°C) of the BTA-EG4/OPEG
system (D), showing that Gel 1 ([BTA- B
EG4] = 80 mM) has a stronger network
than Gel 2 ([BTA-EG4] = 27 mM),
and the BTA-EG4/UPy-EG11/OTAB sys-
tem (G), showing that Gel 1′ ([BTA-EG4] =
93 mM) has a stronger network
than Gel 2′ ([BTA-EG4] = 15.5 mM).
(E and H) Step-strain measurements
Gel 1’ Sol 1’ Gel 2’ Sol 2’
with applied oscillatory strain alternated
between 1% and 1000% for 30-s C D E
periods (w = 1 rad/s, 20°C) in Gel 1 1%
of the BTA-EG4/OPEG system (E)
and Gel 1′ of the BTA-EG4/UPy-EG11/
OTAB system (H).
1000 %
F G H
1%
1000 %
(See figs. S44 to S46 for details of the experi- time. Some of the fibers became many tens respectively—a phenomenon similar to the
mental setup.) When an aqueous droplet of of micrometers long (Fig. 4A). In some cases, dynamic interconnected networks formed by
[BTA]:[CTAB] (1:3, [BTA] = 250 mM), contain- the fibers were connected to other droplets Marangoni flows in the mesoscale position-
ing 5% BTA-Cy3 for visualization, was pipetted (Fig. 4C). In other cases, the fibers both grew ing of amphiphiles when extruded from the
into the aqueous film and thus 100-fold diluted and shrank, and the pulling of droplets by the droplets (29). These results emphasize the crit-
in situ, supramolecular fibers started to grow contracting fibers could be observed (Fig. 4D ical role of concentrations in multicomponent
at several positions out of droplets (Fig. 4, A and movies S1 to S4). We ascribe the growth systems and are useful to understand several
to C). The relatively slow process in thin and the shrinkage to concentration gradients phenomena in biology, where RPTs are critically
films enabled us to follow the growth over from the droplet and against the droplet, important to understand cellular condensates.

RES EARCH | REPOR T
11. F. Helmich et al., Angew. Chem. Int. Ed. 49, 3939–3942

A 5 min B 0 min 3 min (2010).
12. N. Pardi, M. J. Hogan, F. W. Porter, D. Weissman, Nat. Rev.
Drug Discov. 17, 261–279 (2018).
13. E. A. Clark, J. E. G. Lipson, Polymer 53, 536–545 (2012).
14. T. M. Hermans et al., Nat. Nanotechnol. 4, 721–726 (2009).
11 min 15. P. R. Banerjee, A. N. Milin, M. M. Moosa, P. L. Onuchic,
A. A. Deniz, Angew. Chem. Int. Ed. 56, 11354–11359 (2017).
6 min 8 min 16. J. A. Riback et al., Nature 581, 209–214 (2020).
17. Y. Shin, C. P. Brangwynne, Science 357, eaaf4382 (2017).
18. W. M. Jacobs, D. Frenkel, Biophys. J. 112, 683–691 (2017).
12 min 19. R. J. Brea et al., J. Am. Chem. Soc. 140, 17356–17360
(2018).
20. M. P. Hendricks, K. Sato, L. C. Palmer, S. I. Stupp, Acc. Chem. Res.
50, 2440–2448 (2017).
10 min 13 min 21. J. Kang et al., Science 347, 646–651 (2015).
22. G. González-Rubio et al., Science 368, 1472–1477 (2020).
15 min 23. A. Heeres et al., J. Am. Chem. Soc. 125, 14252–14253
(2003).
24. C. M. A. Leenders et al., Chem. Commun. 49, 1963–1965
(2013).
25. E. Vereroudakis et al., ACS Cent. Sci. 6, 1401–1411 (2020).
C 6 min D 11 min 12 min 13 min 26. R. Zieliński, J. Colloid Interface Sci. 235, 201–209 (2001).
27. S. Varela-Aramburu et al., Biomacromolecules 21, 4105–4115
(2020).
28. S. I. S. Hendrikse et al., Chem. Commun. 53, 2279–2282
(2017).
29. A. van der Weijden, M. Winkens, S. M. C. Schoenmakers,
W. T. S. Huck, P. A. Korevaar, Nat. Commun. 11, 4800 (2020).
8 min AC KNOWL ED GME NTS

We thank D. Frenkel, T. de Greef, R. Cardinaels, and P. Korevaar for
discussions. Funding: Supported by NWO (TOP-PUNT grant
14 min 17 min 18 min 10018944); the Dutch Ministry of Education, Culture and Science
(Gravitation program 024.001.035); and the European Commission
(SYNMAT-788618-1 and H2020-MSCA-IF-2017-794016). C.M.
gratefully thanks the French National Research Agency (ANR-17-
11 min EURE-0016). Author contributions: J.M. and C.M. found the effect
serendipitously and performed the first series of experiments. J.M.,
L.S., and E.W.M. designed the study, conceived the overall set
of experiments, and wrote the manuscript. L.S., M.E.J.V., S.D., and
S.M.C.S. contributed to all analyses and the synthesis of the
components. S.W. and L.S. did the SAXS and S.M.C.S. performed
all the cryo-TEM experiments. S.D. and L.S. performed all the
Fig. 4. Kinetic in situ formation of fibers upon dilution. (A to C) TIRFM images from movies S1 and S2 upon optical microscopy experiments and M.F.J.M. did the modeling. All
addition of BTA-EG4/CTAB droplets to a film of PBS over time. The droplets in time adhere to the surface authors critically revised the manuscript, and they all approve the
manuscript. Competing interests: The authors declare no
and fibers are formed after several minutes, using 1% BTA-biotin. The green fluorescence is the result of 5% competing interests. Data and materials availability: All data are
BTA-Cy3; the higher the intensity, the higher the concentration of BTA. The increase of the droplet size results available in the main text or the supplementary materials.
from an increased adhesion of the droplet to the surface. The increase in length of the fibers is the result of License information: Copyright © 2022 the authors, some rights
reserved; exclusive licensee American Association for the
dilution-induced self-assembly/polymerization by diffusion of aggregates from the densely packed droplet to the Advancement of Science. No claim to original US government works.
continuous water film. (D) Snapshots from movies S3 and S4 of the shrinkage (white arrow) and connection www.science.org/about/science-licenses-journal-article-reuse
(yellow arrow) of the fibers due to concentration gradients. Scale bar, 5 mm.
science.org/doi/10.1126/science.abn3438
Materials and Methods
RE FE RENCES AND N OT ES 6. T. Aida, E. W. Meijer, S. I. Stupp, Science 335, 813–817 Supplementary Text
(2012). Figs. S1 to S46
1. R. Merindol, A. Walther, Chem. Soc. Rev. 46, 5588–5619 (2017).
7. E. Mattia, S. Otto, Nat. Nanotechnol. 10, 111–119 (2015). References (30–39)
2. K. Simons, E. Ikonen, Nature 387, 569–572 (1997).
8. Q. Wan, W. P. To, X. Chang, C. M. Che, Chem 6, 945–967 Movies S1 to S4
3. W. Baumeister, J. Walz, F. Zühl, E. Seemüller, Cell 92, 367–380
(2020). MAPLE Files: Derivation_model_1eq.mw, Derivation_model_2eq.mw
(1998).
MATLAB Scripts and Function
4. G. Vantomme, E. W. Meijer, Science 363, 1396–1397 (2019). 9. S. Li et al., J. Am. Chem. Soc. 140, 10794–10802 (2018).
5. J.-F. Lutz, J.-M. Lehn, E. W. Meijer, K. Matyjaszewski, Nat. Rev. Mater. 10. D. A. Wilson, R. J. M. Nolte, J. C. M. van Hest, Nat. Chem. 4, Submitted 19 November 2021; accepted 11 May 2022
1, 16024 (2016). 268–274 (2012). 10.1126/science.abn3438

RES EARCH
PALEONTOLOGY are connected by short, parallel cross-linkages

forming a lattice (Fig. 3F and fig. S3I). The
Ultrastructure reveals ancestral vertebrate microfibrils are extremely fine, best measured
with transmission electron microscopy to be
pharyngeal skeleton in yunnanozoans 12 ± 2 nm in diameter (N = 16) (Fig. 3, G and
H). Carbonaceous particles locally aggregate
Qingyi Tian1,2, Fangchen Zhao2*, Han Zeng2, Maoyan Zhu2,3, Baoyu Jiang1* forming a “beads-on-a-string” appearance
with 50- to 60-nm intervals (Fig. 3H).
Pharyngeal arches are a key innovation that likely contributed to the evolution of the jaws and braincase We sought and found the first branchial
of vertebrates. It has long been hypothesized that the pharyngeal (branchial) arch evolved from an arch, lateral to a stout, dark circumoral struc-
unjointed cartilaginous rod in vertebrate ancestors such as that in the nonvertebrate chordate ture (description is given in the supplementary
amphioxus, but whether such ancestral anatomy existed remains unknown. The pharyngeal skeleton of text section of the supplementary materials) at
controversial Cambrian animals called yunnanozoans may contain the oldest fossil evidence constraining the rostral end of the body. The first branchial
the early evolution of the arches, yet its correlation with that of vertebrates is still disputed. By arch is often fully or partly obscured by the
examining additional specimens in previously unexplored techniques (for example, x-ray microtomography, underlapping circumoral structure, but close
scanning and transmission electron microscopy, and energy dispersive spectrometry element mapping), inspection reveals that the first arch, identical
we found evidence that yunnanozoan branchial arches consist of cellular cartilage with an extracellular matrix to the remaining arches, has bamboo-like seg-
dominated by microfibrils, a feature hitherto considered specific to vertebrates. Our phylogenetic analysis ments delineated by first-order septa (fig. S1,
provides further support that yunnanozoans are stem vertebrates. F, L, and M), attached gill filaments (fig. S1,
D, F, and H), and microfibrils in its matrix
D
(fig. S3, C and D).
ebate over the origin and early evolu- S1B) (10). The stacked discoid structures in the The seven branchial arches are aligned sub-
tion of vertebrates revolves around the bamboo-like bar were compared to those in parallel to each other and slightly inclined
pharynx (1–4), but there is little high- the cellular cartilages of embryonic verte- toward the rostral end (Fig. 1, C and F). The
resolution information available in the brates (8, 10), but this claim lacks further arches are connected by dorsoventrally curved,
critical part of the vertebrate tree—its anatomical support (12). The arches attach parenthesis-shaped horizontal rods at both
stem (5). Yunnanozoans from the early Cam- to slightly bent horizontal rods at both their ends (Fig. 1, F to H, and figs. S4 and S5; see
brian Chengjiang fauna (~518 million years dorsal and ventral ends (8, 11). The dorsal also figure 16 in (13); measurements are given
old) of China (6) are the oldest relatives of and ventral horizontal rods were previously in supplementary text). These dorsal and
crown-group vertebrates, although their phy- interpreted as blood vessels, margins of the ventral horizontal rods run subparallel to
logenetic position remains sharply debated— endostyle trough, or simply folds (8, 11, 12, 14). each other through most of the length of the
they have been variously classified as stem These debates call for further study based on pharynx then gradually converge between the
vertebrates, cephalochordates, hemichordates, new specimens. Here, we reexamine the pha- seventh pair of arches and the gonads (fig.
or even stem deuterostomes on the basis of ryngeal skeleton of yunnanozoans on the basis S5H). Locally, the rods show poorly preserved,
disputed anatomical interpretations (7–13). of 127 newly collected specimens and provide putative bamboo-like structures (figs. S4D and
Three species have been erected in the clade, high-resolution anatomical and ultrastructural S5, F and G). No gill filaments occur on the
Yunnanozoon lividum Hou, Ramsköld, and correlation with that of vertebrates. rods, but their matrix does contain microfib-
Bergström, 1991 (7), Haikouella lanceolata Chen, Newly collected specimens confirm that the rils, as in the branchial arches (fig. S3S). The
Huang, and Li, 1999 (10), and H. jianshanensis branchial arches comprise bamboo-like bars branchial bars articulate with the horizontal
Shu et al., 2003 (11), but the latter two species and lanceolate gill filaments (Fig. 1, D to F, rods at joints, evidenced by the common pres-
were subsequently revised as junior synonyms and fig. S1, A and B). Dorsoventrally preserved ence of disarticulation between the two struc-
of Y. lividum (13) (Fig. 1, A and B). This revised specimens show that the gill filaments are tures (figs. S1I and S5F). The horizontal rods
classification is adopted in this paper. Seven external to the bars (Fig. 1H and fig. S5, B, D, are not to be confused with nearby dark lines
pairs of bilaterally symmetrical branchial arches and G). All the bars consist of ~25 straight, called ventral and dorsal thin horizontal lines
have long been recognized in the pharynx of first-order septa that form a series of stacked (“vtl” in fig. S5, B, C, and G; and “dtl” in fig. S1,
yunnanozoans (Fig. 1C). The first pair of mouth- discoid structures measuring ~170 mm wide K and N). These thinner lines might be lon-
surrounding arches were once interpreted as and 140 mm thick (Fig. 1, D and E, and fig. gitudinal blood vessels, because they occur in
blood vessels in the upper and lower lips (10, 12) S1B). Each discoid structure is subdivided the regions of the aortae of vertebrates (10–12).
or oral skirts (11), but others have argued for into two to four cellular chambers by irreg- Our findings can be interpreted in a com-
their close similarity to the remaining arches ular second-order septa (N = 1 to 3) between parative biological context. In vertebrates,
(13). Each branchial arch consists of a bamboo- the two neighboring first-order septa, which chondrocytes are stacked parallel to the arch
like bar formed by evenly spaced, transverse are straight and regularly organized (Fig. 1, D axis during chondrogenesis, forming the clas-
septa and two rows of lanceolate gill filaments and E). X-ray computed microtomography sical stacked discoid structures of pharyngeal
that attach to both sides of each segment (fig. (micro-CT) of two different animals reveals arches in extant vertebrates (15, 16). The ar-
that the cellular chambers are 23 to 84 mm rangement, size, shape, and number of cellular
1
State Key Laboratory for Mineral Deposits Research, School of
in diameter (N = 43) (Fig. 2 and fig. S2). chambers in the branchial bars of yunnanozoans
Earth Sciences and Engineering and Frontiers Science Center The arches are the most commonly pre- match the chondrocytes in the pharyngeal
for Critical Earth Material Cycling, Nanjing University, Nanjing served structures in yunnanozoans (11), with arches of embryonic and primitive vertebrates
210023, China. 2State Key Laboratory of Palaeobiology and abundant carbonaceous residues (Fig. 3I and and in the oral cirri of cephalochordates (Fig. 4D).
Stratigraphy, Nanjing Institute of Geology and Palaeontology
and Center for Excellence in Life and Palaeoenvironment, fig. S3, N, O, and Q). Under scanning elec- The microfibrils in the branchial arches of
Chinese Academy of Sciences, Nanjing 210008, China. tron microscopy (SEM), these residues com- yunnanozoans have characteristics of fibrillin
3
College of Earth and Planetary Sciences, University of Chinese prise microfibrils that are thin, slightly wavy, microfibrils, including their diameter of ~12 nm
Academy of Sciences, Beijing 100049, China.
*Corresponding author. Email: fczhao@nigpas.ac.cn (F.Z.); largely parallel, and densely packed (Fig. 3, (17, 18), wavy appearance, parallel bundling,
byjiang@nju.edu.cn (B.J.) D and F, and fig. S3). Locally, the microfibrils and cross-linkages (19). They differ notably
Tian et al., Science 377, 218–222 (2022) 8 July 2022 1 of 5

RES EARCH | REPOR T
A F
cs
B
ba4-r
f ba1
mo ba
ba3-l ba3-r ba2-r

gonads
C G
ba4-r
ba3-r
ba2-r
gonads cs
D
ba1
gonads
ba2-r ba3-r ba4-r ba5-r ba6-r
ba3-l ba4-l ba5-l ba6-l ba7-l
D
ba5-r
ba6-r ba4-l
ba6-l ?ba2-l
ba7-r
H ba5-r ba4-r ba3-r ba2-r

ba6-r
f
bamboo-like bar ba7-r
E
us
cc
cc fs
discoid lss
structure
f
Fig. 1. Pharyngeal structures of yunnanozoans. (A and B) Line drawings of a (black arrows) sides in specimens NIGP 176267a (F), EC00086a (G), and
yunnanozoan (A) and its pharynx (B). (C to E) An overview of the branchial EC00308a (H). Abbreviations: ba1–7, branchial arches 1 to 7 (with “-r” and “-l”
arches in specimen NIGP 176256a (C). An enlargement (D) and line drawing (E) meaning right and left, respectively); cc, cellular chamber; cs, C-shaped
of the boxed area in (C) show detailed anatomy of each arch. (F to structure; f, filament; fs, first-order septa; lss, lobe-shaped structure; mo, mouth
H) Lateral [(F) and (G)] and ventral views (H) show the dorsoventrally curved, opening; us, U-shaped structure. Scale bars: 1 mm in (B), (C), and (F) to (H);
parenthesis-shaped, horizontal rods in both the dorsal (white arrows) and ventral and 200 mm in (D).
from collagen fibrils, which have thicker diam- The attached gill filaments with matrix mi- chial arches with stacked discoid structures,
eters ranging from 20 to 500 nm and repeat- crofibrils (fig. S3I) are possibly cartilaginous and presence of the horizontal bars at both
ing banding patterns (every ~67 nm) instead gill ray–like structures, as previously inter- ends of the branchial arches. We ran phyloge-
of beads (20). Unlike collagen fibrils that are preted (12). netic analyses on a newly combined set of
the major component of cartilages in jawed Turning next to the horizontal rods that characters, both without and with the char-
vertebrates (gnathostomes) (20, 21), fibrillin connect the branchial arches in yunnanozoans: acter modifications demanded by the findings
microfibrils are the dominant extracellular The presence of the microfibrils and putative of this study (fig. S6). These analyses support
matrix in some parts of cyclostome skeletons. septa again indicate their cartilaginous nature. that yunnanozoans are the earliest branching
The prominent examples occur in the first As skeletal elements, the horizontal rods perhaps stem vertebrates (Fig. 4A and fig. S6B).
two pharyngeal arches of larval lampreys (17) compare to the subchordal and hypobranchial The branchial arches of yunnanozoans also
and in the posterior lingual cartilage of a hag- bars in the branchial basket of lampreys (Fig. 4, B shed light on the early evolution of the ver-
fish (18). Similar matrix microfibrils also char- and C) (15). Comparable horizontal elements also tebrate pharyngeal arches. Because the first
acterize the pharyngeal skeletons of extant existed in the stem hagfish Myxinikela (22) arch in yunnanozoans is lateral to the circu-
cephalochordates (21). The common presence and the fossil jawless fish Euphanerops (23). moral structure and identical to the remain-
of microfibrils further confirms that yunna- These new findings reveal that the branchial ing arches, it completes a serial pattern of
nozoan branchial arches are cartilaginous, skeleton of yunnanozoans shares three char- similar arches that is also seen in another
and this cartilage shares both anatomical acteristics broadly present in crown-group stem vertebrate, Metaspriggina (Fig. 4) (5).
and ultrastructural characteristics with that vertebrates: dominance of microfibrils in the Yunnanozoans are consistent with the con-
of primitive vertebrates and cephalochordates. matrix of a cartilage with chondrocytes, bran- ventional hypothesis that every pharyngeal

RES EARCH | REPOR T
ba5-r
A ba6-r D cc
ba7-l
f
ba1 fs
ba6-l ba5-l ba4-l ba3-l ba2-l
f
B ba5-r ba4-r
ba6-r ba3-r
ba2-r
ba7-r
ba1 E cc
ba7-l D, E
ba6-l ba5-l ba4-l ba2-l f
ba3-l
ba6-r ba5-r ba4-r ba3-r ba2-r

C
ba7-r f
fs
ba1 cc
ba6-l
Fig. 2. X-ray computed microtomography of the branchial arches. (A to C) Photomicrographs of part (A) and mirrored counterpart (C) of specimen NIGP 176268
and its three-dimensional micro-CT reconstruction (B). (D and E) A selected slice and rendered image of the third left branchial arch show the cellular chambers.
Abbreviations as in Fig. 1. Scale bars: 1 mm in (A) to (C), and 100 mm in (D) and (E).
A B C
D
C
B
cs
us
F brightness
D E G H 900 700
F H
I 40 Si
CO Al
Counts
20
Mg S K Fe
0
0 1 2 3 4 5 6 7 Energy (keV)
Fig. 3. Ultrastructure of the branchial arches of yunnanozoans. (A) An overview between the microfibrils (white arrowheads). (G and H) Transmission electron images
of the branchial arches of specimen NIGP 176258a. Abbreviations as in Fig. 1. show the microfibrils and the possible beads at 50-nm intervals (white arrowheads,
(B) Enlarged view of the fifth arch [the boxed area in (A)]. (C) Backscattered scanning one bead is circled) in specimen NIGP 176263. The inset in (H) is the brightness
electron microscopy image of a straight septum [the boxed area in (B)]. profile between the two cyan crosses, with positions of the possible beads marked
(D) Secondary electron image shows that the carbon residues consist of bundles by orange bars. (I) Energy dispersive spectrometry spectrum based on SEM at the
of microfibrils. (E) An overview of the branchial arches of specimen NIGP 176255. white cross in (F). Scale bars: 1 mm in (A) and (E); 200 mm in (B); 50 mm in (C);
(F) Secondary electron image of the white point area in (E) shows the cross-linkages 100 nm in (D), (F), and (G); and 20 nm in (H).

RES EARCH | REPOR T
A B C D
cephalo
nt -chordates
cephalo
-chordates
nt
mo
mo sb sr
urochordates
dhr
dtl dtl vtl vhr
yunnanozoans †
cs
yunnanozoans † ?nt
cs mo lss us
mo ba
lss us
vtl vhr
larval lampreys
v
† nt
ys
mo
t la mpre
eb a dul
ul hb
cb
†
sc
†
ul nt
mo ma hy
lampreys
mo ma hy
c hb ba
† zebrafish
nt
jawed eb
vertebrates †
mo cb
pharyngeal skeletons
ma hy mo
missing/unknown
ma
serial non-serial gill rays
with serially patterned bipartite units
Fig. 4. Phylogenetic relationships and morphological comparisons of pharyn- (D) Chondrocyte arrangement in the boxed areas of the pharyngeal skeletons in
geal skeletons among representative clades of chordates. (A) Simplified (B) and (C). The left half in cephalochordates is from oral cirri, whereas the right half
consensus tree from Bayesian analyses (from fig. S6B), with morphological state of is from pharyngeal skeletons. Abbreviations: ba, branchial arches; cb, ceratobran-
pharyngeal skeletons shown at the tips. This state is indicated as “serial” (green chials; dhr/vhr, dorsal/ventral horizontal rod in yunnanozoans; dtl/vtl, dorsal/ventral
box), meaning the first arch resembles the other arches to form a series, or as thin horizontal line in yunnanozoans; eb, epibranchials; hb, hypobranchial bars in
“nonserial” (purple box), meaning the first arch has extensive skeleton and distinct lamprey; hy, hyoid skeleton; ma, mandibular skeleton; nt, notochord; sc, subchordal
from other arches. Cephalochordates are colored in gray, indicating that they have bar in lamprey; sr/sb, skeleton in primary/secondary bar in cephalochordates; ul,
two alternate units of serially patterned arches. Encircled “V” indicates total-group upper lip cartilages in lamprey. For other abbreviations, see Fig. 1. Scale bars: 50 mm
vertebrates; encircled “C” means crown vertebrates. Fossil taxa are indicated with a in (D). Images in (B) and (C) are not scaled. References and additional explanation
single-dagger symbol. (B and C) Left and ventral views of the pharyngeal skeletons. of the drawings are available in supplementary text.
arch of the ancestral vertebrates was a typical gina acquired a pre-gnathostome condition, 14. X. Hou, P. Cong, Y. Li, Acta Palaeontol. Sin. 48, 402–413
branchial arch (24). having already lost the ancestral basket and (2009).
15. W. M. Martin, L. A. Bumm, D. W. McCauley, Dev. Dyn. 238,
The yunnanozoan basket, with its succes- gained the more separated arches. 3126–3138 (2009).
sive, unjointed branchial arches that are con- 16. C. B. Kimmel et al., Dev. Biol. 203, 245–263 (1998).
nected by dorsal and ventral horizontal rods, RE FERENCES AND NOTES 17. G. M. Wright, J. H. Youson, Am. J. Anat. 165, 39–51 (1982).
is cyclostome-like (Fig. 4, B and C). Indeed, 1. J. Mallatt, J. Zool. 204, 169–183 (1984). 18. G. M. Wright, F. W. Keeley, J. H. Youson, D. L. Babineau,
2. J. Mallatt, Zoolog. Sci. 25, 990–998 (2008). Am. J. Anat. 169, 407–424 (1984).
many researchers claim that the ancestral ver-
3. T. Miyashita, Biol. Rev. Camb. Philos. Soc. 91, 611–657 (2016). 19. E. C. Davis, R. A. Roth, J. E. Heuser, R. P. Mecham, J. Struct. Biol.
tebrates had such an external basket (3) and 4. C. Hirschberger, V. A. Sleight, K. E. Criswell, S. J. Clark, 139, 65–75 (2002).
not the deeper, jointed, and less interconnected J. A. Gillis, Mol. Biol. Evol. 38, 4187–4204 (2021). 20. D. E. Birk, P. Brückner, in The Extracellular Matrix: an Overview,
arches seen in jawed fishes. This scenario, how- 5. S. C. Morris, J.-B. Caron, Nature 512, 419–422 (2014). R. P Mecham, Ed. (Biology of Extracellular Matrix Series,
6. C. Yang, X.-H. Li, M. Zhu, D. J. Condon, J. Chen, J. Geol. Springer, Berlin Heidelberg, 2011), pp. 77–115.
ever, is challenged by the isolated and bipartite Soc. London 175, 659–666 (2018). 21. G. M. Wright, F. W. Keeley, P. Robson, Cell Tissue Res. 304,
branchial bars in the higher stem verte- 7. X. Hou, L. Ramskold, J. A. N. Bergstrom, Zool. Scr. 20, 395–411 165–174 (2001).
brates from the Cambrian, Metaspriggina and (1991). 22. T. Miyashita, Can. J. Zool. 98, 850–865 (2020).
8. J.-Y. Chen, J. Dzik, G. Edgecombe, L. Ramsköld, G. Q. Zhou, 23. P. Janvier, M. Arsenault, Geodiversitas 29, 143–216 (2007).
its relatives, the branchial skeletons of which
Nature 377, 720–722 (1995). 24. G. De Beer, The Development of the Vertebrate Skull
are more gnathostome-like (Fig. 4) (5). This 9. D. Shu, X. Zhang, L. Chen, Nature 380, 428–430 (1996). (Clarendon Press, 1937).
either implies that the cyclostome basket was 10. J.-Y. Chen, D.-Y. Huang, C.-W. Li, Nature 402, 518–522
secondarily derived and thus is merely converg- (1999). AC KNOWL ED GME NTS
11. D. Shu et al., Science 299, 1380–1384 (2003). We acknowledge M. Benton for comments and Z. Yang and
ent on the original, yunnanozoan basket—
12. J. Mallatt, J. Y. Chen, J. Morphol. 258, 1–31 (2003). Y. Wang for discussion. We also acknowledge the following
which is unlikely, given the real similarities we 13. P.-Y. Cong, X.-G. Hou, R. J. Aldridge, M. A. Purnell, Y. Z. Li, individuals for their laboratory assistance: J. Tang, C. Wang,
found here—or else it implies that Metasprig- Palaeontology 58, 45–70 (2015). Y. Fang, and S. Wu at the Nanjing Institute of Geology and

RES EARCH | REPOR T
Palaeontology, Chinese Academy of Sciences (NIGPAS); J. Chen, Investigation: Q.T., F.Z., H.Z., and B.J. Funding acquisition: F.Z., SUPPLEMENTARY MATERIALS
H. Liu, P. Jiang, and Z. Xia at the State Key Laboratory for Mineral M.Z., and B.J. Writing – original draft: Q.T., F.Z., and B.J. Writing – science.org/doi/10.1126/science.abm2708
Deposits Research, Nanjing University; G. Wang at Shanghai Jiao review & editing: Q.T., F.Z., H.Z., M.Z., and B.J. Competing interests: Materials and Methods
Tong University; and B. Wang, X. Lei, X. Chen, L. Hu, Z. Yin, The authors declare that they have no competing interests. Data Supplementary Text
R. Wu, T. Zhao, W. Zhang, and X. Zhao. We are grateful to two and materials availability: The specimens studied here are all stored Figs. S1 to S6
anonymous reviewers for their constructive comments and in NIGPAS. The measurements and datasets for the phylogenetic Tables S1 to S4
suggestions. Funding: Funding was provided by the Strategic analysis are available in the supplementary materials. Computed References (25–64)
Priority Research Program (B) of the Chinese Academy of Sciences tomography data are deposited in MorphoSource at https://www. MDAR Reproducibility Checklist
[XDB26000000 (B.J., F.Z., and M.Z.)], the National Science morphosource.org/projects/000378422?locale=en. License Data S1 to S3
Foundation of China [42288201 (B.J.); 41921002 and 42072006 information: Copyright © 2022 the authors, some rights reserved;
(F.Z. and M.Z.)], and the Fundamental Research Funds for the exclusive licensee American Association for the Advancement of
Central Universities [0206-14380137 (B.J.)]. Author contributions: Science. No claim to original US government works. https://www. Submitted 5 September 2021; accepted 17 May 2022
Conceptualization: F.Z. and B.J. Fossil collection: F.Z. and M.Z. science.org/about/science-licenses-journal-article-reuse 10.1126/science.abm2708

RES EARCH
STRUCTURAL BIOLOGY table S1). To investigate the mechanisms under-
Conformational selection guides b-arrestin

lying b-arrestin recruitment at ACKR3, we first
measured b-arrestin-2 recruitment in response
to a panel of ACKR3 agonists, including the
recruitment at a biased G protein–coupled receptor endogenous chemokine CXCL12 [median effec-
tive concentration (EC50) = 0.75 nM] (18), the
Andrew B. Kleist1,2†‡, Shawn Jenjak1†, Andrija Sente3, Lauren J. Laskowski4, small molecule CCX777 (EC50 = 0.95 nM; max-
Martyna Szpakowska5, Maggie M. Calkins4, Emilie I. Anderson4, Lisa M. McNally4, imum effect (Emax, % of CXCL12) = 75 ± 2%]
Raimond Heukers6§, Vladimir Bobkov6, Francis C. Peterson1, Monica A. Thomas1,2¶, (19), and a recently described peptide LIH383
Andy Chevigné5, Martine J. Smit6, John D. McCorvy4, M. Madan Babu7,8, Brian F. Volkman1* (EC50 = 4.8 nM; Emax = 83 ± 1%) (Fig. 1C) (20).
We also identified a potent, extracellular-
G protein–coupled receptors (GPCRs) recruit b-arrestins to coordinate diverse cellular processes, but targeting, ACKR3-competitive antagonist
the structural dynamics driving this process are poorly understood. Atypical chemokine receptors nanobody called VUN701 [median inhibitory
(ACKRs) are intrinsically biased GPCRs that engage b-arrestins but not G proteins, making them a model concentration (IC50) = 1.47 mM] using nano-
system for investigating the structural basis of b-arrestin recruitment. Here, we performed nuclear body phage display (Fig. 1C; fig. S1, B to F; and
magnetic resonance (NMR) experiments on 13CH3-e–methionine–labeled ACKR3, revealing that materials and methods). These four ligands
b-arrestin recruitment is associated with conformational exchange at key regions of the extracellular display a range of activities for b-arrestin-2
ligand-binding pocket and intracellular b-arrestin–coupling region. NMR studies of ACKR3 mutants recruitment, allowing us to sample inactive
defective in b-arrestin recruitment identified an allosteric hub in the receptor core that coordinates (i.e., VUN701-bound) and active (i.e., CCX777-,
transitions among heterogeneously populated and selected conformational states. Our data suggest that LIH383-, and CXCL12-bound) b-arrestin–
conformational selection guides b-arrestin recruitment by tuning receptor dynamics at intracellular recruiting states of ACKR3 (Fig. 1D).
and extracellular regions. NMR spectroscopy allows simultaneous
characterization of receptor conformation
O
at multiple sites (21). We previously purified
13
nce thought of as G protein–coupled tural comparisons of GPCRs bound to antago- CH3-e-Met–labeled, ACKR3-bound CCX777
receptor (GPCR) off-switches, b-arrestins nists and b-arrestin–biased ligands suggest that in lauryl maltose neopentyl glycol/cholesteryl
are now known to coordinate diverse, ligand-specific conformational changes gov- hemisuccinate (MNG/CHS) micelles for NMR
G protein–independent signaling re- ern b-arrestin recruitment (12, 13). Despite this studies (22), and we used this method for other
sponses (1). In a phenomenon called progress, how ligand-binding pocket changes ACKR3-ligand complexes here (Fig. 1E, figs. S2
biased signaling,some GPCR ligands (biased are transmitted to the b-arrestin interface re- and S3, and materials and methods). As with
ligands) preferentially recruit b-arrestins or mains poorly understood. Other findings have other GPCRs, copurification with ligands was
G proteins (2). Because they select one pathway, complicated our understanding of b-arrestin necessary to produce sufficient quantities of
and therefore a specific functional outcome, recruitment by GPCRs. The identification of ACKR3 for NMR studies (23). Structural ho-
biased ligands have demonstrated advantages multiple, distinct b-arrestin–recruiting GPCR mology modeling of ACKR3 (fig. S4 and
over conventional ligands in preclinical (3–5) conformations suggests there may be multiple materials and methods) demonstrates eight
and clinical (6, 7) testing. More than 30% of conformational solutions to b-arrestin recruit- native Met residues (excluding the N-terminal
US Food & Drug Administration–approved ment (15), but raises the question of how such Met) distributed in ACKR3′s tertiary structure
drugs target GPCRs (8), so understanding the different conformations could elicit similar for NMR labeling. NMR labels serve as sensi-
mechanistic basis by which GPCRs recruit outcomes. Conversely, nearly identical confor- tive probes reporting on regional changes in
b-arrestins has important implications for mations of b-arrestin– and G protein–bound GPCR conformation and dynamics (24, 25). In
drug development. GPCRs (9–11) suggest that conformational this setting, Met2125x39 and Met1383x46 [where
Recent GPCR–b-arrestin structures (9–11) changes alone might not account for b-arrestin the superscripts indicate GPCRdb nomenclature
and biophysical studies of b-arrestin–biased recruitment. Finally, some studies of b-arrestin (26)] are well positioned to report on confor-
ligands (12–14) provide mechanistic insights recruitment use ligands that provoke re- mational changes associated with b-arrestin re-
into b-arrestin recruitment. For instance, struc- sidual signaling at G proteins (16), making it cruitment at ACKR3 because they are located
challenging to isolate the specific molecular in key regions of the ligand-binding pocket
1
Department of Biochemistry, Medical College of Wisconsin, changes leading to b-arrestin recruitment. (27, 28) and intracellular effector binding in-
Milwaukee, WI 53226, USA. 2Medical Scientist Training Given these complexities, studies of b-arrestin terface (29), respectively.
Program, Medical College of Wisconsin, Milwaukee, WI recruitment could benefit from functionally Ligand interactions with TM5 play key roles
53226, USA. 3MRC Laboratory of Molecular Biology,
Cambridge CB2 0QH, UK. 4Department of Cell Biology, decoupled receptors that exclusively recruit in GPCR activation (27, 30), but how confor-
Neurobiology, and Anatomy, Medical College of Wisconsin, b-arrestin. Atypical chemokine receptors (ACKRs) mational changes in TM5 lead to b-arrestin
Milwaukee, WI 53226, USA. 5Immuno-Pharmacology and represent one such naturally occurring system. recruitment is unclear. NMR spectra for the
Interactomics, Department of Infection and Immunity,
Luxembourg Institute of Health (LIH), L-4354 Esch-sur-
ACKR3 is an intrinsically b-arrestin–biased four ligand-bound states show chemical shift
Alzette, Luxembourg. 6Amsterdam Institute for Molecular GPCR that recruits b-arrestin but does not acti- perturbations (CSPs) for the Met2125x39 peak,
and Life Sciences, Division of Medicinal Chemistry, Faculty of vate G protein (Fig. 1A) (17). Here, we performed which remains colinear along the 1H axis but
Sciences, Vrije Universiteit, 1081 HZ Amsterdam,
Netherlands. 7Department of Structural Biology, St. Jude
nuclear magnetic resonance (NMR) studies of variable along the 13C axis (Fig. 2, A and B).
Children’s Research Hospital, Memphis, TN 38105, USA. ACKR3, showing that b-arrestin recruitment This indicates that ACKR3 exists along a two-
8
Center for Data Driven Discovery, St. Jude Children’s at an intrinsically biased GPCR is guided by state equilibrium, with the peaks at either ex-
Research Hospital, Memphis, TN 38105, USA. tuning its conformational equilibrium. treme defining the conformational end points
*Corresponding author. Email: bvolkman@mcw.edu
†These authors contributed equally to this work. ‡Present address: (31). In the CCX777-bound state, downfield
Harriet Lane Pediatric Residency Program, Johns Hopkins Child- RESULTS peak positions [~18.2 to 19 parts per million
ren’s Center, Baltimore, MD 21287, USA. §Present address: QVQ In agreement with prior studies, ACKR3 re- (p.p.m.)] indicate that 13CH3 is in a trans rota-
Holding BV, 3584 CL Utrecht, Netherlands. ¶Present address:
Department of Anesthesiology and Critical Care Medicine, Johns cruits b-arrestin but does not activate G pro- mer (end point 1), whereas in the CXCL12- and
Hopkins Medicine, Baltimore, MD 21287, USA. tein signaling (Fig. 1, A and B; fig. S1A; and LIH383-bound states, upfield peak positions
Kleist et al., Science 377, 222–228 (2022) 8 July 2022 1 of 7

RES EARCH | REPOR T
Fig. 1. b-arrestin–biased signaling at ACKR3 and structural characterization recruitment. Structures are not to scale. CXCL12 PDB: 2KEC; nanobody from
by NMR. (A) CXCL12 activates b-arrestin but not G protein at ACKR3. (B) CXCL12- PDB 6KNM used to represent VUN701, and LIH383 and CCX777 were modeled
mediated cAMP inhibition of CXCR4 (positive control) and ACKR3 as shown by in PyMol. LIH383 sequence is FGGFMRRK (20). The chemical structure of CCX777 is
Glosensor assay (EC50 = 0.21 nM, CXCR4) (top). CXCL12-mediated b-arrestin-2 shown in (19). (E) 1H-13C heteronuclear single quantum coherence NMR spectra
recruitment to ACKR3 as shown by Tango assay (EC50 = 3.9 nM, ACKR3) (bottom). of WT-ACKR3 with various ligands at 310 K. Assigned Met residues are labeled.
N = 3 in triplicate in both assays. Error bars indicate SEM. (C) b-arrestin recruitment Asterisk denotes inferred assignments from other ligand-bound states (see the
to ACKR3 as shown by Nano-BiT assay. VUN701 dose response alone could not materials and methods). Negative contour peaks are shown in semitransparency and
be fit (black circles). Purple circles reflect VUN701 dose response with CXCL12 dashed lines. Peaks marked “a” encompass natural abundance peaks from buffer
at 3.2 nM . See text and table S1 for EC50 and Emax. All conditions, N = 3 in and detergent components (see also figs. S2G and S3). All spectra are shown at the
duplicate. Error bars indicate SEM. (D) Summary of ligand potency at b-arrestin same contour except LIH383, which was lowered to represent M2125x39.
(~16 to 16.5 p.p.m.) indicate that 13CH3 is in state is selected. Antagonists such as VUN701 has equivalent efficacy for b-arrestin recruit-
a gauche rotamer (end point 2) (Fig. 2B and may fail to constrain the region sampled by ment as wild type (WT) (fig. S5F), indicating
fig. S5, A and B) (22, 32–34). Met2125x39, preserving conformational hetero- that Met2125x39 reports on but does not itself
By contrast, in VUN701-bound ACKR3, the geneity (seen by increased exchange relative to mediate functional changes in the TM5 region.
Met2125x39 peak appears between the two end the active state) and precluding TM5-binding Does this occur in other GPCRs? We iden-
points at the random coil 13CH3 position, indi- pocket stabilization required for ACKR3 acti- tified all ligand-contacting residues among
cating fast conformational exchange between vation. Binding data show that Ala mutagen- structures of (i) GPCRs bound to b-arrestin–
gauche and trans rotamers in the inactive, esis of Met2125x39 similarly affects CXCL12 and biased ligands (or full-length b-arrestin; joint-
b-arrestin–nonrecruiting state (Fig. 2B). In- VUN701, suggesting that NMR changes be- ly called b-arresting–recruiting states) and
deed, CSPs of Met2125x39 from 13CH3 random tween inactive and active states likely represent (ii) inactive-state structures of the same GPCRs
coil vary by ligand type (fig. S5A), suggest- functionally important regional conformational (fig. S6, A and B). Using contact network analysis
ing that b-arrestin recruitment is associated changes, as opposed to differences in direct (29, 35), we found that ligands in b-arrestin–
with decreasing exchange among Met2125x39 ligand contacts at the probe residue (fig. S5, recruiting states make similar numbers of
rotamers irrespective of which c3 rotameric C to E, and table S1). Indeed, Met212Ala 5x39 contacts with residues in TM5 but significantly

RES EARCH | REPOR T
Fig. 2. Conformational changes in the ligand-binding pocket and intracellular contacts with the ligand TRV023 (b-arrestin–biased; black outline) in TM5,
region characterize the b-arrestin–recruiting state. (A) ACKR3 model ECL2, and TM7 are shown as blue spheres (PDB 6OS1). The position of
depicting M2125x39 and M1383x46 probes. (B) Overlay of M2125x39 peaks from NMR probe 5x39 in AT1R is shown. Stabilization of TM5-ECL2-TM7 by biased
ligand-bound ACKR3 complexes at 310 K. The 16 to 16.5 and 18.5 to 19 p.p.m. ligands is depicted as a lock (right). (E) Met1383x46 peaks in all four ligand-
peaks (13C) correspond to gauche and trans rotameric states, respectively. bound states. Upfield peak positions (1H: ~1.3 p.p.m.) of M1383x46 among
Ligand-specific b-arrestin activity is depicted at right. Open and closed agonist-bound states supports ring-current shifts due to aromatic side chain
locks depict conformational sampling and restriction, respectively (bottom). interactions. Peaks are marked “a” as in Fig. 1. (F) 3x46 contact residues
(C) Ligand-residue interactions were compared between antagonist and at 2x43 and 7x53 exclusively in active-state complexes and 6x37 exclusively in
b-arrestin–biased ligand-bound GPCRs (see the materials and methods). inactive-state complexes (see the materials and methods). (G) b-arrestin
Shown is a comparison of the mean number of ligand contacts with TM5, recruitment of WT ACKR3 versus Ala mutants of 3x46-contacting mutants by
ECL2, and TM7 residues. *P < 0.05; **P < 0.005, unpaired t test. (D) AT1R NanoBiT (N = 3). See also table S1.
more contacts with residues in ECL2 and TM7 region (Fig. 2A) (29). In the three agonist- the b-arrestin–coupling region sampled by
than do antagonists (Fig. 2C). Thus, TM5 may bound states (CXCL12-, CCX777-, and LIH383- M1383x46.
anchor agonist interactions with ECL2 and bound), the Met1383x46 NMR peak has nearly By contrast, the inactive, VUN701-bound
TM7 (27, 36), creating a tripartite lock involving identical chemical shifts (Fig. 2E), indicating ACKR3 state is characterized by CSPs of the
the 5x39 region (Fig. 2D and fig. S6C), whereas that all three agonists elicit a shared confor- Met138 3x46 peak in both 1 H and 13C values
antagonists fail to stabilize all three lock points mation of Met1383x46. The upfield position (Fig. 2E), indicating diminished ring-current
(ECL2, TM5, and TM7) simultaneously. (~1.3 p.p.m.) of Met1383x46 in the 1H dimension shift (1H dimension), depletion of the gauche
Large-scale conformational changes underly- likely reflects ring current shifts caused by prox- rotamer (13C dimension), and more gauche/
ing GPCR activation are driven by local re- imity to an aromatic side chain (33, 38, 39). trans exchange ( 13C dimension). As in the
arrangements of residues at key positions in These data indicate that the agonist-specific binding pocket, VUN701 causes the smallest
the GPCR structure (so-called microswitch conformations of Met2125x39 (i.e., trans ver- CSP from random coil 13CH3-e-Met versus the
residues (37). We next examined Met1383x46, a sus gauche) in the TM5-binding pocket are three agonists, suggesting that agonists reg-
microswitch residue in the b-arrestin–coupling funneled into a common conformation at ulate b-arrestin recruitment by decreasing

RES EARCH | REPOR T
Fig. 3. Mutational inactivation of b-arrestin recruitment through the Peaks are marked “a” as in Fig. 1. See fig. S8A and legend for the Met1383x46
conserved polar network. (A) Position of Asn1273x35 in the ACKR3 model assignment method. (E) Normalized CSPs for Met2125x39 in the 13C dimension
relative to Met2125x39 and Met1383x46. (B) b-arrestin-2 recruitment with CXCL12 (y axis) and Met1383x46 in the 1H dimension (x axis) from random coil for
for the ACKR3 mutants Asn127Lys3x35 and Asn127Ser3x35 as shown by the Asn1273x35 mutants and CXCL12- and VUN701-bound WT-ACKR3. Arrows depict
Tango assay. All conditions, N = 3 in triplicate. Error bars indicate SEM. See also transitions among CXCL12-bound WT-ACKR3 and mutant ACKR3. (F) Depiction of
table S1. (C) Close-up view of 1H-13C-HSQC in the Met2125x39 region (right), differences between WT-ACKR3 and Asn127Lys3x35 in CXCL12-bound states.
highlighting VUN701-WT-ACKR3 (first panel), CXCL12-ACKR3 Asn127Lys3x35 Despite being bound to CXCL12 (blue key), Asn127Lys3x35 locks ACKR3 in the
(second panel), CXCL12-WT-ACKR3 (third panel), and CXCL12-ACKR3 Asn127Ser3x35 inactive state, abrogating CXCL12’s effects on the Met2125x39 and Met1383x46
(fourth panel). (D) Overlay of 13C-HSQC in the Met1383x46 region for WT-ACKR3- probes. (G) Comparison of residue-residue interactions AT1R–b-arrestin–
VUN701, ACKR3-Asn127Lys3x35-CXCL12, ACKR3-Asn127Ser3x35-CXCL12, and biased ligand and AT1R-antagonist bound states (top) reveals the formation of
WT-ACKR3-CXCL12 complexes. A shaded triangle suggests peak noncolinearity. a 3x39–7x46 interaction in the b-arrestin state (bottom).
conformational heterogeneity in this region of and 7x53 and 6x37 have been shown to stabi- fig. S7, D to F) and the pronounced ring cur-
the b-arrestin–coupling region relative to the lize GPCRs in G protein signaling (7x53) and rent shift (1H) of Met1383x46 (Fig. 2E) in active
inactive state (fig. S7A). inactive states (6x37), respectively (29). but not inactive states of ACKR3 suggest that
What is the role of 3x46, a known micro- To investigate whether 3x46 and active- Met1383x46 reports on dynamic alterations in
switch involved in G protein activation (29), in state contacts from GPCR-b–arrestin complexes Tyr3157x53 associated with b-arrestin recruit-
b-arrestin recruitment among other GPCRs? (2x43 and 7x53) might play a role in b-arrestin ment (33, 38, 39). Active-state interactions be-
We calculated all intramolecular, residue- recruitment to ACKR3, we tested Met138Ala3x46, tween 3x46 and 7x53 may thus lock this region
residue interactions among structures of Ile84Ala2x43, and Tyr315Ala7x53 mutants for in place, decreasing local conformational het-
(i) b-arrestin-bound and (ii) inactive-state b-arrestin recruitment (Fig. 2G and table S1). erogeneity and increasing the likelihood of
structures of GPCRs with resolved side chains Although Met138Ala3x46 and Ile84Ala2x43 min- b-arrestin engagement. Indeed, receptors con-
(b1AR and rhodopsin) (fig. S7, B and C), finding imally affected ACKR3 function, Tyr315Ala7x53 tact the b-arrestin finger loop, a key structural
that 3x46 contacts 2x43 and 7x53 in b-arrestin- almost completely abolished b-arrestin re- region anchoring GPCR-b–arrestin interactions,
bound structures and 6x37 in inactive structures cruitment, demonstrating the essential role directly at or in the vicinity of 3x46 and 7x53
(Fig. 2F and fig. S7, D to F). Of note 7x53 is a for this residue on ACKR3 activation. The in b1AR- and rhodopsin-arrestin complexes
highly conserved Tyr among class A GPCRs (37), close proximity of 3x46 to Tyr7x53 (Fig. 2F and (11, 40). By contrast, a more heterogeneous

RES EARCH | REPOR T
Fig. 4. Dynamic control ing that it acts as an inverse agonist at this

of b-arrestin recruit- mutant (fig. S8C and table S1). As shown by
ment to ACKR3. Alloste- NMR, the VUN701-bound Asn127Ser3x35 mu-
ric regulation of GPCR tant reverts to the inactive (i.e., trans/gauche
b-arrestin activation by interconverting) state at Met2125x39, and the
coordinated transitions in Met1383x46 peak is diminished, suggesting tran-
conformational heteroge- sition to a more dynamic state at both probes
neity. Ligands constrain (fig. S8, D to G).
(agonists) or promote Among analyzed GPCR structures (i.e.,
(antagonists) conforma- those bound to b-arrestin or b-arrestin–biased
tional heterogeneity ligands), AT1R is most closely related to ACKR3
by altering intermolecular and is the only one that shares Asn3x35. Cal-
(ligand-residue) and culation of intramolecular, residue-residue
intramolecular (residue- interactions of AT1R bound to antagonists
residue) interactions and b-arrestin–biased agonists revealed that
throughout the GPCR whereas 3x35 interacts with 7x46, a key residue
structure. for b-arrestin recruitment (12) in both states,
7x46 interacts with 3x39, which is also involved
in b-arrestin recruitment (12), only in the active
state (Fig. 3G). This rearrangement in turn
shifts the register of TM3 and TM7 relative
to one another in the region above 3x46 and
7x53 (Fig. 3G), suggesting a mechanism by
which 3x35 mutations in ACKR3 might be
transmitted to the b-arrestin–coupling re-
inactive state (characterized by rotameric ex- VUN701, suggesting the occupancy of a third, gion as observed at the intracellular probe
change and a diminished ring current shift distinct conformation in this ACKR3 mutant. Met138 3x46 . In effect, Asn 3x35 mutations in
at Met1383x46) may reflect absent interactions Although Met1383x46 in this mutant occupies ACKR3 may function by promoting (active)
with 7x53 and a lower propensity for b-arrestin a gauche rotamer (13C: ~16.1 p.p.m.), its 1H or disrupting (inactive) 3x39–7x46 interac-
engagement (Fig. 2F). value is closer to that of VUN701-bound ACKR3, tions, resulting in decreased (active) or in-
To what extent are dynamic changes in suggesting fewer local aromatic interactions creased (inactive) conformational exchange
extracellular (Met2125x39) and intracellular resulting in a diminished ring current shift. at the intracellular probe.
(Met1383x46) probes allosterically linked through We speculate that this downfield 1H shift re-
the receptor core during ACKR3 activation? flects increased conformational exchange be- DISCUSSION
Conserved polar core residues, including the tween both gauche rotamers (i.e., + and –) in Some studies have identified distinct confor-
residue Asn3x35 (12, 41, 42), coordinate a sodium the mutant state, which would weaken inter- mational changes associated with b-arrestin
ion (inactive state) and a water network (active actions with an aromatic side chain compared recruitment (12, 13), but others have shown
state) to regulate GPCR activation and bias (Fig. with a single gauche rotamer (i.e., + or –) in the multiple b-arrestin competent conformations
3A) (42). We investigated the effects of known active state (fig. S8B). Regardless, pronounced (14). How can b-arrestin be both sensitive to and
inactivating and constitutively activating muta- downfield 1H shifts and peak broadening of tolerant of GPCR conformational changes?
tions of conserved polar core residue Asn1273x35 Met1383x46 in the CXCL12-bound, inactivat- The results presented here help to recon-
on ACKR3 activation. Consistent with prior re- ing Asn127Lys3x35 mutant reveal that, as in the cile this apparent contradiction by revealing
sults, the Asn127Lys3x35 mutation showed com- ligand-binding pocket, the conserved core mu- that whereas b-arrestin is tolerant of diverse
plete inactivity in b-arrestin recruitment (Fig. 3B), tation abrogates b-arrestin-2 recruitment by GPCR conformations in some parts of the re-
whereas Asn127Ser3x35 acts as a constitutively increasing conformational heterogeneity at a ceptor, it may have more stringent conforma-
active mutant (Fig. 3B and table S1) (43). key position in the b-arrestin–coupling region tional requirements in other parts. At TM5
How do Asn1273x35 mutants affect the con- (Fig. 3E). Thus, a single mutation in the re- of the ligand-binding pocket, we found that
formation and stability of the binding pocket ceptor core, Asn127Lys3x35, destabilizes ACKR3 multiple conformational solutions (15) are
and intracellular probes as observed by NMR at extracellular and intracellular sites, over- compatible with b-arrestin recruitment. These
chemical shifts? In the context of CXCL12, the riding the effects of CXCL12 to decrease con- multiple-binding-pocket conformations are
Asn127Lys3x35 mutation shifts the Met2125x39 formational exchange present in WT ACKR3 funneled into a single active conformation as
peak downfield relative to its position in the (Fig. 3F). monitored at the intracellular probe Met1383x46.
WT-ACKR3 complex (Fig. 3C and fig. S8A), In the CXCL12-bound, constitutively active NMR and structural evidence for active-state
resembling the VUN701-bound state charac- Asn127Ser3x35 mutant receptor, Met2125x39 interactions between 3x46 and 7x53 suggest
terized by trans/gauche rotamer exchange and Met1383x46 peaks overlay those in the that the stringent intracellular conformational
at Met2125x39. This reveals that the receptor CXCL12-WT-ACKR3 state (Fig. 3, C to E, and requirements comprise the intracellular regions
core mutation exerts long-range, allosteric ef- fig. S8A) despite this state’s unresponsiveness of TM3 and TM7. Indeed, although studies of
fects that enhance conformational exchange to CXCL12, indicating that the activating muta- AT1R bound to multiple b-arrestin agonists
in the extracellular ligand-binding pocket. The tion promotes conformational homogeneity have revealed diverse intracellular conforma-
Asn127Lys3x35 mutation shifts the Met1383x46 (no rotameric exchange) extracellularly and tions, most variation was in TM5 and TM6,
peak downfield in both the 1H and 13C values intracellularly, even as it decouples allosteric with only modest variation in TM7 (14).
to an intermediate position between those in transmembrane communication. By contrast, Conformational control, although important,
CXCL12- and VUN701-bound states (Fig. 3D). VUN701 decreases the elevated basal b-arrestin may not fully account for b-arrestin recruit-
This peak is not colinear with CXCL12 and recruitment of ACKR3 Asn127Ser3x35, suggest- ment. The observation that intracellular and

RES EARCH | REPOR T
extracellular probes both transition between cellular region (46). There are several possible 21. I. R. Kleckner, M. P. Foster, Biochim. Biophys. Acta 1814,
conformational heterogeneity (inactive) and reasons for this difference. Principally, com- 942–968 (2011).
22. A. B. Kleist et al., Methods Cell Biol. 149, 259–288 (2019).
homogeneity (active) suggests that b-arrestin pared with extensive GPCR–G protein interface
23. X. Zhang, R. C. Stevens, F. Xu, Trends Biochem. Sci. 40, 79–87
recruitment is governed in part by tuning the (mediated primarily through the a5-helix of (2015).
conformational spectrum sampled by the re- G protein), GPCR-b-arrestin interactions are 24. R. Sounier et al., Nature 524, 375–378 (2015).
ceptor. The regulation of b-arrestin recruit- mediated by a smaller, less structured finger 25. J. J. Liu, R. Horst, V. Katritch, R. C. Stevens, K. Wüthrich,
Science 335, 1106–1110 (2012).
ment in this manner could help to explain loop. Other reasons include dependence of
26. V. Isberg et al., Trends Pharmacol. Sci. 36, 22–31 (2015).
the apparent agnosticism of b-arrestin for G protein (but not b-arrestin) engagement 27. J. D. McCorvy et al., Nat. Struct. Mol. Biol. 25, 787–796 (2018).
specific GPCR conformations (14): Altera- on outward, destabilizing motions of TM6 28. T. Warne, C. G. Tate, Biochem. Soc. Trans. 41, 159–165 (2013).
tion of receptor stability in key regions may (14, 25), and requirement of G protein (but 29. A. J. Venkatakrishnan et al., Nature 536, 484–487 (2016).
tune a GPCR’s ability to couple b-arrestin ir- not b-arrestin) coupling to be linked to GTP 30. S. G. Rasmussen et al., Nature 469, 175–180 (2011).
31. B. F. Volkman, D. Lipson, D. E. Wemmer, D. Kern, Science 291,
respective of conformation (e.g., at Met2125x39). hydrolysis. 2429–2433 (2001).
Likewise, globally different GPCR confor- We propose roles for receptor conforma- 32. A. S. Solt et al., Nat. Commun. 8, 1795 (2017).
mations may be similarly tuned to couple tions and conformational equilibria in regulat- 33. G. L. Butterfoss et al., J. Biomol. NMR 48, 31–47 (2010).
b-arrestins by decreasing conformational ing b-arrestin recruitment, but other aspects of 34. Y. Kofuku et al., Nat. Commun. 3, 1045 (2012).
35. M. Kayikci et al., Nat. Struct. Mol. Biol. 25, 185–194 (2018).
heterogeneity of the same key epitopes of the b-arrestin coupling must also be considered.
36. M. Szpakowska et al., Br. J. Pharmacol. 175, 1419–1438 (2018).
b-arrestin interface. For instance, the observed conformational 37. R. Nygaard, T. M. Frimurer, B. Holst, M. M. Rosenkilde,
Ligand-specific modulation of the confor- changes might lead to differences in GPCR T. W. Schwartz, Trends Pharmacol. Sci. 30, 249–259 (2009).
mational spectrum observed at NMR probes kinase recruitment between antagonists and 38. R. E. London, B. D. Wingad, G. A. Mueller, J. Am. Chem. Soc.
130, 11097–11105 (2008).
supports a role for conformational selection agonists, resulting in distinct phosphorylation 39. S. J. Perkins, K. Wüthrich, Biochim. Biophys. Acta 576,
in b-arrestin recruitment. In this model, a pre- patterns that alter GPCR functional properties 409–423 (1979).
existing spectrum of conformations is nar- (47). Indeed, b-arrestin interaction with phos- 40. Y. Kang et al., Nature 523, 561–567 (2015).
rowed upon perturbation of the system (44), phorylated GPCRs in the absence of core en- 41. G. Fenalti et al., Nature 506, 191–196 (2014).
42. V. Katritch et al., Trends Biochem. Sci. 39, 233–244 (2014).
in this case by addition of various ligands. In gagement has been shown to be sufficient for
43. N. Montpas et al., J. Biol. Chem. 293, 893–905 (2018).
the ligand-binding pocket, the inactive confor- GPCR internalization and b-arrestin–mediated 44. T. R. Weikl, F. Paul, Protein Sci. 23, 1508–1518 (2014).
mation of M2125x39 is a composite of multiple signaling (but not G protein desensitization) 45. L. Ye, N. Van Eps, M. Zimmer, O. P. Ernst, R. S. Prosser, Nature
active conformations, suggesting that b-arrestin (48). Nevertheless, other studies have cor- 533, 265–268 (2016).
related the extent of allosteric coupling be- 46. A. Manglik et al., Cell 161, 1101–1111 (2015).
recruitment is governed not by switching be-
47. A. Sente et al., Nat. Struct. Mol. Biol. 25, 538–545 (2018).
tween distinct on and off conformations, but tween agonists and b-arrestin with a particular 48. T. J. Cahill3rd et al., Proc. Natl. Acad. Sci. U.S.A. 114,
rather by selecting one of many active con- ligand’s efficacy in b-arrestin recruitment 2562–2567 (2017).
formations (active) versus facilitating rapid (49), suggesting at least some role for GPCR 49. R. T. Strachan et al., J. Biol. Chem. 289, 14211–14224
conformational changes dictating b-arrestin (2014).
exchange between them (inactive). The same is
only partially true intracellularly: Met1383x46 engagement.
exists as a composite of trans/gauche rotamers In summary, our results show that coor- AC KNOWL ED GME NTS
in the inactive state that is narrowed to gauche- dinated, allosterically linked changes in recep- We thank A. Dishman for comments on the manuscript and
only rotamers in the active state. Nevertheless, tor dynamics regulate b-arrestin recruitment J. Campbell and colleagues at Chemocentryx for supplying CCX777.
Funding: This work was supported by the National Institutes of
Met1383x46 only experiences ring current shift to the intrinsically biased receptor ACKR3. Health (grant F30CA196040 to A.B.K.; grant R01AI058072 to B.F.V.;
effects in the active state. Although this might Our results provide a framework with which grant F30HL134253 to M.A.T.; grant R35GM133421 to J.D.M.; and
suggest that the active-state conformation of to understand the molecular changes re- grant T32 GM080202 to the Medical Scientist Training Program
at Medical College of Wisconsin to A.B.K. and M.A.T.); the State of
Met1383x46 is not accessible to ACKR3 in the quired for b-arrestin recruitment and provide Wisconsin Tax Check-Off Program for Cancer Research and
inactive state, it could also reflect sensitivity insights that may facilitate the design of biased the Medical College of Wisconsin Cancer Center (B.F.V.); the
limitations of 13C compared with other NMR therapeutics. Luxembourg National Research Fund (Pathfinder “LIH383,” INTER/
FWO “Nanokine” grant 15/10358798 to A.C.); INTER/FNRS grants
nuclei (45).
20/15084569, PoC “Megakine” 19/14209621, AFR-3004509,
How might different ligands enhance or and PRIDE 11012546 “NextImmune” to A.C.; F.R.S.-FNRS-Télévie
RE FERENCES AND NOTES
suppress changes in conformational plastic- grants 7.4593.19, 7.4529.19, and 7.8504.20 to A.C.; European
1. Y. K. Peterson, L. M. Luttrell, Pharmacol. Rev. 69, 256–297 Union’s Horizon 2020 MSCA Program (grant 641833 ONCORNET
ity to regulate b-arrestin? b-arrestin agonists (2017). and 860229 ONCORNET2.0 to M.J.S.); American Lebanese Syrian
might be described as stabilizing a particular 2. D. Wootten, A. Christopoulos, M. Marti-Solano, M. M. Babu, Associated Charities (ALSAC grant to M.M.B.); and the UK Medical
set of conformations that facilitate b-arrestin P. M. Sexton, Nat. Rev. Mol. Cell Biol. 19, 638–653 (2018). Research Council (MRC grant MC_U105185859 to M.M.B. and
3. A. Manglik et al., Nature 537, 185–190 (2016). A.S.). Author contributions: A.B.K. and B.F.V. conceived of the
binding (Fig. 4). Indeed, contact network anal-
4. C. L. Schmid et al., Cell 171, 1165–1175.e13 (2017). project. B.F.V. supervised the project. A.B.K. and S.J. made
ysis of inactive- and active-state structures 5. L. M. Slosky et al., Cell 181, 1364–1379.e14 (2020). constructs, purified protein, and collected and analyzed NMR data.
of GPCRs reveals enhanced intermolecular 6. N. K. Singla et al., Pain Pract. 19, 715–731 (2019). A.B.K., S.J., and B.F.V. analyzed data, interpreted results, and wrote
(ligand-residue) and intramolecular (residue- 7. N. Singla et al., J. Pain Res. 10, 2413–2424 (2017). the manuscript. A.B.K. and S.J. produced the figures. A.B.K. and
8. A. S. Hauser, M. M. Attwood, M. Rask-Andersen, H. B. Schiöth, A.S. performed contact network analysis. A.B.K., A.S., and M.M.B.
residue) interactions in b-arrestin active states, D. E. Gloriam, Nat. Rev. Drug Discov. 16, 829–842 (2017). interpreted contact network analysis data. M.J.S., R.H., and V.B.
suggesting that b-arrestin agonists might sta- 9. D. P. Staus et al., Nature 579, 297–302 (2020). identified VUN701 and performed nanobody screening and
bilize specific conformations by organizing 10. W. Huang et al., Nature 579, 303–308 (2020). characterization. F.C.P. cloned VUN701, optimized pulse sequences,
11. Y. Lee et al., Nature 583, 862–866 (2020). and oversaw NMR data collection and analysis. M.A.T. made and
denser contact networks. Antagonists fail to 12. L. M. Wingler et al., Science 367, 888–892 (2020). validated ACKR3 homology model. A.C. identified and provided
fasten these locks, leaving them open (or con- 13. C. M. Suomivuori et al., Science 367, 881–887 (2020). LIH383 and provided input on binding and functional assays. A.C.
formationally heterogeneous), which might 14. L. M. Wingler et al., Cell 176, 468–478.e11 (2019). and M.S. performed and analyzed fluorescence binding assays
15. L. M. Wingler, R. J. Lefkowitz, Trends Cell Biol. 30, 736–747 and some b-arrestin and binding assays. L.J.L., M.M.C., E.I.A., and
disfavor b-arrestin recruitment (Fig. 4). (2020). L.M.M. performed and G protein and some b-arrestin assays under
Whereas NMR data and structural analysis 16. C. Wang et al., Science 340, 610–614 (2013). the supervision of J.D.M.. All authors provided comments on the
point to the association of b-arrestin recruit- 17. S. Rajagopal et al., Proc. Natl. Acad. Sci. U.S.A. 107, 628–632 manuscript. Competing interests: B.F.V. and F.C.P. have an
ment with increased stability, other studies (2010). ownership interest in Protein Foundry, LLC. R.H. is CSO of QVQ
18. J. M. Burns et al., J. Exp. Med. 203, 2201–2213 (2006). Holding BV. A patent application has been filed on “Novel Selective
show that G protein activation is associated 19. M. Gustavsson et al., Nat. Commun. 8, 14135 (2017). ACKR3 Modulators and Uses Thereof” (applicant: Luxembourg
with conformational heterogeneity at the intra- 20. M. Meyrath et al., Nat. Commun. 11, 3033 (2020). Institute of Health; inventors: A.C. and M.S.; PCT application no.:

RES EARCH | REPOR T
PCT/EP2020/061981). Data and materials availability: CCX777 exclusive licensee American Association for the Advancement of Tables S1 and S2
was supplied through a materials transfer agreement (MTA) with Science. No claim to original US government works. https://www. References (50–91)
Chemocentryx. LIH383 was supplied to B.F.V. through a MTA with science.org/about/science-licenses-journal-article-reuse MDAR Reproducibility Checklist
A.C. VUN701 was supplied to B.F.V. through a MTA with M.J.S.
All data are available in the main text or the supplementary materials. SUPPLEMENTARY MATERIALS View/request a protocol for this paper from Bio-protocol.
NMR peak assignments are deposited in the Biological Magnetic science.org/doi/10.1126/science.abj4922
Resonance Bank (BMRB) under entries 51451 to 51454. License Materials and Methods Submitted 16 May 2021; accepted 7 June 2022
information: Copyright © 2022 the authors, some rights reserved; Figs. S1 to S7 10.1126/science.abj4922

RES EARCH
ACTUATORS can be found in table S1. Butyl acrylate (BA)

and isobornyl acrylate (IBOA) were selected
A processable, high-performance dielectric elastomer as comonomers to lower the modulus and im-
prove the toughness of copolymers, respec-
and multilayering process tively (18). They were also important to reduce
the viscosity of prepolymer solutions (19).
Ye Shi1,2†, Erin Askounis1†‡, Roshan Plamthottam1†, Tom Libby3, Zihang Peng1, Kareem Youssef1, CN9021, a urethane diacrylate (UDA) with a
Junhong Pu1, Ron Pelrine3§, Qibing Pei1* high molecular weight, was selected as the
flexible long-chain cross-linker, and propoxy-
Dielectric elastomers (DEs) can act as deformable capacitors that generate mechanical work in response lated neopentyl glycol diacrylate (PNPDA) was
to an electric field. DEs are often based on commercial acrylic and silicone elastomers. Acrylics require used as the short-chain cross-linker (fig. S1).
prestretching to achieve high actuation strains and lack processing flexibility. Silicones allow for 2,2-dimethoxy-2-phenylacetophenone (DMPA)
processability and rapid response but produce much lower strains. In this work, a processable, high- and benzophenone (BP) were used as cophoto-
performance dielectric elastomer (PHDE) with a bimodal network structure is synthesized, and its initiators to ensure complete curing through
electromechanical properties are tailored by adjusting cross-linkers and hydrogen bonding within the the bulk and surface of the film.
elastomer network. The PHDE exhibits a maximum areal strain of 190% and maintains strains higher We developed systematic synthetic strat-
than 110% at 2 hertz without prestretching. A dry stacking process with high efficiency, scalability, and egies to tune the bimodal system’s stress-strain
yield enables multilayer actuators that maintain the high actuation performance of single-layer films. responses and viscoelasticity. First, short-chain
cross-linkers with softer and more-extended
D
chains were used to replace previously explored
ielectric elastomers (DEs) earned their resulting in progressively greater strain that short and stiff molecules, such as hexanediol
artificial muscle moniker because of proceeds until failure or induces unstable snap- diacrylate (HDDA), thus achieving more tun-
their large electrically induced actua- through. EMI on some DEs can be suppressed able stretchability and tensile strength (Fig. 1C
tion strain, high energy density, fast by applying a constant strain before actua- and figs. S2 and S3) (8). The cross-link density/
response speed, and mechanical com- tion, known as prestretching (8, 9). This, how- gel fraction (fig. S4) as well as the stress-strain
pliancy, all of which reproduce or in some ever, requires a rigid frame to maintain the responses were further controlled by changing
aspects exceed the multifunctional performance applied strain. the concentrations of the short-chain cross-
of natural muscles (1, 2). When a DE film is The contradiction between low driving volt- linker, PNPDA (Fig. 1C) (8). Without PNPDA,
sandwiched between compliant electrodes, it ages and high energy output also limits DEA the DE shows a similar long stress-strain
acts as a deformable capacitor and is known applications. Multilayer DEAs are able to scale plateau to that of non-prestretched VHB (fig.
as a dielectric elastomer actuator (DEA) (3). up force and energy outputs without increas- S5) (5, 6). With the bimodal network struc-
Under a voltage, the electric field across the ing voltages (2, 10, 11), but currently used wet ture, our DE stiffens after a critical stretch
DE generates a strong electrostatic interaction stacking methods—where an uncured DE film ratio in its non-Gaussian region (20), which
between the electrodes, known as a Maxwell is deposited on a cured DE—show low effi- can suppress EMI. This critical stretch ratio
stress, which compresses the film in the thick- ciency, poor scalability, low yield, and suboptimal shifts to a smaller value as the cross-link den-
ness direction and expands it in area. On the actuation performance as a result of layer non- sity increases. Second, the concentration of
basis of the working mechanisms, a high- uniformity. In addition, wet techniques are hydrogen bonds was optimized in the DE
performance DE should have sufficiently high not applicable for precured films, such as network to modify the viscoelasticity while
elastic strains, a large dielectric constant, high VHBs (12). maintaining its stress-strain relationship. A
dielectric strength, and an actuation stability There have been important advances in the small amount of acrylic acid (AA) comonomer
without premature failure. development of DE materials, including poly- (2.5 parts of weight) was added, which pro-
It has been challenging to tailor the electro- acrylates with optimized cross-linking networks vides side groups to form hydrogen bonds with
mechanical properties of an elastomer to meet (13), bottle-brush polymers (6), interpenetrating themselves, as well as with the –NH– groups
all these requirements, and the selection of DE networked elastomers (14), and high–dielectric on the CN9021 and PNPDA cross-linkers (fig.
materials is therefore limited (4). Commercial constant elastomers (15, 16). A bimodal-networked S6) (21). The mechanical loss factor of DE, a
3M very high bond (VHB) acrylate tapes and DE material was reported to suppress EMI with- measurement of the viscoelasticity, was then
silicone elastomer resins are the most widely out prestretching (8). However, it suffered from decreased from ~0.22 to ~0.11 at room temper-
used DE materials. VHBs, however, suffer from low maximum strains (<90% areal strain) and ature and low frequencies (fig. S7) without
high viscoelastic losses, whereas silicones tend high viscoelastic loss, which limited its energy changing the strain stiffening behavior (table
to exhibit low maximum strains and dielectric and power output. S2) and high elasticity (fig. S8). After AA is added,
strength (5). Additionally, many soft conven- We built a bimodal-networked elastomer hydrogen bonds partially replace covalent cross-
tional elastomers exhibit a long stress-strain using two cross-linkers with different chain links in the network. Hydrogen bonds act as
plateau (5, 6), and their performance is limited lengths and tailored its electromechanical weak, physical cross-links and can dynami-
by electromechanical instability (EMI) (3, 7). properties for high actuation performance. cally dissociate, leading to a lower glass transi-
Under constant voltage, the electric field in- DE films were fabricated through solution tion temperature (Tg) (fig. S9) and higher
creases as the DE film thickness decreases, processing and cured under ultraviolet (UV) chain mobility in the network (22). As more
light. The long chain segment in the bimodal AA is added, hydrogen bonds become highly
1
network ensures large elongation, and the sec- concentrated, and the density of covalent bonds
Department of Materials Science and Engineering, University ond relatively short chain segment raises the decreases (fig. S7), which results in inhibited
of California, Los Angeles, Los Angeles, CA, USA. 2ZJU-UIUC
Institute, Zhejiang University, Hangzhou, China. 3SRI modulus at modest strains to resist the rapidly actuation performance (fig. S10). At frequen-
International, Menlo Park, CA, USA. increased Maxwell stress during actuation and cies above 20 Hz, the storage modulus and loss
*Corresponding author. Email: qpei@seas.ucla.edu suppress EMI (Fig. 1A) (8, 17). Figure 1B shows factor of DE also increase rapidly (fig. S11).
‡Present address: Exponent, Inc., Menlo Park, CA, USA. the molecular structures of reactants. Detailed The static actuation performance of DEs was
§Present address: Pelrine Innovations, Longmont, CO, USA. formulations and nomenclature of samples tested in a diaphragm configuration (Fig. 1D)
Shi et al., Science 377, 228–232 (2022) 8 July 2022 1 of 5

RES EARCH | REPOR T
Fig. 1. Synthesizing bimodal-networked DEs with soft cross-linkers and prestretching. The complete curves of PNPDA 0 and VHB 4905 are shown in
hydrogen bonds for high-performance actuation. (A) Schematic illustration fig. S5. (D) Static actuation of PHDE, bimodal DE materials with different
of the design rules for DE with bimodal network structure. (B) Molecular concentrations of PNPDA, and VHB 4905 with 300% biaxial prestretching,
structures of acrylates and diacrylates used for the synthesis of bimodal- measured on a diaphragm. Carbon grease was used as the compliant electrodes.
networked DE. (C) Stress-strain relationships of PHDE, bimodal DE materials with (E) Cyclic actuation of PHDE and bimodal DE without AA under a 2-Hz square
different concentrations (weight %) of PNPDA or HDDA, and VHB 4905 without wave voltage of 4.5 kV for 100 cycles.
Fig. 2. Energy density and power density properties of PHDE. (A) Energy density of single-layer PHDE films measured at specified loads. (B) Energy density at different
frequencies. The driving voltage is 2.5 kV. (C) Average power density of single-layer PHDE tested in pure-shear mode with different loads. (D) Performance comparison
of PHDE with natural muscles and other DEAs reported in the literature. (E) Ball toss by the PHDE actuator. (F) A jumping robot fabricated with a PHDE actuator.

RES EARCH | REPOR T
(23). A moderate ratio of short-chain cross- an aluminum ball, which is 20 times as heavy precursor acting as a binding layer are sprayed
linker to long-chain cross-linker was found as itself, to 12.1 cm high (Fig. 2E and movie S3), on the glass/PHDE films. The PET/PHDE is
to be ideal for high actuation strains. The DE reaching an energy transfer efficiency of ~10% aligned onto a glass/PHDE/CNT/precursor
with 10 wt % of PNPDA exhibited a maximum (fig. S20). A miniature jumping robot based on and laminated in a vacuum laminator. After
strain of 189% with EMI fully suppressed, as a single-layer PHDE actuator reached a height UV curing the binding layer, the glass is re-
indicated by the modeling study (fig. S12). of 1.2 cm, which was ~67% of the theoretical moved and the two-layer PHDE stack, which
After the small amount of AA was added, the height of this jumping system (Fig. 2F, fig. S21, is now on PET, is aligned onto another glass/
DE maintained this high and stable actuation and movie S4). PHDE/CNT/precursor to add the third PHDE
and improved its response speed owing to To scale up the energy and power outputs layer. These processes are repeated until the
reduced viscoelasticity. Under a 2-Hz square of DEAs at low voltages, we developed a dry stack is fabricated. Figure 3B and fig. S22 show
wave voltage, the DE with AA exhibited an av- stacking method to fabricate multilayer DEAs. the pictures of a 10-layer PHDE stack with an
erage areal actuation of 110%, which is su- This method was demonstrated with PHDE array of 20 actuators. Its cross section was ex-
perior to those of the DE without AA and of but is applicable for other DEs, including pre- amined by scanning electron microscopy (SEM),
other previously reported DEs (Fig. 1E, movie formed DE films. As illustrated in Fig. 3A, to and the image in Fig. 3C shows uniform thick-
S1, and table S3). Thus, the DE with 10 wt % of fabricate an N-layer PHDE stack, N-1 films are ness of PHDE layers and solid sealing between
PNPDA and an additional 2.5 parts of AA is cast on glass, and one film is prepared on poly- them. The interfacial electrode with inter-
finalized as the processable, high-performance ethylene terephthalate (PET), with uniform penetrating structure allowed strong bonding
dielectric elastomer (PHDE) composition for thickness. Two substrates were selected to pro- between layers, which was proved by a swelling
film preparation and device fabrication. PHDE vide different mechanical properties and ad- test (fig. S23), and the stress-strain curve and
showed high dielectric strength with an ap- hesions, thus avoiding delamination during dynamic moduli of a 10-layer PHDE stack are
parent electrical breakdown field of 330 V/mm, stacking. Carbon nanotube (CNT) electrodes very close to those of single-layer films (fig.
low leakage currents (fig. S13), a high dielec- and a thin layer of uncured acrylic polymer S24). The dielectric constant of the PHDE
tric constant of 5.35, and a small dielectric
dissipation factor of 0.014 at 1 kHz (fig. S14).
The force output, energy density, and power
density of the single-layer PHDEs were tested
in a pure-shear mode (fig. S15) (23, 24). The
film measured 6 cm by 1 cm by 40 mm and
~0.030 g of active material. Under preloaded
isometric conditions, PHDE was able to output
a stable force of ~0.95 N (~0.40 MPa stress) at
2.5 kV with a 1.2 N preload without prestretch-
ing (fig. S16A). The force maintained ~0.67 N
at 2 Hz and ~0.4 N at 20 Hz (fig. S16B). Under
isotonic conditions, PHDE films performed a
specific work of ~35 J/kg with a 50-g load and
~88 J/kg with a 100-g load during contraction
at 2.5 kV (Fig. 2A and movie S2). This maxi-
mum energy density is more than twice that of
natural muscles (0.4 to 40 J/kg). PHDE films
maintained energy densities of ~67 J/kg at
5 Hz and ~16 J/kg at 20 Hz with a 100-g load
(Fig. 2B and movie S2), and the corresponding
power densities were calculated to be ~670 W/kg
and ~640 W/kg, respectively (Fig. 2C) (24).
The peak energy and power density at 10 Hz
was caused by resonant responses of planar
actuator systems (24, 25). These power den-
sities are about one order of magnitude as
high as those of natural muscles (26) and can
be achieved in the 101-Hz range or lower. At
frequencies greater than or equal to 50 Hz,
the films were not able to effectively perform
work. Notably, the blocked force and energy
density can be tuned by adjusting the con-
centration of short-chain cross-linkers for dif-
ferent applications (figs. S17 and S18). PHDE
exhibited high stability when performing work
(fig. S19). The maximum energy and power
output densities of PHDE are superior to Fig. 3. Multilayer PHDE actuators enabled by dry stacking method. (A) Diagram of dry stacking for
those of natural muscles and DEAs developed multilayer PHDE actuator fabrication. (B) Photo of a 10.16 cm by 12.7 cm (4 inch by 5 inch) 10-layer PHDE
in recent years (Fig. 2D) (2, 24, 27–31), which stack with 20 actuators. (C) SEM images of cross-sectional view of a PHDE stack. (D) Static actuation
enable lightweight, energy-dense actuation sys- of single-layer and 10-layer PHDE films, tested on a diaphragm. CNT was used as compliant electrodes.
tems. A PHDE film was demonstrated to toss (E) Frequency response of single-layer and 10-layer PHDE actuators under 3.5 kV (square wave).

RES EARCH | REPOR T
stack is measured to be 5.37, and its dissipa-

tion factor increases to 0.020 at 1 kHz, possibly
because of the dielectric loss at interfaces.
Compared with wet stacking methods
(2, 10, 11), our dry stacking method shows
many advantageous features. It is compatible
with established, large-scale materials process-
ing techniques, and the film dimensions and
thickness can be easily tuned (fig. S25). Addi-
tionally, this method enables a high yield by
screening individual films before stacking
(fig. S26). Because film preparation, electrode
deposition, and lamination can be conducted
in parallel, the process is very efficient. The
efficiency can be further improved by mod-
ularly dry stacking PHDE stacks on each other,
as shown in fig. S27. The actuation strain of
10-layer PHDE actuators is comparable to that
of single-layer actuators at the same low driv-
ing voltages (Fig. 3D). The strain is slightly
lower at higher voltages, possibly because of
greater stiffness from the inactive border sur-
rounding the 10-layer actuators (movie S5) and
decreased conductivity of CNT electrodes at
highly stretched states. Moreover, the multi-
layer actuator maintained high response speed
and reached strains of ~110% at 2 Hz and ~60%
at 20 Hz (Fig. 3E). Additionally, our dry stack-
ing method can linearly scale up the force,
energy, and power outputs with the number of
layers and experiences less than a 20% reduc-
tion (fig. S28).
The multilayer PHDE stacks enable new
actuator designs. We fabricated spider actua-
tor, which were identified as a potential DEA
design but never realized, in part because they
are not compatible with high prestretching.
The basic concept uses radially symmetric in- Fig. 4. Actuators enabled by multilayer PHDE. (A) A PHDE spider actuator lifts a 200-g load at 2.5 kV.
clined legs to couple axial force to radial load- (B) Schematic illustration of fabrication of PHDE roll actuators. The inset shows the photos of a PHDE
ing of a circle of DE film, giving the linkage roll. (C) A PHDE roll actuator lifts a 100-g load at 0.5 Hz. (D) Photograph of a PHDE roll pump.
nonlinear mechanical advantage that can be (E) Normalized flow rates (L/L0) during various mechanical manipulations of the pump. The pump is
tuned to maximize performance. We built one driven at 2.5 kV at 10 Hz while bent at 90°, twisted axially, and hammered. The initial flow rates and the
instantiation of this linkage, tuned for high recovery flow rates are also shown.
axial force, and tested it with four-layer PHDE
stacks with an active area of ~4.8 cm2. As shown
in Fig. 4A and movie S6, the spider actuator However, the roll is made entirely of poly- rate, compactness, light weight, low noise,
successfully lifted a 200-g load with ~3.0-mm mers (except for a small amount of conductive and high mechanical compliancy make it a
linear displacement at 2.5 kV, reaching a spe- materials). It is highly robust and remains promising device for flexible medical implants
cific work of ~74 J/kg. functional after they are bent, twisted, and and soft robotics.
We also designed multifunctional roll actua- hammered (fig. S30 and movie S8). We further
tors. As shown in Fig. 4B, a two-layer PHDE integrated two check valves to build a tubular REFERENCES AND NOTES
strip was rolled into a free-standing tube with pump from a PHDE roll (Fig. 4D). The flow 1. R. Pelrine, R. Kornbluh, Q. Pei, J. Joseph, Science 287,
a total of ~18 active layers. The roll was first rate (L) could be tuned by controlling the 836–839 (2000).
characterized as a linear actuator (fig. S29) driving voltage and frequency. The peak flow 2. M. Duduta, E. Hajiesmaili, H. Zhao, R. J. Wood, D. R. Clarke,
Proc. Natl. Acad. Sci. U.S.A. 116, 2476–2481 (2019).
and lifted a 100-g load by 1.4 mm at 0.5 Hz, rate reached ~20.4 mL/min at 2.5 kV and 3. Z. Suo, Acta Mech. Solida Sin. 23, 549–578 (2010).
which translated to an output energy den- 10 Hz, which translated to a specific flow rate 4. L. J. Romasanta, M. A. Lopez-Manchado, R. Verdejo,
sity of 15.6 J/kg (Fig. 4C and movie S7). The of ~100,000 mL/kg per minute (fig. S31 and Prog. Polym. Sci. 51, 188–211 (2015).
displacement at 5 Hz was 1.1 mm (fig. S29C movie S9). This value is about two orders of 5. Y. Qiu, E. Zhang, R. Plamthottam, Q. Pei, Acc. Chem. Res. 52,
316–325 (2019).
and movie S7), with a power density calculated magnitude higher than that of a miniature 6. M. Vatankhah-Varnoosfaderani et al., Adv. Mater. 29, 1604209
to be 122 W/kg. These values are lower than pump (TCS Micropumps, MGD 1000S) (32). (2017).
those determined in the pure-shear mode Owing to its robust nature, the PHDE roll 7. X. Zhao, Z. Suo, Phys. Rev. Lett. 104, 178302 (2010).
8. X. Niu et al., J. Polym. Sci. B Polym. Phys. 51, 197–206
and the spider actuators, possibly because pump showed nearly full performance recov-
(2013).
the inactive layers restrict its linear strain, ery after it was bent, twisted, or hammered 9. L. Jiang, A. Betts, D. Kennedy, S. Jerrams, J. Phys. D
and the circumferential strain is not coupled. during operation (Fig. 4E). Its tunable flow Appl. Phys. 49, 265401 (2016).

RES EARCH | REPOR T
10. M. Duduta, R. J. Wood, D. R. Clarke, Adv. Mater. 28, 24. E. Acome et al., Science 359, 61–65 (2018). directed the project. Y.S., E.A., R.Pl., R.Pe., T.L., and Q.P. designed
8058–8063 (2016). 25. C. Cao et al., Soft Sci. 1, 1 (2021). the actuators, analyzed the data, discussed the results, and wrote
11. S. Rosset, H. R. Shea, Appl. Phys. Rev. 3, 031105 (2016). 26. S. I. Rich, R. J. Wood, C. Majidi, Nat. Electron. 1, 102–112 the manuscript. Competing interests: Q.P., E.A., and Y.S. are
12. G. Kovacs, L. Düring, S. Michel, G. Terrasi, Sens. Actuators (2018). applying for patents related to the described work. The other
A Phys. 155, 299–307 (2009). 27. Y. Chen et al., Nature 575, 324–329 (2019). authors declare that they have no competing interests. Data and
13. L.-J. Yin et al., Nat. Commun. 12, 4517 (2021). 28. X. Ji et al., Sci. Robot. 4, eaaz6451 (2019). materials availability: The data that support the findings of this
14. S. M. Ha, W. Yuan, Q. Pei, R. Pelrine, S. Stanford, Adv. Mater. 29. H. S. Jung et al., Int. J. Control. Autom. Syst. 15, 25–35 study are available in the main text and the supplementary
18, 887–891 (2006). (2017). materials. License information: Copyright © 2022 the authors,
15. C. Ellingford, C. Bowen, T. McNally, C. Wan, Macromol. Rapid 30. S. Xu, Y. Chen, N.-P. Hyun, K. P. Becker, R. J. Wood, Proc. Natl. some rights reserved; exclusive licensee American Association
Commun. 39, e1800340 (2018). Acad. Sci. U.S.A. 118, e2103198118 (2021). for the Advancement of Science. No claim to original US
16. P. Caspari, S. J. Dünki, F. A. Nüesch, D. M. Opris, J. Mater. 31. H. Zhao et al., Adv. Funct. Mater. 28, 1804328 (2018). government works. https://www.science.org/about/science-
Chem. C 6, 2043–2053 (2018). 32. V. Cacucciolo et al., Nature 572, 516–519 (2019). licenses-journal-article-reuse
17. C. Tugui et al., J. Mater. Chem. C 3, 8963–8969 (2015).
18. M. Hayashi, S. Matsushima, A. Noro, Y. Matsushita, Macromolecules ACKN OWLED GMEN TS SUPPLEMENTARY MATERIALS
48, 421–431 (2015). We acknowledge R. Kornbluh of SRI International for valuable science.org/doi/10.1126/science.abn0099
19. P. K. T. Oldring, Ed., Chemistry & Technology of UV & EB suggestions in designing the ball-tossing and jumping-robot Materials and Methods
Formulations for Coatings, Inks & Paints, vol. 2 (SITA experiments. Funding: This work was supported by the Defense Supplementary Text
Technology, 1991). Advanced Research Projects Agency (DARPA), award no. Figs. S1 to S31
20. M. Rubinstein, S. Panyukov, Macromolecules 35, 6670–6686 HR001119C0043. Author contributions: Y.S., E.A., R.Pl., and Q.P. Tables S1 to S3
(2002). conceived the idea. E.A., Y.S., and R.Pl. performed materials References (33–44)
21. H. Guo, Y. Han, W. Zhao, J. Yang, L. Zhang, Nat. Commun. 11, synthesis, characterizations, and thin film preparation. Y.S. and Movies S1 to S9
2037 (2020). R.Pl. performed device fabrication and characterizations, including
22. C. L. Lewis, K. Stewart, M. Anthamatten, Macromolecules 47, multilayer stacks, planar jumping robots, and roll actuators. Submitted 27 October 2021; resubmitted 27 March 2022
729–740 (2014). T.L. fabricated and tested spider actuators. K.Y., Z.P., and J.P. Accepted 1 June 2022
23. F. Carpi et al., Smart Mater. Struct. 24, 105025 (2015). helped with film fabrication and material characterizations. Q.P. 10.1126/science.abn0099

WORKING LIFE
By Hannah Gurholt
Revising my mental health checklist
“S
truggling Ph.D. student checklist, December 2021: Anxiety. Caffeine addiction. Depression.
Emotional support cat that then causes stress. Imposter syndrome. Stress twitches. Hair loss.”
When I recently came across these words, I couldn’t believe I had written them just 6 months
earlier. Since then, I have tried three different antidepressants and acquired an unruly num-
ber of pill cutters. I have experienced migraines, drowsiness, weight loss, weight gain, and
fatigue. At times I have still felt numb, lonely, and frustrated. But I have also felt hopeful and
reinvigorated, confident in the knowledge that I am on the path of healing.
My troubles started early in gradu- take the plunge and start on antide-
ate school. In addition to the typi- pressants. After the appointment, I
cal stressors of starting a Ph.D., felt hopeful—an emotion I hadn’t
my advisers decided to move to a experienced in months. I called my
new university. They wanted me parents and assured them I would
to go with them, whereas many of be all right.
my classmates and other profes- My optimism was temporar-
sors wanted me to stay. I felt pulled ily dampened when, shortly after
in two directions, and as a devout starting the medicines, I was hit
people-pleaser, my ever-present with intense migraines. But soon
anxiety began to rear its ugly head. I began to see real benefit. I was
I started seeing a therapist to sleeping again. My nausea dissi-
help unpack the loaded question of pated and my appetite returned.
what I wanted to do with my life. I My eyebrows grew back. My smile
shared my hesitancy about medica- felt more genuine and my laughter
tion, having lost a beloved family less forced. I no longer felt like an
member to suicide while they were imposter in my own body. I picked
taking antidepressants. My thera-
pist and I decided we would explore
“Antidepressants aren’t up my old hobbies of running, read-
ing, drawing, and doing puzzles,
every other option. We practiced
coping mechanisms for when I felt
magic pills … but they helped me and developed new ones, including
cleaning and being an obnoxious
particularly anxious. We discussed when nothing else could.” cat mom. I returned to therapy
how to respond to the never-ending with newfound light in my eyes. I
questions from colleagues about whether I was moving. I left began to work from local coffee shops, which helped keep
each session with a positive feeling in my stomach. me motivated, focused, mentally charged, and caffeinated.
But despite our efforts, anxiety and depression continued I visited my advisers’ new university and finally decided to
to creep up on me. With classes being held online due to move with them, lifting a massive weight off my shoulders.
the pandemic, I spent my days at a desk in the corner of my That isn’t to say everything was suddenly perfect. The first
bedroom, searching the familiar walls for any distraction medications made me feel emotionally numb, so my doc-
from reading papers. I spent nights lying awake with my tor and I tried a different one, which hit me with extreme
thoughts. When I did sleep, I woke to frightening anxiety. drowsiness and weight gain. But my doctor and I will keep
I battled nausea, appetite loss, fatigue, panic attacks, and working to find the medications that work best for me, and
lack of motivation. I lost my eyebrow hair. My period be- I will continue therapy as well. Antidepressants aren’t magic
came irregular to the point of spotting every day, with ane- pills, and the side effects are no joke. But they helped me
mia close behind. My friends and family were concerned when nothing else could.
ILLUSTRATION: ROBERT NEUBECKER
about my rapid weight loss. When a panic attack forced me These days, I have a new checklist. Healing Ph.D. student
to miss an important meeting with my advisers, I sched- checklist, July 2022: Drowsiness. Caffeine addiction. Ap-
uled an emergency session with my therapist. She could tell petite changes. Weight fluctuations. Newfound motivation.
everything she needed to know from my eyes, glassy and Transparency. Hope. j
sunken with defeat. It was time to consider medication.
I went to a doctor, and with her support, I decided to Hannah Gurholt is a Ph.D. student at Washington University in St. Louis.

I NS I GHTS
PRIZE ES SAY
GRAND PRIZE WINNER
Jennifer Hampton Hill
Jennifer Hampton
Hill received an
undergraduate
degree from Cali-
fornia State Poly-
technic University
Humboldt and her
PhD in biology from the University
of Oregon in Eugene in Dr. Karen
Guillemin’s laboratory. She is
currently finishing a postdoctoral
fellowship at the University of Utah
in Salt Lake City in the laboratories
of Drs. June Round and Charles
Murtaugh. Her research focuses
on host-microbe interactions
that shape pancreatic develop-
ment and health. www.science.org/
doi/10.1126/science.abq6051
Micrograph of mouse pancreatic islets composed of insulin-producing b cells and other hormone-secreting cells
MICROBIOME
From bugs to b cells

Do ancient microbial irritants offer early life
protection against diabetes?
By Jennifer Hampton Hill acterized by the destruction of pancreatic
insulin-producing b cells (see the figure).
PHOTOS: (LEFT TO RIGHT) COURTESY OF RACHEL MERRILL; JENNIFER HAMPTON HILL
O
ne-hundred years ago, the dis- However, a cure still evades discovery, and
covery of the hormone insulin by its realization will require a dual approach
Frederick Banting and Charles Best that tempers autoimmunity while simulta-
transformed the prognosis of type neously restoring endogenous insulin pro-
1 diabetes (T1D). Practically over- duction. Focusing on the latter, our work
night, a diagnosis of T1D went from revealed that the resident microbiome is a
meaning certain suffering and death to promising source of previously undiscov-
being a manageable, albeit difficult, condi- ered modulators of host b cells.
tion that could result in a long and full life. Animals evolved within a microbial
Over the century since that Nobel Prize– world, and it is plausible that early multi-
winning discovery, there has been monu- cellular life used microbial cues to gather
mental progress in the understanding of information about the environment, e.g.,
the basic etiology of diabetes, which is now local nutrient availability, to set their me-
known to be an autoimmune disease char- tabolisms (1, 2). Many animals, including
fruit flies and mice, have reduced meta-
bolic rates when grown in the absence
Departments of Pathology and Human Genetics,
University of Utah, Salt Lake City, UT, USA. of microbes—an artificial situation that
Email: jennifer.hill@path.utah.edu could be interpreted by these organisms

as a dearth of nutrients (3). We investi- FINALIST important functions, such as to stimulate
gated whether animals use microbial cues Apollo Stacy developmental programs (11). In the case
to tune the number of cells producing the Apollo Stacy re- of T1D, decreased gut microbiome diversity
conserved metabolic regulator, insulin. To ceived his under- is characteristic of and can be predictive
optimize energy needed for development graduate degree of disease onset, suggesting that at-risk
of these cells (4), animals need to match from Washington children may be missing specific microbes
insulin production capacity to environ- University in St. with disease-mitigating functions (12, 13).
mental nutrient availability. Thus, we hy- Louis and his PhD The BefA-containing Enterobacteriaceae
pothesized that hosts incorporate informa- from The Univer- bacteria are abundant colonizers of the
tion from their resident microbiomes to sity of Texas at Austin. Currently a healthy infant gut during peak neonatal
accomplish this optimization. postdoctoral fellow at the National b cell expansion (6, 14). Failure to develop
We leveraged a vertebrate zebrafish Institutes of Health, Apollo will be a sufficient b cell mass in early childhood
model to study development of pancreatic starting his laboratory at the Cleve- can lead to metabolic disorders, which sug-
b cells with or without microbiota, by com- land Clinic Lerner Research Institute gests that missing out on microbial cues,
paring sterile or germ-free (GF) larvae and in 2022. His research investigates such as BefA, may predispose individuals
their conventionally reared (CV), microbe- how host-derived metabolites shape to diabetes. If so, fortification of these mi-
carrying counterparts. We found that CV the ecological balance between crobial activities in children who carry risk
commensals and pathogens and thus
larvae had significantly more b cells than alleles for T1D could be a strategy to pre-
host susceptibility to inflammatory
GF larvae, despite both groups having vent or delay disease.
disease. www.science.org/doi/10.1126/
equivalent nutrient supplies from their Bacteria did not evolve BefA to prevent
science.abq6054
yolk sacks (5). Taking a reductionist ap- T1D in their hosts. Long before animals de-
proach, we systematically added back indi- veloped Reg proteins, bacteria were mak-
FINALIST
vidual zebrafish gut bacteria and their se- ing AMPs to compete with other microbes.
creted products until we identified a single Irina Leonardi Animal cells have always had to sense and
protein that was sufficient to restore GF Irina Leonardi defend themselves against bacterial mem-
b cell mass (5). The protein we identified received under- brane disrupters. But information could
was previously unknown, and we named it graduate degrees be gleaned amidst the onslaught. Because
b cell expansion factor A (BefA). from ETH Zürich bacterial growth and competition is most
Searches for BefA-related sequences un- and a PhD from the pronounced in nutrient-rich environments,
University of Zürich.
covered homologs in a handful of bacteria the intensity of membrane attack could be
She performed her
of the Enterobacteriaceae family commonly used to gauge metabolic needs. Perhaps
postdoctoral work in the laboratory
found in healthy mammalian microbiomes insulin-producing cells in early animals
of Dr. Iliyen Iliev at Weill Cornell Medi-
after birth, when b cells normally undergo cine. Her research investigated the
were particularly sensitive in response to
substantial expansion (5, 6). To test whether cellular mechanisms of fungal rec- AMPs because damage to these precious
BefA might elicit similar responses in mam- ognition in the gut and the local and cells would jeopardize survival. Possibly,
malian species, we analyzed b cell devel- systemic consequences of intestinal animals evolved their own Reg-like pro-
opment in gnotobiotic mouse models. Not fungal colonization. She is currently teins, with the dual functions of protecting
only did we find that newborn GF mice had the Scientific Communications Lead against microbial invaders and stimulating
the same paucity of b cells as GF larval fish, at Immunai in New York. www.science. nutrient utilization. Understanding cur-
but purified BefA was sufficient to increase org/doi/10.1126/science.abq6056 rent declining microbiome diversity as a
their developing b cell mass (7). loss of ancient developmental cues allows
We noted that purified BefA elicited its us to imagine future strategies to correct
effect in GF zebrafish and mice whether microbes (9). Notably, these proteins were the environmental information conveyed
it was delivered orally or intraperitone- originally discovered in pancreatic islets during human pancreas development to
ally, suggesting that it might have traveled for their capacity to regenerate b cells, but ward off T1D.
to the pancreas (7). We found that BefA this capacity has not been connected with
could induce proliferation directly in pri- their antimicrobial activity. We explored
1. T. J. Wiles, K. Guillemin, Curr. Opin. Microbiol. 54, 87
mary islet cultures from both hosts (7). this by creating amino acid substitutions (2020).
PHOTOS: (TOP TO BOTTOM) APOLLO STACY; COURTESY OF IRINA LEONARDI
Although small microbial metabolites have at several predicted lipid-binding residues 2. T. C. G. Bosch, M. McFall-Ngai, Curr. Top. Dev. Biol. 141,
frequently been shown to disseminate sys- in BefA. One of these substitutions de- 399 (2021).
3. G. Storelli et al., Cell Metab. 14, 403 (2011).
temically, this is the first observation of the creased BefA’s ability to both disrupt mem- 4. L. C. Murtaugh, Development 134, 427 (2007).
far-reaching effects of a much larger pro- branes and stimulate b cell turnover (7). 5. J. H. Hill, E. A. Franzosa, C. Huttenhower, K. Guillemin,
tein secreted by gut bacteria (8). Conversely, Reg3 was sufficient to induce eLife 5, e20145 (2016).
6. B. E. Gregg et al., J. Clin. Endocrinol. Metab. 97, 3197
To better understand how BefA affects a potent increase in b cell turnover from (2012).
b cells, we resolved the protein’s crystal mouse islets, illustrating both the necessity 7. J. H. Hill et al., bioRxiv 10.1101/2022.03.24.485696
structure. This structure revealed a previ- and sufficiency of membrane perturbation (2022).
8. J. H. Hill, J. L. Round, Cell 184, 2796 (2021).
ously unknown protein fold for a putative in b cell expansion (7). 9. S. Mukherjee, L. V. Hooper, Immunity 42, 28 (2015).
lipid-binding domain that was sufficient The hygiene hypothesis posits that cer- 10. D. P. Strachan, BMJ 299, 1259 (1989).
for BefA’s activity (7). We next deter- tain skyrocketing diseases, such as T1D, 11. M. J. Blaser, S. Falkow, Nat. Rev. Microbiol. 7, 887 (2009).
12. A. D. Kostic et al., Cell Host Microbe 17, 260 (2015).
mined that BefA was capable of binding are the result of changing societal practices
13. P. Zheng, Z. Li, Z. Zhou, Diabetes Metab. Res. Rev. 34,
to membrane lipids and perturbing lipid that have reduced microbial exposures and e3043 (2018).
bilayers—a trait common to antimicrobial decreased microbiome diversity (10). As 14. C. J. Stewart et al., Nature 562, 583 (2018).
proteins (AMPs), like the Reg proteins, se- a result, humans may have lost microbes
creted in the gut to prevent incursion by that were relied on in the past to perform 10.1126/science.abq6051

Science - July 8, 2022 USA

Uploaded by

Copyright:

Available Formats

Science - July 8, 2022 USA

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Science - July 8, 2022 USA

Uploaded by

Copyright:

Available Formats

E D I TOR IAL

Inclusion doesn’t lower standards

removing symbols of exclusion is helpful, why wouldn’t a

SCIENCE science.org 8 JULY 2022 • VOL 377 ISSUE 6602 129

IN BRIEF Colombia hits conservation goal

other lawsuits that would further reduce

New water lily species named

134 8 JULY 2022 • VOL 377 ISSUE 6602 science.org SCIENCE

Scientists find new hummingbird colors

SCIENCE science.org 8 JULY 2022 • VOL 377 ISSUE 6602 135

Scientists decry reversal of U.S. abortion rights

136 08 JULY 2022 • VOL 377 ISSUE 6602 science.org SCIENCE

advantaged backgrounds and may not be MICROBIOLOGY

SCIENCE science.org 8 JULY 2022 • VOL 377 ISSUE 6602 137

Lab studies show that one of the coronavirus’ protein-

United States issued only 40,000 or fewer

Bad news for Paxlovid? reported in JBC Accelerated Communica-

Resistance may be coming SARS-CoV-2 variants, including Alpha,

138 8 JULY 2022 • VOL 377 ISSUE 6602 science.org SCIENCE

SCIENCE science.org 8 JULY 2022 • VOL 377 ISSUE 6602 139

140 8 JULY 2022 • VOL 377 ISSUE 6602 science.org SCIENCE

Mass layoff looms for Japanese researchers

SCIENCE science.org 8 JULY 2022 • VOL 377 ISSUE 6602 141

PHOTO: CREDIT GOES HERE AS SHOWN; CREDIT GOES HERE AS SHOWN

142 8 JULY 2022 • VOL 377 ISSUE 6602 science.org SCIENCE

By Gabriel Popkin, in Chile’s Villarrica National Park

SCIENCE science.org 8 JULY 2022 • VOL 377 ISSUE 6602 143

sity who advises SPUN. “It’s amazing there

APPRECIATED OR NOT, fungi are integral to

through mycelial networks. In one highly

144 8 JULY 2022 • VOL 377 ISSUE 6602 science.org SCIENCE

SCIENCE science.org 8 JULY 2022 • VOL 377 ISSUE 6602 145

146 8 JULY 2022 • VOL 377 ISSUE 6602 science.org SCIENCE

At another stop, Furci charmed a couple

SCIENCE science.org 8 JULY 2022 • VOL 377 ISSUE 6602 147

At another stop, Furci charmed a couple

SCIENCE science.org 8 JULY 2022 • VOL 377 ISSUE 6602 147

148 8 JULY 2022 • VOL 377 ISSUE 6602 science.org SCIENCE

SCIENCE science.org 8 JULY 2022 • VOL 377 ISSUE 6602 149

of native ventricular microanatomy from dif-

150 8 JULY 2022 • VOL 377 ISSUE 6602 science.org SCIENCE

of native ventricular microanatomy from dif-

150 8 JULY 2022 • VOL 377 ISSUE 6602 science.org SCIENCE

SCIENCE science.org 8 JULY 2022 • VOL 377 ISSUE 6602 151

MEMBRANES transport to occur. As solvent concentration

Bridging membrane inside the swollen polymer, the mechanis-

transport models ence of continuous solvent pathways. The

152 8 JULY 2022 • VOL 377 ISSUE 6602 science.org SCIENCE

Less is more when forming gels by dilution

A dilution-responsive supramolecular gel-sol-gel-sol cascade Orthogonal self-assembly motifs

Dilute Dilute Dilute

Concentrate Concentrate Concentrate

SCIENCE science.org 8 JULY 2022 • VOL 377 ISSUE 6602 153

pramolecular network. This component com- EVOLUTION

9. J. A. Riback et al., Nature 581, 209 (2020).

ACK N OW LED GMENTS

154 8 JULY 2022 • VOL 377 ISSUE 6602 science.org SCIENCE

SCIENCE science.org 8 JULY 2022 • VOL 377 ISSUE 6602 155

vibration, temperature, and pain—get re- through corticofugal projections arising

156 8 JULY 2022 • VOL 377 ISSUE 6602 science.org SCIENCE