5 JANUARY 2024, VOL 383, ISSUE 6678 Science

Citation preview

EDI TO R I A L

Genuine images in 2024

I

the pilot period, authors generally provided a satisfactory response and fixed the problems so that the paper could proceed to further review or acceptance. However, papers that should not be published were detected. Going forward, if the authors’ response is unsatisfactory or raises additional concerns, we will probe further and take steps that could include rejecting the paper. If image integrity concerns are raised about a paper that a Science journal already published, we will use Proofig to carefully examine the suspicious images, which will inform subsequent actions (e.g., correction or retraction). Among other changes at the journals, Ali Shilatifard, the academic editor of Science Advances, has decided to step down after 10 years of exemplary service. Because of the outstanding growth of the journal, five new section editors have been appointed to lead social and interdisciplinary science; neuroscience; Earth, environmental, ecological, and space sciences; physical and materials sciences; and biomedicine and life sciences. Meanwhile, in this issue of Science, the News staff highlights areas in research and policy that are likely to attract attention in 2024. From the effects of El Niño on global temperatures to forging AI regulation across the world to the outcome of a US presidential election, there will be no shortage of events and issues to keep an eye on. Science will cover all of this and more, in its research, news, and commentary sections. And Science’s newsletter, ScienceAdviser, will deliver exciting highlights from the world of science and Science every weekday morning. In addition, this month, the Science Press Package team— the group responsible for presenting content from the Science journals to thousands of journalists—will launch a weekly press package called “Policy Pak” that will be filled with policy-oriented content from across Science’s six journals. It will be aimed at the producers of the many newsletters relied on by policy-makers to keep abreast of scientific issues. As I wrote in my end-of-year editorial last month, 2024 will bring a lot of challenges. Through better monitoring for research errors, careful curation of content, and sharing information with policy-makers, we hope to build stronger trust and integrity in science in the coming year.

H. Holden Thorp Editor-in-Chief, Science journals. [email protected]

p

n recent years, the research community has become increasingly concerned with issues involving the manipulation of images in scientific papers. Some of these alterations—involving images from experimental techniques such as microscopy, flow cytometry, and western blots—are inadvertent and may not change the conclusions of papers. But in rare cases, some are done deliberately to mislead readers. Image sleuths who can detect these alterations, like the scientific integrity consultant Elisabeth Bik, have risen to prominence, as has the website PubPeer, where many of the detected flaws are posted. High-profile incidents, such as one involving the laboratory of former Stanford University President Marc Tessier-Lavigne, have eroded public confidence in science and harmed careers of investigators who missed doctored images coming from their own laboratories. To address these problems, in 2024, the Science family of journals is adopting the use of Proofig, an artificial intelligence (AI)–powered image-analysis tool, to detect altered images across all six of the journals. Proofig screens images for duplication and other types of manipulation. It is similar to the iThenticate plagiarism detection software (which Science has been using for 7 years), but it works on images rather than text. Science has been piloting Proofig for several months with clear evidence that problematic figures can be detected prior to publication, so its use will expand to all papers under consideration that present relevant images. This should help identify both honest mistakes and fraudulent activity before a decision is made on publication. Prior to the pilot phase, Science had been conducting “human-eye” image checks on a portion of papers, so Proofig is a natural next step. This tool will enhance Science’s review process and reduce the potential for human error because it captures many more alterations. Proofig will be applied after a research paper is revised by authors. After analyzing the images, the tool generates a report flagging duplications and other abnormalities, such as rotation, scale distortion, and splicing. The paper’s editor reviews the findings and determines whether the AI-detected issues may be problematic. (In some cases, figures may have intentional rotations or duplications that are explained in the paper.) If so, the editor contacts the authors to request an explanation. During

y y g y ,

PHOTO: CAMERON DAVIDSON

g

“…we hope to build stronger trust and integrity in science in the coming year.”

–H. Holden Thorp

10.1126/science.adn7530

SCIENCE science.org

5 JANUARY 2024 • VOL 383 ISSUE 6678

7

NEWS p g y

Edited by Jeffrey Brainard

In October 2023, boats on Puraquequara Lake in Brazil were stranded after drought lowered the water level.

y g

IN BRIEF

y

AREAS TO WATCH

P

olitical upheaval and policy disputes could create uncertainties for scientists this year as a momentous U.S. election looms and European parties grapple over green policies, for example. As the effects of the pandemic caused by SARSCoV-2 continue to recede, trials of Long Covid treatments may serve up their first fruits. The Pacific Ocean warming called El Niño will likely help boost global temperatures to new records, providing a continual reminder that time is running out for science-based interventions to keep the world from warming beyond “safe” thresholds. Science’s news staff forecasts other areas of research and policy likely to make news this year.

8

5 JANUARY 2024 • VOL 383 ISSUE 6678

C L I M AT E S C I E N C E | An El Niño warming in the eastern Pacific Ocean is likely to grow stronger over the next few months and might help push the average global surface temperature 1.5°C above preindustrial levels for the first time. Part of a regularly oscillating climate pattern, the El Niño is expected to worsen drought in the Amazon and Australia. The shift, which began last year, is suspected of helping make 2023 the hottest year in modern history, as temperatures in the first 11 months averaged more than 1.4°C above preindustrial levels—higher than rising emissions of greenhouse gases alone can explain. El Niño has suppressed

sciencemag.org SCIENCE

,

What’s coming up in 2024

El Niño may extend record heat

calms a hyperactive immune system, and vagus nerve stimulation recalibrates the autonomic nervous system, potentially ameliorating symptoms such as brain fog and abnormal heart rate. Even if these therapies aren’t successful, scientists hope trial results will clarify the biology fueling the condition, helping them identify what to test next.

the ocean’s ability to absorb heat, and a drop in Sun-blocking pollution also played a role—effects that will continue this year.

U.S. politics shape science

PHOTO: JAIME SALDARRIAGA/AP

| Four years after the pandemic caused by SARS-CoV-2 began, millions of people are debilitated by Long Covid, a syndrome that includes crushing fatigue, persistent headaches, and shortness of breath. With no treatments clearly confirmed to help, patients and their doctors are experimenting with various drugs and dietary supplements. This year, scientists hope some of the first rigorous placebo-controlled clinical trials of possible Long Covid treatments will report outcomes, though they may be preliminary. Some, such as Paxlovid, other antiviral drugs, and monoclonal antibodies, target the SARS-CoV-2 virus itself, which may persist in the body. Other candidates take aim at different abnormalities identified in Long Covid research: Intravenous immunoglobulin

C L I N I CA L R E S E A R C H

A N T H R O P O L O GY | Many Indigenous people have been hesitant to engage with researchers after centuries of colonial war and decades of exploitation by some scientists. But glimmers of reconciliation are becoming more common, as more Indigenous-led research projects and genuine collaborations aim to incorporate centuries of Indigenous knowledge of the natural world into Western science. This year, new partnerships may build on high-profile examples from 2023, such as a study of the ancestry of the horse in the North America. The U.S. National Science Foundation is supporting expanded efforts with $30 million over 5 years for a new Center for Braiding Indigenous Knowledges and Science, and the National Institutes of Health is providing $9 million to Stanford University and the Native BioData Consortium, a biobank run by Native geneticists, to create a COVID-19 database for Native people. Although many points of tension remain, the new

5 JANUARY 2024 • VOL 383 ISSUE 6678

9

,

SCIENCE sciencemag.org

First fruits of Long Covid trials

Teaming with Indigenous experts

y

I N F E CT I O U S D I S E A S E S | A strategy of releasing lab-modified mosquitoes to prevent the spread of dengue fever is poised to scale up this year after a series of successes. The Aedes aegypti mosquitoes, developed and tested by the nonprofit World Mosquito Program (WMP), carry the bacterium Wolbachia pipientis, which blocks them from transmitting certain viruses and spreads to their offspring when they mate with

wild insects. A trial in Indonesia showed the strategy reduces dengue cases and hospitalizations. And in the fall of 2023, researchers reported that dengue cases fell by at least 95% in treated areas in the largest continuous release of the modified mosquitoes yet, in Colombia’s Aburrá Valley. WMP, which has projects in 14 countries, is now expanding. It plans to construct the world’s largest Wolbachia mosquito production facility, in Brazil, which will begin operations this year. WMP says it also expects the World Health Organization to issue formal guidance this year on deploying the mosquitoes—a step that could lead more countries to adopt the method.

y g

Antidengue mosquitoes take off

A mosquito modified to fight dengue lays eggs at a World Mosquito Program factory in Colombia.

y

| Governments around the world last year announced ambitious plans to increase their oversight of artificial intelligence—a race to regulate AI that will likely speed up this year. U.S. agencies face a daunting job to flesh out policy directives announced in November 2023 by President Joe Biden’s administration aimed at setting standards for the responsible development of AI. Members of Congress have also been active, introducing more than 150 bills designed to ensure that AI is a boon, not a threat, to economic development, public health, civil liberties, and national defense. The European Union appears to be closer to adopting ethical safeguards. But specific rules may quickly become obsolete because of the seemingly endless stream of powerful new AI applications, such as chatbots based on large language models.

POLICY

| Two experiments working together could reveal how the tiny masses of elusive particles called neutrinos stack up. They come in three types—electron, muon, and tau—that morph into one another, a phenomenon that might help explain how the universe generated more matter than antimatter. First, physicists have to flesh out their theoretical model. They know that two of the neutrinos have nearly the same mass, but not whether there are two light neutrinos and one heavier one, or vice versa. Physicists working with the T2K experiment in Japan and the NOvA experiment in the United States study neutrinos by shooting them hundreds of kilometers through Earth to huge detectors. This year, they plan to release a joint analysis that could indicate which of the two alternatives is correct. Finding that the electron neutrino is light would complicate planned experiments to search for a type of nuclear decay that would prove the neutrino is its own antiparticle. PA R T I C L E P H YS I C S

g

The race to regulate AI

Bringing order to neutrino mass

p

D E M O C R AC Y | The U.S. national elections in November will cap what promises to be a bumpy political ride for scientists this year. An ongoing partisan stalemate in Congress over annual spending bills could lead to a government shutdown that would disrupt grantmaking at every science agency. Most observers expect that the final budget agreement will contain at most small increases for individual agencies rather than the large boosts President Joe Biden requested and Congress had promised to deliver. The winner of the race for president will shape policy on myriad issues involving science, including climate change, pandemic preparedness, innovation, and immigration—topics on which the likely major-party candidates, Democrat Biden and his predecessor, Republican Donald Trump, have starkly different views.

NE WS | I N B R I E F

year will likely bring new models of collaboration, and more respect from some Western scientists, in these and other projects—for example in Native American genetics and the effects of climate change on Arctic ecosystems.

Pushback on EU green ambitions POLICY

| With far-right, nationalist parties

Fusion megaproject faces delay

Costly Europa probe to launch P L A N E TA RY S C I E N C E | NASA’s $5 billion Europa Clipper probe is set to launch in October aboard a SpaceX Falcon Heavy rocket as the agency’s most expensive planetary science mission since the Viking Mars probes of the 1970s. Europa, one of Jupiter’s large moons, has an icy crust only kilometers thick, beneath which swirls a vast salty ocean—a potential nursery for life. When Clipper arrives in 2030, it won’t land or sample this ocean directly, but will swoop past the moon 50 times, scanning its surface and gathering clues about its interior. Scientists had hoped the probe would follow up on hints from telescopic surveys of an active plume erupting from Europa into space. But a recent campaign with NASA’s JWST space telescope found no evidence of it.

g

| Managers of ITER, the giant experimental fusion reactor decades in the making, are expected to announce this year a new completion date, postponing “first plasma” well beyond the longstanding target of 2025. The facility, being built in France at a cost of tens of billions of dollars to its international partners, is meant to demonstrate the viability of fusion as a carbon-free energy source, but it has been beset with problems. The COVID-19 pandemic slowed manufacturing of components, segments of the reactor vessel were wrongly shaped and didn’t fit together, cooling pipes were corroded, and the French nuclear regulator was unconvinced of its safety. ITER Director-General Pietro Barabaschi, who took control 1 year ago after the death of his predecessor, Bernard Bigot, is struggling to get E N E R GY

construction back on track. He plans to reveal the revised schedule this year but told a fusion energy conference in October 2023, “It will not be good news.”

p

polling high for pan-European elections in June, observers anticipate steppedup opposition to the bloc’s sprawling green agenda. The European Green Deal, approved in 2020, aims to make the bloc “the first climate-neutral continent” and reach zero greenhouse gas emissions by 2050, with new climate-friendly measures planned across various policy areas, including energy, conservation, and transport. Reflecting those goals, about one-third of the funding under the massive Horizon Europe program will go to research related to climate change. Regardless of whether far-right parties end up forming the third biggest political force in the European Parliament, the majority group—the conservative European People’s Party—has already cast itself as friendly to farmers and industry by pushing back against

policies aimed at reducing pesticide use, for example. Legislators could also seek to limit the funding for the implementation of green legislation that has to date largely withstood conservative pushback—such as the Nature Restoration Law, which sets targets to restore degraded ecosystems.

SCIENCEMAG.ORG/NEWS Read more news from Science online. y y g y ,

Technicians at the Jet Propulsion Laboratory examine NASA’s Europa Clipper spacecraft, scheduled for launch in October to explore an icy moon of Jupiter.

10

5 JANUARY 2024 • VOL 383 ISSUE 6678

sciencemag.org SCIENCE

N E WS

IN DEP TH

p

A company called Running Tide has released tons of woody material off Iceland’s coast.

GEOENGINEERING

g

Plant waste buried at sea to fight climate change Carbon in forestry or agriculture debris could remain locked on sea floor for centuries

y

By Saima Sidik

11

,

5 JANUARY 2024 • VOL 383 ISSUE 6678

y

SCIENCE science.org

so the two layers don’t mix much at all—one reason why very little oxygen makes it to the sea floor. Without oxygen, microbes are limited in their ability to convert the carbon in biomass back into greenhouse gases, such as methane, and even if some methane is produced, chemical reactions in the sulfate-rich waters will break it down. And because the layers don’t mix, any trace greenhouse gases that are produced will be locked in the depths for hundreds or thousands of years. “There’s all these additional processes that add more layers of security,” Angel says. The advantages are enough to lure investors hoping to sell credits for the carbon removed from the atmosphere. Carbon credit marketplace Supercritical recently became Rewind’s first customer, and this summer the company plans to start sinking biomass in burlap sacks—possibly including forestry residue, river driftwood, and agricultural waste. Bulgaria, Romania, Turkey, and Georgia have all shown interest in the project, Angel says. Frontier Climate—a group that makes commitments to buy future credits from carbon sequestration startups—recently awarded $250,000 R&D grants to Rewind and another firm, Houston-based Carboniferous, which hopes to sink sugarcane waste in an oxygen-starved region of the Gulf of Mexico known as Orca Basin. The waste is abundant on Gulf Coast farms, says Morgan

y g

PHOTO: JENNIFER JOHNSON/RUNNING TIDE

D

ror Angel, a marine ecologist at the University of Haifa, had for years heard his archaeologist colleagues talk about ancient shipwrecks on the bottom of the Black Sea that were perfectly preserved by the low-oxygen environment. “You can see ropes,” Angel says. “It’s something which is quite spectacular.” Now, Angel wants to combat climate change by purposefully adding to the wreckage, sinking waste wood to the sea floor, where carbon that the trees stored up while living can remain locked away for centuries. Angel is a science lead for an Israeli company called Rewind, one of many companies riding a wave of investment in technologies that could help limit global warming by drawing carbon out of the atmosphere and locking it up. Whereas some carbon capture schemes require expensive machines and complex chemistry, burying terrestrial biomass at sea is exceedingly simple: All it takes are tugboats, barges, and woody waste from forestry and agriculture. The approach has advantages over another popular ocean-based carbon capture strategy: growing, and sinking, massive amounts of seaweed or phytoplankton. Because the plant material is grown on land rather than in the ocean, it is less likely to rob nutrients from the surrounding water and upset the

ecology. Industrial agriculture and forestry have an extensive infrastructure for growing, processing, and transporting plants, in contrast to marine farming, which has never been attempted at scale. And because woody plants are tough and unlikely to degrade, they are good at hanging on to their carbon. “Decomposers don’t like to eat them—they don’t get much out of it,” says Ning Zeng, a climate scientist at the University of Maryland. At the same time, the approach may fall short of what’s needed to fight climate change. To keep warming below 2°C, the world needs to capture and store about 10 billion tons of carbon dioxide per year by midcentury, according to the International Energy Agency. But terrestrial biomass can be sunk only where supplies of waste are located near suitable bodies of water. By one recent estimate, the approach could sequester a few tens of billions of tons of carbon dioxide in total—just a fraction of the need. “The terrestrial biomass thing is not going to solve the full problem,” says ocean engineer Kate Moran from Ocean Networks Canada, a group that is assessing the efficacy of carbon capture strategies. “It’s going to be a small piece of the pie if it is deemed to be more beneficial than risky.” But, she adds, “We need all the tools in the toolbox.” In the Black Sea, Rewind has one of the world’s great carbon burial sites. The sea is much saltier at the bottom than at the top,

NE WS | I N D E P T H

But researchers fear a plan to increase outside oversight of six top schools could create new problems By Dennis Normile

T

,

science.org SCIENCE

y

wenty years ago, five universities in Japan were among the world’s top 100, according to an annual compilation by the ShanghaiRanking Consultancy. The University of Tokyo led the contingent in 19th place, with Kyoto University 30th. But by 2023, Tokyo had tumbled to 27th and Kyoto to 39th despite repeated efforts to keep the universities globally competitive. The other three schools dropped out of the top 100 entirely. The decline has raised alarms, and Japan’s government has launched a new effort to reverse it. Last month, lawmakers approved legislation that requires six top-ranked universities to establish new management policy councils designed to give outside experts a greater voice in decisions. The councils, which will include the university president and at least three members independent of the school, will weigh in on “significant operational policies and oversee the execution of the president’s duties,” the education ministry’s higher education bureau said in a statement. And Masahito Moriyama, Japan’s education and science minister, has asserted

that the measure will “enhance education and research.” But critics fear the councils will only add to already cumbersome university bureaucracies and push academic scientists to emphasize applied research. And because Moriyama’s ministry will vet council members, they worry the new law will erode university autonomy and open the door to political meddling. The councils will lead to “increasingly complex and mysterious decisionmaking,” predicted sociologist Ryosuke Nishida of the Tokyo Institute of Technology in a critique published by The Asahi Shimbun, a prominent Japanese news outlet. All sides agree that Japan’s research establishment has endured what some academics call “the lost decades.” In addition to the slide in university rankings, since the early 2000s Japan has dropped from second to fifth in the number of papers indexed in Clarivate’s Web of Science, according to Japan’s National Institute of Science and Technology Policy (NISTEP). And it fell from fifth to 12th in the number of papers ranking in the top 1% of citations, trailing countries including Spain and Iran. Multiple factors have contributed to the decline, says Masatsura Igami, a science policy researcher at NISTEP. But he singles out a

y g

5 JANUARY 2024 • VOL 383 ISSUE 6678

Japan tries, again, to boost global ranking of its universities

y

12

SCIENCE POLICY

g

Saima Sidik is science journalist based in Somerville, Massachusetts.

p

Raven, a biogeochemist at the University of California, Santa Barbara and the company’s chief science officer. “It’s already sitting in piles,” she says. “The alternative for this material is essentially that it degrades, releases methane, and requires tending so it doesn’t light on fire.” Carboniferous is now applying for permission to test its strategy from the Environmental Protection Agency. Portland, Maine–based Running Tide is combining terrestrial and marine biomass in one carbon capture strategy. The company takes waste wood from a forestry operation in Nova Scotia that would otherwise be burned or left to decay and presses it together to create floating “buoys” no bigger than a basketball that are seeded with seaweed spores. The buoys are released off the coast of Iceland, where ocean currents carry them over a deep region with little oxygen. Eventually they become waterlogged and sink, along with any seaweed that has grown en route. Last summer, Running Tide sold its first carbon credits to Shopify, and the company says it has sunk tens of thousands of tons of material into the North Atlantic Ocean. Marine scientist David Koweek of the nonprofit Ocean Visions, which has previously supported Running Tide’s research, lauds the simplicity of sinking terrestrial biomass, because technology exists for almost every step in the process. That’s a strong reason why “you might think about doing this,” he says. Beyond that, the benefits are murkier. Even though boats are a climate-friendly form of transportation (trucks emit at least 100 times more carbon per kilometer), Angel says it wouldn’t make sense to ship biomass around the world to get it to favorable sites. And although sunk terrestrial biomass doesn’t steal nutrients from marine life, removing it from land could deplete soil of nutrients. “Over time we’re going to also be losing some of the fertility that crops and forests need,” says Charlotte Levy, a biogeochemist at Carbon180, which advocates for scaling up carbon removal projects. Levy also worries that as innovators find new uses for scrap biomass—for example, as sustainable building materials or biochar, a charcoal-like soil additive—sinking the biomass might not be the most environmentally friendly use. Zeng agrees that sinking terrestrial biomass will be limited to a few areas of the ocean for the foreseeable future. But the urgency of carbon removal demands that every possible scheme be explored thoroughly, he says. “I think every idea deserves $1 billion of support to test it out.” j

Tohoku University, which hosts this advanced synchrotron, is one of several Japanese schools to fall off a list of the world’s top 100 universities.

Megaproject will chart human immune diversity Human Immunome Project’s survey aims to improve drugs and vaccines By Mitch Leslie

T

y ,

13

y g

5 JANUARY 2024 • VOL 383 ISSUE 6678

y

he hepatitis B vaccine is one of the most potent immunizations, usually providing decades of protection against the deadly liver virus. But in about 10% of people it doesn’t work, and in 2020, Amy Huei-Yi Lee, a systems biologist at Simon Fraser University, and her colleagues set out to determine whether they could predict who would benefit. The scientists found that data on recipients’ immune systems such as the abundance of certain proteins and the activity patterns of a few genes foretold whether they would generate defenses against the virus. “We got a sense of what factors drive the vaccine response and what [doesn’t],” Lee says. She and her colleagues were only able to take measurements from a handful of patients, but an ambitious effort slated to begin early this year will collect such data from hundreds of thousands of volunteers throughout the world. Called the Human Immunome Project (HIP) and backed by an international consortium of companies, government agencies, and universities, the effort will probe thousands of immune variables in blood and tissue samples. The result will likely be the world’s largest and most comprehensive immunological database, a resource for scientists investigating immune system differences and how they influence our responses to vaccines and drugs and our vulnerability to illness. “There’s a huge opportunity here in terms of understanding human disease,” says immunologist Mark Davis of Stanford University, who is not involved in the project. And that’s just the start for the effort, which currently operates on about $5 million a year in funding but could ultimately cost billions. An offshoot of a previous effort known as the Human Vaccines Project, HIP

g

SCIENCE science.org

IMMUNOLOGY

p

“stagnation” in research spending and a scientific workforce that has struggled to keep pace with those found in other nations. Many researchers trace Japan’s current academic malaise to a 2003 law that turned the nation’s 86 government-supported national universities into corporate entities with greater independence. At the time, policymakers argued that private entities outperformed public institutions, and that corporatization was a means of “improving efficiency and strengthening universities through competition,” says Kiyoshi Yamamoto, an emeritus professor at the University of Tokyo who specialized in university management. While pursuing that strategy, however, the government steadily cut operating funds for the universities, which included some support for research. Overall, those funds dropped from $87 billion in 2004 to $76 billion in 2022, according to the Japan Association of National Universities. To compensate, universities were allowed to raise tuition, increase revenues from affiliated hospitals, and pursue contract research with industry. The government also launched a number of competitive grant programs intended to promote promising areas of research and strengthen top universities. In 2015, for example, it created the Japan Agency for Medical Research and Development (AMED) to move basic biomedical research discoveries into clinical trials. And just 2 years ago it also launched an initiative that will award up to $70 million annually to each university deemed capable of becoming a world-class institution (Science, 27 May 2022, p. 903). In September 2023, Tohoku University won the first such award. Although the competitive programs have helped some universities, they are “still far

from being able to make up for … the reduction in government funding,” says Futao Huang, a higher education scholar at Hiroshima University. The cost cutting, meanwhile, led to both a dearth of entry-level research jobs and “the creation of a career ladder of low-paid, insecure jobs” that “drove talented young researchers out of academia,” says Robert Geller, a geophysicist and emeritus professor at the University of Tokyo. Enrollment in Ph.D. programs has dropped precipitously. Such developments are now undermining government efforts to boost innovation, some scientists say. To cope with budget cuts, for example, Japan’s national universities increased the clinical duties of their hospital physicians. But that meant they had less time for biomedical research, according to an analysis reported by physician-scientists Shotaro Kinoshita and Taishiro Kishimoto of the Keio University School of Medicine in an October 2023 letter in The Lancet. As a result, “research output has declined,” Kinoshita says—a trend that “may be hindering” AMED’s mission. Critics of the latest tweak to university governance don’t believe it addresses the challenges universities face. Outsiders, they note, already have a voice in managing most universities. At the University of Tokyo, for example, independent advisers serve on the board and hold a majority of the seats on a 28-member Administrative Council. Others worry about a provision in the new law that allows the education minister to veto council appointees, saying it could politicize university governance. As an example, they point to a recent controversy over appointees to the Science Council of Japan (SCJ), which plays a critical role in advising the government and is considered the nation’s foremost association of researchers. Breaking with historic practice, in October 2020 then–Prime Minister Yoshihide Suga blocked the SCJ appointments of six academics who had criticized the policies of his Liberal Democratic Party (LDP), creating an uproar among researchers and the public. Such concerns did not prevent the current LDP government from pushing the new measure through the legislature in late December 2023. Opponents take some solace, however, from resolutions added to the bill that call for the protection of academic freedom and respect for a university president’s council nominations. The resolutions are not legally binding. But Sayaka Oki, a historian of science at the University of Tokyo, says they could give research advocacy groups that “had not seriously tried to cooperate before” a “means to battle against possible harmful outcomes of the new law.” j

NE WS | I N D E P T H

y ,

science.org SCIENCE

y g

5 JANUARY 2024 • VOL 383 ISSUE 6678

ect will then select a few variables that provide the clearest picture of how the immune system is working. They will also provide the basis for an immune monitoring kit, a standard set of assays that all the sites in the second part of the project will use. In the end, HIP will generate nearly 2 trillion immune measurements, which will be publicly available through a central database. With this data haul and other information, HIP will build a predictive AI model that can forecast—based on immune profile, ancestry, economic status, age, and other information—how individuals will respond to stresses or challenges, such as a particular drug or pathogen. The model could help pharmaceutical companies identify opportunities for new treatments and drug reactions to avoid. And by providing a much more detailed view of a population’s health and vulnerability to side effects, the model could enable countries to better decide which drugs are needed by and suitable for their populations, thus allowing them to reduce health care costs, Keirstead says. What HIP is aiming for with its AI ambitions has “never been done before,” Kierstead says, which is probably why this part of the project draws more skepticism from outside researchers. The project intends to generate not just predictive models, but also ones that replicate how the immune system operates. Mathematical biologist Reinhard Laubenbacher of the University of Florida says the AI will detect patterns of responses but doubts it will open a deeper understanding of the immune system. “Data collection efforts like this are tremendously helpful, but we will probably need more than that,” he says. A priority is “building [a] theoretical framework” to understand the information the project will accrue, he says. Another challenge is money. To realize its ambitions, HIP will require a supersize budget, about $1 billion to $3 billion over the next 10 years, Keirstead says. To raise the needed funds, HIP now hopes to go beyond its current partners to philanthropies, governments, and other pharmaceutical companies. “I am targeting everyone. There is not going to be a stone left unturned,” he says. Ensuring that HIP’s far-flung sites follow the same procedures in collecting and analyzing the data will also be a challenge, Lee says, adding that the immune monitoring kits will be a big help in this regard. Thomas says that attracting nonwhite participants could also be difficult, given their mistrust of scientific research like this. “They haven’t seen benefits and have been exploited.” Still, he and others are eager to see what HIP produces. “If they pull this off, it will be big,” Greenplate says. j

y

projects have collected limited categories of information and, in Project Baseline’s case, haven’t made the data available publicly. Another area where research has fallen short is “the understanding of human immune variation and diversity,” says John Tsang, a systems immunologist at Yale University who helped develop HIP’s scientific plan. A litany of factors—including age, sex, diet, living conditions, previous disease exposure, and genetics—shapes how the immune system functions. But Eosinophil most immunological studies are Attackss p parasites and p pathogens g conducted on small, homogenous populations, usually in the United States or Europe, Tsang says. Relying on such a narrow slice of humanity “has biased our understanding,” Thomas says. HIP aims to address that lack of diversity. “We want baseline data from every human population,” Keirstead says. To capture T cell human variety, HIP’s plans call Kills infected cells for up to 300 collection sites on all and orchestrates immune response spo of the inhabited continents. Each site will measure the same set of variables in as many as 10,000 v people, from different socioeconomic levels and a range of ages, from newborns to centenarians. In addition, they will include healthy people as well as individuals who have medical problems such as autoimmune diseases, cancer, and allergies. Macrophage All A volunteers will have to unConsumes bacteria and dergo medical exams and prodestroys damaged cells vide a detailed health history. v Although HIP intends to begin this global data collection phase in 2027, the effort’s first phase, launching this year, will be smaller and likely involve seven to 10 clinical research centers, including facilities outside the w wealthy countries, that are already adept at gathering and anB cell Produces antibodies alyzing immune data, Keirstead says. At each site, the project will study about 500 people, w measuring immune variables including the abundance of different types of immune cells, gene activity, concentrations of metabolic molecules, and DNA sequences. “The idea is that we will w go deep and measure as much as possible,” Tsang says. Neutrophil From this mass of data, the projCaptures and kills pathogens

g

14

Defenders

Immune cells have a variety of roles, and cataloging them may help researchers take a snapshot of immune system health.

p

will also use the data as fodder for new artificial intelligence (AI) models that could predict immune system responses across entire populations, providing valuable insights not just for pharmaceutical companies and governments, but even for doctors and patients. “The impacts will be felt globally,” says neuroscientist Hans Keirstead, he Irvine, California–based project’s CEO. Scientists unconnected to the project say its goal off compiling a basic immune database for the world is feasible. “We have the experience and technology,” says immunologist Allison Greenplate of the University of Pennsylvania. But she and others question how much insight AI will add. “There is a lot off low-hanging fruit we don’t need AI to pick” but that researchers can parse themselves, says immunologist Paul Thomas of St. Jude Children’s Research Hospital. In the field of cardiology, a lipid panel reveals a lot about a patient’s cardiovascular health and risk of disease. Immunology, however, doesn’t have a comparable set of simple measurements that indicate the status of a person’s immune system, Davis says. Some data can provide a rough gauge: Patients with reduced numbers of neutrophils, for instance, are prone to infections. But such dataa are limited. HIP aims to come up with a uniform group off measurements that can, like a lipid panel, provide a readout of the immune system’s functioning. A few public and private efforts have scooped up some basic immune data from large numbers of people, including All of Us, the U.S. National Institutes of Health’s program to gather genomic and medical data from 1 million people, and Project Baseline from the Google offshoot Verily, which tallied information on how individuals responded to COVID-19 infection. But such

ASTRONOMY

Magellanic cloud may be two galaxies, not one Rethink of familiar object may boost odds that its name, offensive to some, will be changed By Daniel Clery

y ,

PHOTO: JUAN CARLOS MUNOZ-MATEOS/ESO

y g

15

y

5 JANUARY 2024 • VOL 383 ISSUE 6678

g

SCIENCE science.org

p

its parts, de los Reyes says, and relies on “better data.” The question remains whether the SMC’s two components are separate obhe Small Magellanic Cloud (SMC), a jects that have drifted close together and are hazy blob easily visible in the Southbound by their mutual gravity or whether ern Hemisphere’s night sky, has long one is debris drawn from the other by a close been considered a lone dwarf galaxy encounter with another galaxy such as the close to the Milky Way. But a study LMC. That the two parts seem to contain simposted online last month, and acilar masses of gas is “suggestive of two syscepted by The Astrophysical Journal, sugtems” without a connection, Murray says. But gests the familiar site is not a single body, de los Reyes counters that the level to which but two, with one behind the other as previous generations of stars have enriched viewed from Earth. the two gas clouds with heavier elements is By tracking the movements of clouds “relatively similar,” suggesting they are “not of gas within the SMC and the young stars entirely unrelated.” recently formed within them, astronomer If further studies confirm that the SMC Claire Murray of the Space Telescope Sciis not one thing, but two, a new ence Institute and her colleagues name or names will be in order. have found evidence of two stellar De los Reyes argues that whatever nurseries thousands of light-years the clouds are called, astronomers apart. If confirmed, the reassessneed to oust Magellan. She points ment will likely amplify calls from out that the objects had been an increasing number of astronoknown and described by peoples mers to change the SMC’s name in the Southern Hemisphere for and that of its neighbor, the Large thousands of years, as well as reMagellanic Cloud (LMC). corded by Italian and Arabic exSixteenth century Portuguese plorers before Magellan’s voyage. explorer Ferdinand Magellan, afThe astronomical community ter whom the galaxies are named, has long celebrated Magellan. was not an astronomer, did not There are Moon and Mars craters discover them, and is recorded as bearing his name, NASA’s Magelhaving murdered and enslaved lan mission to Venus, and the twin Indigenous people during his Magellan Telescopes in Chile and first-ever circumnavigation of the the Giant Magellan Telescope curglobe. As a result, astronomer Mia Magellanic clouds, seen here above Paranal Observatory in Chile, rently under construction there. de los Reyes of Amherst College are due for a name change, some astronomers say. De los Reyes says the atrocities called for renaming the clouds in an opinion piece for Physics magathat Magellan and his crews comzine last year. The idea has since “gotten a lot Combining those two sets of data and asmitted in Argentina and elsewhere during of informal support,” she says. suming any stars that young would still be their travels make him unsuitable for these The two Magellanic clouds are in the gravimoving in concert with the clouds of gas honors and for having his name in the more tational sway of the much larger Milky Way from which they formed, they identified two than 17,000 peer-reviewed papers as a result. and are on course to pass through it in the distinct star-forming clouds. By measuring De los Reyes is now working with the future. The LMC is a disk galaxy, like the how much light from the two clouds is abInternational Astronomical Union (IAU), Milky Way in miniature. But the SMC is irsorbed by dust between them and Earth, they which manages the naming of celestial obregular in shape, two-thirds of the LMC’s calculated that one is more distant than the jects, to present a resolution to its general mass, and often gets overlooked in favor of other. But determining the amount of sepaassembly in August. But what to call the its heftier neighbor. “The SMC got the short ration proved tricky. “Measuring distance to familiar clouds? Murray suggests one of end of the stick,” de los Reyes says. “It’s less gas is very challenging,” Murray says. In the their many Indigenous names. De los Reyes well modeled.” new study, her team describes constructing a says the IAU resolution will include sevSome previous studies have suggested simple computer model of the system. With eral options for members to vote on. She that the SMC may have multiple compothe available data plugged in, it predicts the likes the idea of keeping the acronyms SMC nents, but the truth has not been easy to pin two clouds are 16,000 light-years apart— and LMC—“to maintain continuity in the down. The SMC “has suffered most” in its more than half the distance between Earth literature”—and just changing the “M” to gravitational interactions with the LMC and and the center of the Milky Way. something “more physically meaningful” the Milky Way, Murray says. “It’s simply full This view of the SMC “is much more comsuch as “milky” or “meridional,” a reference of disrupted gas. It’s a train wreck of sorts.” pelling” than previous attempts to untangle to the cloud’s southern location. j

T

So, she and her team set out to reinvestigate both the system’s stars and its gas, using what Murray calls the “latest and greatest” instruments available. To probe the SMC’s gas they turned to the Australian Square Kilometre Array Pathfinder, a radio telescope made up of 36 dishes each 12 meters in diameter. With it, they zoomed in on radio waves emitted by atomic hydrogen gas. They then plotted the location and velocity of thousands of stars in the SMC that are less than 10 million years old using Gaia, an orbiting observatory from the European Space Agency. Over the past decade, Gaia has logged such data for more than 1 billion stars across the Milky Way with unprecedented accuracy.

NE WS | I N D E P T H

Monica Bertagnolli in her office at the National Institutes of Health.

By Jocelyn Kaiser

science.org SCIENCE

,

5 JANUARY 2024 • VOL 383 ISSUE 6678

y

16

y g

W

hen cancer surgeon Monica Bertagnolli learned last year that President Joe Biden wanted her to lead the $47.5 billion National Institutes of Health (NIH), the world’s largest funder of biomedical research, “I didn’t embrace it,” she says with a laugh. She was just a few months into heading NIH’s largest component, the National Cancer Institute (NCI). “My reaction was: ‘But I’m the NCI director and I have plans.’ We were accomplishing some really great things together. And it was just too soon.” But the idea grew on her. She had gotten to know the chiefs of NIH’s 26 other institutes and centers at weekly meetings and found them “dynamic and talented and eager for innovation.” NIH’s acting director, Lawrence Tabak, was lamenting the lack of a permanent NIH chief since geneticist Francis Collins stepped down in December 2021 after 12 years at the helm. And Bertagnolli saw the post as a chance to advance “some things that I am passionate about that I think will help biomedical research overall.” So, she took the job. A few weeks in, she sat down recently for a 45-minute interview in her office between a swirl of meetings, calls,

and the first town hall with NIH staff to fill a nearly 500-seat room since the COVID-19 pandemic. She discussed her goal of making NIH research more “equitable and accessible” by expanding clinical trials to more rural and minority patients. She also wants researchers to bring more rigor to clinical care with the help of artificial intelligence (AI). “We have an opportunity to be able to deliver so much more in terms of real high-quality, wellunderstood results that can help guide care for people,” she says. Colleagues say that although Bertagnolli, age 64, has made a name as a lab scientist and clinical trial leader, she is a physician first. Raised on a ranch in Wyoming, she studied engineering at Princeton University before heading to the University of Utah for medical school and a career as a physician-scientist at Brigham and Women’s Hospital and the Dana Farber Cancer Institute, where she was chief of surgical oncology for 10 years. Dana Farber colleague Suzanne George says Bertagnolli was known to “take the hardest cases” of sarcomas, a rare cancer of bone and soft tissue, offering hope to desperate patients. At the same time, she was making her name as a researcher. Until about 7 years ago, Bertagnolli ran an NIH-funded tumor immunology lab studying how inflammation

y

Improving patient care is top priority for head of world’s largest biomedical research funder

g

Once reluctant, new NIH chief Monica Bertagnolli embraces role

p

BIOMEDICAL RESEARCH

drives colon cancer in mice. The work led to a landmark clinical trial testing whether anti-inflammatory drugs called Cox-2 inhibitors can prevent precancerous colon polyps in people prone to these growths. Bertagnolli still remembers walking down a hallway at Brigham when the trial’s results came out in 2006. Cox-2 inhibitors reduced polyps but also raised the risk of strokes and heart attacks, putting the drugs off limits as a preventive measure. “My oncology colleagues said, ‘We’re so sorry,’” she says. “But cardiologists said, ‘That was great!’”—because the large study firmed up the risks of Cox-2 drugs, which had already led regulators to remove some types from the market. Bertagnolli’s trial experience led to a job heading one of NCI’s large cooperative groups, which run multiple cancer trials. When a 2010 Institute of Medicine report urged NCI to overhaul these groups, she agreed to oversee a merger of three of them— one run by researchers at powerhouse cancer centers since the 1950s, another centered in the Midwest with many community oncologists, and a third consisting of surgeons. Bertagnolli gave these disparate groups of nearly 10,000 cancer specialists what she calls a “completely new identity” as the Alliance for Clinical Trials in Oncology. On its committees, community oncologists from the rural West now sit side by side with researchers from major academic centers. Merging these groups was “extraordinarily multifaceted and complex,” says Elad Sharon, a former staffer in NCI’s clinical trials division who is now at Dana Farber. “She was clearly head and shoulders the most innovative thinker running those shows.” In October 2022, Bertagnolli became the first woman to lead NCI. Two months later, she announced that she had early breast cancer. (She has completed treatment and has a good prognosis.) Within 6 months, she had produced a document called the National Cancer Plan that lays out eight strategies, such as smoking cessation campaigns and wider cancer screening, to achieve Biden’s moonshot goal of cutting the U.S. cancer death rate in half by 2047. She unveiled a new unit at NCI aimed at working with the U.S. Food and Drug Administration to develop innovative clinical trial designs. In May 2023, she was nominated to become the second woman to lead NIH. It took until October for Senator Bernie Sanders (I–VT), who heads the committee that oversaw her nomination, to schedule her confirmation hearing. She was confirmed by a Senate vote of 62 to 36 after a hearing where conservatives pressed her on NIH’s

Peru moves to crack down on fraud in research publishing Two new laws would make it easier to punish scientists who buy authorship and commit other misdeeds By María de los Ángeles Orfila

P

,

17

y

5 JANUARY 2024 • VOL 383 ISSUE 6678

y g

María de los Ángeles Orfila is a journalist in Montevideo, Uruguay.

y

eru’s legislature is poised to approve new laws that would make it easier to investigate and punish researchers who engage in fraudulent publishing practices, including paying to have their names added to a paper. The move comes as Peru’s national science agency seeks to crack down on authorship buying and other unethical practices. It recently removed two scientists accused of violations from a national registry that is key to receiving government grants, job promotions, and salary bonuses. And officials are investigating more than 170 other researchers that a Peruvian media report alleged were involved in publishing fraud, including 72 listed on the national registry who work at 14 universities in Peru. The new legislation will empower universities and government officials to punish such behavior, and would give Peru some of the strongest measures against publishing fraud in Latin America. Shady publishing practices “transcend mere ethical violations” because they enable researchers to obtain government and private funding under false pretenses, says Edward Málaga Trillo, a neurobiologist and member of Congress who is the driving force behind the bills, which lawmakers are expected to finalize early this year. “These individuals are perpetrating fraud.” As in many other nations, Peru’s academic community has been struggling with a rising tide of fake authorship and related problems. One cause, some researchers say, is a 2014 law that aimed to stimulate research by rewarding researchers who boost their publishing output. For example, under a scoring system used by universities, researchers can earn five points for authorship in a high-impact journal, and two points when the journal is lower impact. Amassing points can bring bonus payments and career advancement. The law creates perverse incentives, researchers say. In an October 2023 report, journalists with the TV program Punto Final reported that Peruvian researchers were paying up to $500 to add their names to papers that they did not help write. The reward can be substantial: Some for-profit universities in

Peru pay publishing bonuses of $2500, says biologist Gisella Orjeda Fernandez of the National University of San Marcos, a former head of Peru’s science agency, the National Council of Science, Technology and Technological Innovation (Concytec). “There is a whole environment that incubates this [misconduct],” Orjeda Fernandez says. The 72 researchers that Punto Final alleged were involved in publishing fraud comprise an “alarming” 1% of all scientists on Peru’s national registry, says Concytec President Benjamín Marticorena Castillo. One of the two researchers that Concytec has already removed from the registry was found to have republished articles by a Spanish researcher under their own name. The other expelled researcher had assumed the identity of a Venezuelan scientist. Signs of publishing fraud can be obvious, notes Nahuel Monteblanco, president of Cientificos.pe, a Peruvian group othat investigates misconduct. Many of the papers cited by Punto Final have numerous co-authors who are based in many different nations and have few prior publications on the same subject. “If your colleague consistently publishes 20 articles a year with collaborators from Nepal, Afghanistan, Kuwait, or Indonesia, that’s highly suspect,” Monteblanco says. Existing law, however, gives Concytec and Peru’s 93 public and private universities limited authority to investigate and punish such misconduct. The two bills now before Congress aim to change that. One would amend a law covering universities, whereas the other gives new powers to Concytec. The legislation defines fraud to include “plagiarism, fabrication or falsification of information in publications, projects, reports and any other academic product related to scientific research.” Researchers who commit lesser violations could be suspended from the national registry for 2 to 5 years. Those committing more serious infractions could be permanently suspended and potentially face criminal charges and imprisonment. Peruvian science, Orjeda Fernandez says, “should not hesitate to expel the guilty.” j

g

SCIENCE science.org

SCIENTIFIC MISCONDUCT

p

support of gender-affirming research, fetal tissue studies, and NIH’s role in funding virus research in China that some claim led to the pandemic. Sanders had delayed her hearing because he wants NIH to do more about drug prices. He and others have called on NIH to use its “march-in rights,” exercising a provision in a 1980 law that some experts say allows the agency to reassign a patent on a drug developed with taxpayer funding if a company is charging an unreasonable price. The Biden administration has since proposed that NIH can use that mechanism to control prices— a shift from previous administration policy. The proposal is open for public comment. “People should have fair and equitable access” to NIH-funded drugs, but “the situation is complex,” Bertagnolli says. She is now working to rebuild relationships with members of Congress that frayed during the pandemic, she says. At NIH Bertagnolli plans to push for a goal she’s championed for several years: building a “learning health system.” The idea involves folding patients’ electronic health records into a research database and then using AI tools to extract useful information. The approach could, for example, help diagnose a condition or determine whether a cancer drug tested in a clinical study is working in the real world. Bertagnolli wants NIH’s basic research community to know that her interest in clinical research does not “diminish in any way the importance of fundamental science.” But, she adds, “The research arm needs to be able to work with and embrace our entire health system.” She expects to work across health agencies to beef up standards for electronic health records and promote their adoption so patients can share their data for research. Bertagnolli also faces immediate pressures. A report delivered by her advisers last month, for example, urges NIH-funded institutions to raise the minimum salaries for postdocs by 24% to $70,000, despite an agency budget that could stay level over the next 2 years. “We can make a commitment to do all we can with the extramural research community to make [postdoc raises] possible,” she says. “But sacrifices are going to need to be made. And I don’t think we can entirely foresee how this is going to work out. It’s just too complicated and it’s so intertwined with the dynamics of the academic community and environment.” Colleagues think she is up to the challenges. Cell biologist and Huntsman Cancer Institute CEO Mary Beckerle, who has known Bertagnolli since her Utah medical school days, compares her to the late Supreme Court Justice Sandra Day O’Connor, who also grew up on a ranch: A “bundle of wit, resilience, and can-do spirit,” Beckerle says. j

NE WS

p g

FEATURES y y g

LAID TO WASTE Ukrainian scientists are tallying the grave environmental consequences of the Kakhovka Dam disaster

y ,

By Richard Stone, in Kyiv and Odesa, Ukraine

I

n the predawn hours of 6 June 2023, a pair of explosions rocked the Kakhovka Dam, a 3-kilometer-long hydropower facility on the Dnipro River in southern Ukraine. Waking up that morning to the unfolding catastrophe, “I couldn’t believe it,” recalls Volodymyr Osadchyi, director of the Ukrainian Hydrometeorological Institute (UHMI). “I thought it had to be fake news.” But footage captured by a Ukrainian military drone showed water from one of Europe’s largest reservoirs gushing through a 18

5 JANUARY 2024 • VOL 383 ISSUE 6678

gaping breach in the dam. Over the next 4 days, 18 cubic kilometers of water surged downstream, inundating more than 620 square kilometers and affecting 80 settlements. Scores of people died, and many more are unaccounted for. Up to 1 million people lost access to drinking water. In October, the Ukrainian government pegged the cost of the disaster, which it blames on Russia, at roughly $14 billion. Nearly half that figure— $6.4 billion—is an estimate of lost ecosystem services due to chemical pollution and

habitat destruction along the Dnipro, one of Europe’s largest rivers. Assessing environmental harm in the midst of a war in which the Dnipro itself delineates more than 300 kilometers of the front line is not easy. But Osadchyi and other Ukrainian researchers have been sobered by what they’ve found so far. The toll includes heavy damage to a unique sturgeon breeding facility, flooding of nature reserves and agricultural land, and a death blow to countless organisms adapted to brackish estuaries near the confluence of science.org SCIENCE

N E WS

p g

PHOTO: ALEXEI KONOVALOV/TASS VIA ZUMA PRESS

long Dnipro (the Dnieper, for Russians)

5 JANUARY 2024 • VOL 383 ISSUE 6678

19

,

ORIGINATING IN RUSSIA, the 2200-kilometer-

flows across Ukraine’s northern frontier with Belarus, near Chornobyl, wending through Kyiv and the rest of the country before spilling into the Dnipro-Bug Estuary and the Black Sea. After World War II, Soviet engineers built or refurbished six hydroelectric dams along the river. They completed the Kakhovka Dam, the final and largest one, in 1956. Soon after Russia launched its fullscale invasion of Ukraine on 24 February 2022, it captured the Kakhovka Dam and Nova Kakhovka, a city on the Dnipro’s left bank built for the hydroelectric station’s workers. In a TV address that October, Ukrainian President Volodymyr Zelenskyy claimed the Russian military had rigged the dam with explosives and warned that its destruction “would mean a large-scale disaster.” At the time, Ukrainian troops were on the verge of recapturing Kherson, a major city downstream on the Dnipro, triggering concerns Russia might blow up the dam to unleash floodwaters that would slow the Ukrainian advance. It would not have been the first use of water as a weapon during the conflict. In March 2022, the Ukrainian army destroyed a dam on the Irpin River, north of Kyiv, to

y

SCIENCE science.org

to thrive when humans are displaced. For instance, because it has been too dangerous to fish in much of the Black Sea, fish stocks are thought to be rebounding. With hunting banned in a wintering ground for migratory birds on the southern steppe, near the Black Sea coast, “the situation for many populations of birds is so much better now,” says Vasiliy Kostiushyn, an ornithologist at the I.I. Schmalhausen Institute of Zoology (IZAN), also part of NAS. The ecological legacy of the Kakhovka disaster is similarly nuanced, as new habitats emerge on terrain scoured by floodwater and in the former reservoir’s driedout lakebed. The Ukrainian government has vowed to rebuild the dam after the war. But some experts hope it will change its mind and allow a natural recovery—and perhaps even accelerate efforts to rewild parts of the lower Dnipro Basin. “Kakhovka is a tragedy,” says marine ecologist Galyna Minicheva, director of NAS’s Institute of Marine Biology (IMB). “But it is also a huge and unprecedented natural experiment.”

y g

the Dnipro and the Black Sea. Billions of mussels are rotting on the former reservoir’s desiccated lakebed. And the Kakhovka’s destruction has added a new wrinkle to a puzzle that arose earlier in the war: the unexplained deaths of dolphins and porpoises in the Black Sea off Ukraine’s southern coast. Such ecological miseries pale in comparison with the atrocities Russian forces have committed against Ukrainians during a nearly 2-year war that has claimed hundreds of thousands of lives. But when the long and grinding war finally ends, Ukrainians will confront environmental damage that extends well beyond the Dnipro to widespread chemical contamination of agricultural fields and forests from shelling, the wanton destruction of protected areas, and the laying of innumerable mines that experts say will take decades to clear. “The environmental cost of the war has been immense,” says Sergei Mosyakin, director of the M.G. Kholodny Institute of Botany, part of the National Academy of Sciences of Ukraine (NAS). “The ecocide that Russia has inflicted on our country will be studied for generations to come.” Yet as in all wars, nature has found places

y

Billions of zebra and quagga mussels are rotting in the dessicated lakebed of the Kakhovka Reservoir (left), which largely emptied after explosions tore apart the Kakhovka Dam in June 2023, unleashing a destructive flood.

NE WS | F E AT U R E S

The morning of the disaster, Sergiy Afanasyev’s first thoughts were for colleagues at the S.T. Artyushchyk Production Experimental Dnipro Sturgeon Breeding Plant. Located southwest of Kherson in Dniprovs’ke, the 40-year-old station annually stocked the Dnipro’s reservoirs with some 1.5 million juvenile sturgeons adapted to the region, including the critically endangered Danube sturgeon (Acipenser gueldenstaedtii). Ukrainian troops evacuated its staff before a 4-meter wave of turbid, polluted water flooded the station and its breeding ponds. But the sturgeons, which

bog down Russian troops marching on the capital. The next month it employed a similar tactic, punching a hole in a dam at the Oskil Reservoir in the Kharkiv region. Then in September 2022, Russia hit a dam on the Inhulets River with cruise missiles, unleashing clayey floodwaters on Kryvyi Rih, Zelenskyy’s hometown. Those and other strikes on water infrastructure ruined cropland and wrecked local ecosystems. But even though Ukraine clawed back Kherson that November, the Kakhovka Dam, still under Russian occupation, stayed intact for another half a year.

Reservoir bogs

Dnipro River

Nikopol

Ukraine

Odesa

Shrubs and grasses Marshy vegetation Water bodies Sand and silt

y

Zaporizhzhia Nuclear Power Plant

AROUND 9 A.M. on 4 October, Osadchyi and

Restoring the great meadow Former lakebed below the embankment would become steppe grassland. Proposed embankment

30

0

30

Steppe grassland site

When the Kakhovka blasts came, they dwarfed the previous attacks. The breach was wider than modeling of an explosion had anticipated. “Something else made everything way worse,” says Amin Tavakkoli Estahbanati, a remote sensing specialist at the University of Houston. Analyzing synthetic aperture radar measurements from 2015 to 2023, he and his colleagues reported at the American Geophysical Union meeting in San Francisco last month that the dam had begun to deform months before the blasts, possibly because of faulty operations and poor maintenance. 20

5 JANUARY 2024 • VOL 383 ISSUE 6678

are acutely sensitive to toxicants, could not be rescued. “There is a very low possibility that any survived,” says Afanasyev, director of NAS’s Institute of Hydrobiology. The breach would also have been devastating for the scarce wild sturgeons in the Dnipro. In late spring, the fish swim upriver from the Black Sea to spawning grounds just below the Kakhovka Dam. Early June is peak breeding season, Afanasyev says. “The explosion wiped them out.” Upstream of the dam, nearly 90% of the Kakhovka Reservoir drained, exposing 1870 square kilometers of former lakebed,

science.org SCIENCE

,

km

y

0

Depth (meters)

Grygorii Derkach wriggled under a barbed wire fence meant to keep civilians out of a death zone along the Dnipro near Kherson, less than 1 kilometer from Russian forces on the opposite bank. Ukrainian army officers urged the UHMI hydrologists to secure their precious samples of river water quickly, before the Russians began their daily shelling. “We were in a sour mood,” Osadchyi says—a lingering disquiet from the previous day, when the duo had driven through the ruined village of Chornobaivka. As the pair edged toward the Dnipro that cool, sunny morning, they heard what sounded like a motorbike engine. “Hit the ground!” a soldier shouted. Osadchyi dove under a tree as a Russian reconnaissance drone swooped in, just 20 meters overhead. “It was terrifying,” he says. The drone departed, and the rattled scientists scrambled to fill their 40-liter plastic jug. Lugging it back to their car in Kherson’s River Port District, they heard shells explode near the area they’d occupied moments earlier. Back at their laboratory, the UHMI researchers learned new details about water quality in the Dnipro. Although the nasty compounds in the reservoir sediments have largely washed out of the water column, the river is still freighted with nitrogen, phosphorus, and toxicants from sewage and

y g

Kakhovka Dam Nova Kakhovka

g

Kyiv Dnipro River

p

The Kakhovka Dam breach on 6 June 2023 inundated settlements and unique ecosystems along the lower Dnipro River. Nearly 90% of the reservoir drained, exposing 1870 square kilometers of former lakebed. One proposal for postdisaster recovery calls for building a 50-kilometer-long embankment that would create a narrower, deeper reservoir and allow for the return of steppe grassland.

including dense beds of zebra and quagga mussels that once filtered and cleansed the reservoir’s water. Now, as much as 500,000 tons of dead bivalves are rotting in the desiccated lakebed. It could take a few years for the soft body parts to decompose fully, and their shells much longer, says Volodymyr Yuryshynets, a parasitologist at the Institute of Hydrobiology. Normally, Afanasyev would have immediately dispatched scientists to assess the Dnipro’s health. But because of the “extreme danger of such investigations” along the war’s front line, he says, his institute first enlisted volunteers—fishers and others who remained along the river— to report on fish stocks and send water samples to Kyiv for analysis. Those samples enabled the scientists to verify that sediments swept downstream were laden with manganese and other heavy metals, and long-lived organic compounds such as polychlorinated biphenyls (PCBs) that had accumulated over decades in sediments behind the dam. “The colloidal particles were quite toxic,” Afanasyev says. To bring the picture into sharper focus, however, the scientists knew there was no substitute for seeing the disaster area for themselves.

NE WS | F E AT U R E S

rine perch (Sander marinus). The species had vanished from the region’s watersheds until 2016, when fishers in the Dnipro-Bug Estuary reported its surprising reappearance. “It’s very sensitive to toxicants and lower salinity,” Afanasyev says. “Now it only exists in a museum.”

y ,

science.org SCIENCE

y g

5 JANUARY 2024 • VOL 383 ISSUE 6678

y

22

g

senic, and copper—and toxicants such as petroleum byproducts and PCBs. The double punch dealt a devastating blow to marine life that couldn’t swim away from the flood. Mollusks and other near-shore sedentary creatures in the bay were wiped out en masse, Minicheva says. The surge of nutrients also ignited masULTIMATELY THE FLOODWATERS reached the sive blooms of short-lived cyanobacteria, sea. On a warm September day, 4 months commonly known as blue-green algae. after the breach, sunbathers lolled on OdeThen, bacteria that consume the algae rapsa’s beach and swimmers frolicked without idly depleted dissolved oxygen, choking fear in the Black Sea thanks to a metal net coastal waters. strung between piers that prevents RusIn August 2023, Ukraine’s military relaxed restrictions on Black Sea research, hoping to glean evidence for war crimes charges. IMB scientists now have permits to dive at four locations in Odesa Bay. They are chronicling what Minicheva expects to be a prolonged period of ecosystem recovery. And they are warily eying an invasive sea snail, the veined rapa whelk DOWNSTREAM IN THE lower (Rapana venosa). The predaDnipro Basin, the surge of tor from the western Pacific polluted floodwaters from Ocean first appeared in the the breach pummeled rare Black Sea in the 1940s and habitats, including Oleshky in recent years has become Sands National Nature Park, a major pest in Ukrainian an 80-square-kilometer prewaters, where it dines on enserve east of Kherson. The demic oysters and other molsecond largest expanse of lusks. “I worry that Rapana Days after Russia launched its full-scale invasion in 2022, Galyna Minicheva risked sand in Europe, it is home to could be the big winner” as conducting fieldwork in the highly saline Kuialnyk Estuary near Odesa, Ukraine. the endangered sandy blind creatures vie to recolonize mole-rat (Spalax arenarius). “Many surely sian mines from drifting into shore. On a marine habitats damaged by the freshdrowned,” says IZAN zoologist Oleksiy hillside patio overlooking the beach, howwater surge, says IMB marine biologist Vasylyuk, who also leads the nonprofit ever, Minicheva’s team was all business. Mikhail Son. Ukrainian Nature Conservation Group. An IMB marine biologist in a wetsuit exResearchers are also hoping the marine Other exceptional endemic species— amined a basket of mollusks and other sea measurements shed light on a mystery. In including Nordmann’s birch mouse (Sicista creatures. A colleague labeled a glass jar March 2023, in a stunning claim that reloriger), a tree-dwelling ant (Liometopum of seawater that will be analyzed for disceived widespread press coverage, a Ukraimicrocephalum), a sand gadfly, two spesolved oxygen and salinity. If it weren’t for nian ecologist asserted that at least 50,000 cies of pearl knapweeds, and a rare birch the war, a more salubrious place to do scidolphins died over 6 months in 2022, pri(Betula borysthenica)—inhabit nearby ecoence would be hard to imagine. “We have marily because of underwater mine blasts systems, including the 900-square-kilomea saying,” Minicheva said with a smile. “In and acoustic damage from high-energy ter Black Sea Biosphere Reserve that’s now Kyiv, science is academic. In Kharkiv, it’s submarine sonar. Other experts dismissed in Russian hands. Vasylyuk fears some of applied. In Odesa, science is shikarna. that figure as a wild exaggeration—“simply those populations are now extinct. But it’s Luxurious.” crazy,” says UkrSCES acting Director too dangerous, he says—and in RussianFor more than a year, Ukraine’s miliViktor Komorin. held areas, impossible—to get out into the tary deemed research in Black Sea waters The truth looks less alarming. In 2022, field and check. too perilous to undertake. But measureabout 900 porpoise and dolphin stranding When the pulse of sludge-choked freshments close to shore in Odesa Bay, where deaths were photo verified in the westwater reached the Dnipro-Bug Estuary, it the Dnipro-Bug Estuary meets the Black ern Black Sea—roughly twice the annual hammered fish species adapted to brackSea, showed that the flood caused salinaverage recorded from 2018 to 2021, says ish water. Taking heavy losses, according ity to drop from 15 parts per million to IZAN mammalogist Pavel Goldin. But in to forecasts by Afanasyev’s team, were 4 parts per million—“practically fresh2023 there were only a handful. And necthe Black Sea roach (Rutilus frisii), the water,” Minicheva says. Monitoring staropsies he and UkrSCES zoologist Karina Dnieper barbel (Barbus borysthenicus), tions in the bay run by the Ukrainian Vishnyakova performed on several of those and the Sarmatian bleak (Alburnus sarScientific Centre of Ecology of the Sea dead animals have revealed no obvious maticus). Particularly heartbreaking, he (UkrSCES) then recorded high concentralinks to pathogens or toxicants released by says, is the presumed demise of the estuations of metals—including cadmium, arthe floods or oil spills. They also did not

p

agricultural runoff. “The amount of wastewater pouring in is the same as before,” says Osadchyi, who notes that determining pollutant concentrations is a “complex and meticulous task.” Compounding that bad news is the loss of mussels that filtered the water. “To a great degree, they purified the reservoir” when alive, Afanasyev says. Until the mussels rebound, wastewater entering the Dnipro above Kakhovka Dam will wash downstream largely unfiltered. The emptied reservoir itself might also be creating a health threat. There, the Dnipro now wends through a mucky patchwork of about 9000 newly formed small lakes and ephemeral ponds. Scientists worry these wetlands are incubating disease-transmitting mosquitoes and other bloodsucking insects, populations of which could explode next spring, Yuryshynets says. “It’s a potential danger for the region.”

p

find internal hemorrhaging—a hallmark of exposure to loud sonar soundings or underwater explosions. Further tests are underway at the University of Padua and the University of Veterinary Medicine Hannover. But so far, “The cause of the deaths is not straightforward,” says Goldin, who has been sharing the results with Ukrainian prosecutors investigating possible war crimes. A fuller picture of the harm inflicted on the Black Sea will only come into focus after hostilities cease. On the eve of the fullscale invasion, UkrSCES had been gearing up for the most ambitious Ukrainian expedition in years to monitor the Black Sea environment. The war scuttled that voyage, and a missile strike on Odesa’s port in July 2022 damaged its research vessel—the Belgica, donated by Belgium in September 2021 and renamed the Borys Aleksandrov after IMB’s previous director, who perished in a fire at the institute in 2019. Restoration work must wait until after Ukraine’s victory, says Komorin, who has already begun to plan an expedition with Ukraine’s Black Sea allies: Bulgaria, Georgia, Romania, and Turkey.

g

After a centurieslong absence, wild asses called kulans are thriving in a Ukrainian reserve.

Reporting for this feature was supported in part by the Richard Lounsbery Foundation. 5 JANUARY 2024 • VOL 383 ISSUE 6678

23

,

think the dam should be rebuilt,” Vasylyuk says. Instead, he and others see great potential in letting nature take its course. “My dream is that the lower Dnipro watershed will be the next hot spot for rewilding,” says Igor Studennikov, executive director of the Centre for Regional Studies. As a model, he points to a recent effort to restore the Tarutino Steppe, southwest of Odesa near the Danube delta region. A decade ago, Oleg Diakov, an ecologist with Rewilding Ukraine, and colleagues began to restore native grasses and animals such as kulans (Equus hemionus)—a native wild ass that disappeared from the Ukrainian steppe centuries ago—to a 5200-hectare sanctuary. After the war, Diakov says, similar reserves could be established around areas in the lower Dnipro Basin where extensive minefields will deter agriculture and other land uses for years to come. “We’d already identified this area before the war as having the greatest potential for large-scale ecological restoration,” he says. But those aspirations— and a full accounting of the environmental cost of the war—will remain unfulfilled, Mosyakin says, until after Ukraine reclaims territory under occupation. “The sooner the war is over,” Goldin says, “the sooner our ecosystems will get a chance to recover.” j

y

SCIENCE science.org

basin soared, and in winter, when ice cover suffocated fish. This approach also offers “a chance to right an historical wrong,” Osadchyi says. When Soviet engineers impounded the reservoir, he notes, they inundated a swath of steppe and wetlands known as the Velykyy Luh, or Great Meadow—an area prized by Cossacks who have lived in southern Ukraine for centuries. A smaller reservoir would spare some of that prized territory, preserving native willows, poplars, and other vegetation that are now colonizing hundreds of square kilometers of the drying basin. Climatologist Svitlana Krakovska, head of the Ukrainian delegation to the United Nations Intergovernmental Panel on Climate Change, is skeptical of plans to reclaim the old lakebed. Before the disaster, she notes, climate modeling forecasted that the lower Dnipro Basin would become much drier. Evaporation at the Kakhovka Reservoir would have mitigated local warming, she says. Without a reservoir, “The region will only get hotter and drier.” Krakovska also anticipates that the willows and poplars emerging in the former reservoir’s lakebed will wither as the water table recedes. “Sure, there will be a different ecosystem,” she says. “But I’m afraid it’s impossible to go back to what it was like there in the middle of the 20th century.” Still, some experts would prefer to try to rewind the clock even further. “I don’t

y g

sion on whether the Kakhovka Dam will be rebuilt. For the Ukrainian government, a paramount concern is economic recovery. Reimpounding the reservoir could entice residents back to abandoned homes, weekend dachas, and fishing boats along the former shoreline. And it would ease concerns about a future restart of the Zaporizhzhia Nuclear Power Plant, which drew cooling water from the reservoir and, since the dam’s destruction, has had to rely on hastily dug wells for water to cool its reactor cores and spent fuel. Osadchyi, who is part of an expert group from NAS that is assessing options for future water infrastructure there, says a failure to rebuild the reservoir “would be another disaster.” Refilling it would allow mussels to reestablish themselves and begin to filter the water again and provide timely water releases for downstream needs such as supplying municipalities, irrigating crops, and sustaining ecosystems. Afanasyev favors an option he calls “build back better.” It would involve creating a narrower and deeper reservoir by repairing the Kakhovka Dam and confining the water within a 50-kilometer-long barrier (see graphic, p. 20). The rationale is that much of the Kakhovka Reservoir was less than 2 meters deep. “It was essentially stagnant,” Afanasyev says. Fish kills, he says, occurred frequently: in summer, when water temperatures in the shallow

y

PHOTO: REWILDING UKRAINE

SCIENTISTS ARE ALSO awaiting a final deci-

Nuclear disasters–in–waiting

The Zaporizhzhia Nuclear Power Plant, as seen from a Ukrainian position across the Dnipro River.

p y ,

PHOTO: DMYTRO SMOLIENKO/SIPA VIA AP IMAGES

y g

5 JANUARY 2024 • VOL 383 ISSUE 6678

y

SCIENCE science.org

The presence of IAEA observers at the Zaporizhzhia station since September 2022 has deterred the theft of dirty-bomb ingredients. But a major missile strike on one of its spent fuel repositories could turn the plant itself into a dirty bomb, spreading radioactive contamination in a radius of up to 30 kilometers, says Volodymyr Borysenko, a nuclear engineer with the National Academy of Sciences of Ukraine’s Institute for Safety Problems of Nuclear Power Plants (ISPNPP). Even a smaller strike could contaminate the reactor complex. And the spent fuel is also at risk from repeated electricity blackouts that have struck the plant, the latest in early December 2023. Diesel-fueled generators can supply power for up to 10 days, but a prolonged outage could be dangerous, as power is needed to pump cooling water into the plant’s uranium reactor cores and pools holding spent fuel. A lesser known radioactive risk is situated about 150 kilometers upstream from the Zaporizhzhia plant on the Dnipro River. During the Cold War, the Prydniprovsky Chemical Plant was one of Europe’s largest uranium ore processing facilities. The complex accumulated some 40 million tons of tailings—leftovers of milling uranium—and other foul residues before it closed in 1992. By early 2022, Ukraine, with help from the European Union, had fenced off highly contaminated areas. But a missile or artillery strike on a tainted building or dump could disperse radioactive dust over the nearby city of Kamianske. One relative bright spot is Chornobyl, where Ukrainian scientists are restoring labs damaged early in the war. But large parts of the exclusion zone remain off limits because of the threat of mines and unexploded ordinance, says ISPNPP Director Anatolii Nosovskiy. Complicating matters for radiation monitoring, he says, the Ukrainian army has built defensive fortifications in the zone, near the border with Belarus. Ukraine’s darkest fear is a Russian attack with a tactical nuclear weapon, a saber that Putin and other officials rattle from time to time. The State Emergency Service of Ukraine and other agencies need to develop plans for aid and evacuation, says Oleg Voitsekhovych, head of radiation monitoring at the Ukrainian Hydrometeorological Institute, which reports to the State Emergency Service. “To be honest, we are not yet ready.” He’s fervently hoping for reason to prevail. “The most important thing,” Voitsekhovych says, “is to stop this madness and prevent the escalation of nuclear terrorism on the territory of Ukraine.” –R.S.

g

H

aving taken a heavy toll on Ukraine’s ecosystems and water resources, the war with Russia threatens to create a another environmental disaster: damage to the region’s extensive nuclear infrastructure—including 15 power reactors and three research reactors. “There continues to be a highly precarious nuclear safety and security situation across Ukraine,” International Atomic Energy Agency (IAEA) Director General Rafael Mariano Grossi said in a statement after explosions were heard near the Khmelnitsky Nuclear Power Plant and its two Soviet-era reactors on 28 November 2023—the second near-miss in a single month at the site. “All of Ukraine’s nuclear facilities remain vulnerable, either directly if hit by a missile or indirectly if their off-site power supplies are disrupted.” Russia’s assault on Ukrainian nuclear sites began on the very first day of the full-scale invasion. On 24 February 2022, troops overran the Chornobyl Nuclear Power Plant, infamous for the explosion and fire there in 1986 that sent a plume of radioactive smoke into Western Europe. During 5 weeks of occupation, Russian soldiers ransacked labs and kicked up radioactive soil and dust as they dug trenches and slogged through contaminated forests in the exclusion zone around the defunct plant. To the east that spring, Russian troops frequently shelled the Kharkiv Institute of Physics and Technology, damaging a hall containing a subcritical nuclear reactor. Shelling has also flared up repeatedly around the Zaporizhzhia Nuclear Power Station, a complex of six reactors that constitutes Europe’s largest nuclear power plant. Russia captured the plant in March 2022 and the reactors were shut down 6 months later, eliminating the risk of a core meltdown. Still, a prodigious amount of nuclear material remains there: The reactor halls hold 1380 tons of fresh and spent uranium oxide fuel, and two repositories store an additional 2100 tons of spent fuel laced with nasty longlived radionuclides—the ingredients, many Ukrainians fear, of a “dirty bomb” that would use conventional explosives to spread radioactive isotopes. Russian propaganda often claimed that Ukraine plans to fashion a dirty bomb from such materials. After Russian President Vladimir Putin amplified those accusations in October 2022, “Russian hysteria over a dirty bomb was rampant,” says Maksym Strikha, a physicist at the Academy of Sciences of the Higher School of Ukraine who tracks Russian nuclear disinformation. For their part, Ukraine and its allies worry Russia might detonate a dirty bomb and blame Ukraine.

21

p

PERSPECTIVES g

EVOLUTION

y

Evolutionary paths to new phenotypes Ecological model systems inform on innovative traits in plants and animals

y g

By Kathryn R. Elmer

SCIENCE science.org

PHOTO: CHIEN C. LEE

plexity can coalesce has been a challenge to explain since the time of Darwin (1). The stepwise changes in phenotypes over long times, involving many different forms and parts with complex genetic bases, make their existence seem vanishingly unlikely (4). Indeed, during the evolution of these composite traits, the intermediate states might not be beneficial or even functional, for example, as separate components or in disparate combinations. However, advances in quantitative and evolutionary genetics have shown that the combinations of traits and states that are beneficial for an organism can shift substantially and swiftly over time and space. Although it is challenging to predict, the history of selection and evo5 JANUARY 2024 • VOL 383 ISSUE 6678

27

,

School of Biodiversity, One Health & Veterinary Medicine, University of Glasgow, Glasgow, UK. Email: [email protected]

Springboard trapping is a composite trait that arose in carnivorous pitcher plants through coincident spontaneous evolution.

y

A

daptation by natural selection cannot take any evolutionary path; it operates within the constraints of genetic variation and environmental context, with futures contingent on the past. Therefore, how new suites of traits arise is an enduring issue and is key to understanding the diversity of life (1). On pages 108 and 114 of this issue, Chomicki et al. (2) and Stankowski et al. (3), respectively, investigate two different cases of fascinating biological complexity that arose through convergent and convoluted evolutionary paths—one in carnivorous pitcher plants (Nepenthes gracilis and Nepenthes pervillei) and another in

marine periwinkle snails (Littorina saxatilis). The studies use different approaches to reconstruct evolution to reveal how complex phenotypic traits arise in unexpected ways. The results advance understanding not only of the specific traits that are studied—feeding structures in plants and live-bearing (as opposed to egg-laying) in snails—but also how evolution in general might arrive at apparently unlikely combinations. Highly complex organs or composite traits are sets of biological features that, when they come together, have a new synergy and function. As highlighted by Chomicki et al., an example of a composite trait is that of bird wings, which transformed from existing skin and forelimbs into new structures that enable flight (1). Other classic examples include animal eyes, for which many different parts evolved from different ancestral forms to operate together. How such com-

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

10. 11. 12. 13. 14.

10.1126/science.adm9239

C

ontrolling the size and curvature of materials with dimensions in the nanometer scale is a challenge despite the abundance of natural nanostructures, such as those that interfere with light and produce color on butterfly wings. Block copolymers are a class of materials that can spontaneously self-assemble into ordered nanostructures with a range of potential applications, including photocatalysis, nanomembrane separations, photonics, and nanomedicine. However, such achievements have been limited by the challenge of reproducibly synthesizing block copolymers that form a desired network structure. On page 70 of this issue, Lee et al. (1) report a process for reliably producing network nanostructures by linking small functional groups to the end of a diblock copolymer. By varying the chemistry of end groups and linking groups, the authors demonstrate a simple set of design rules for unlocking an unprecedented range of network nanostructures using easily synthesized polymers. A diblock copolymer comprises two chemically distinct polymers that are covalently bound, forming a simple polymer architecture (a single chain) that can selfassemble. A block is a segment of the polymer chain where all the monomers are chemically identical. At equilibrium, diblock copolymers exhibit a range of nanostructures that can be controlled by varying the block fraction (2), which is the volumetric fraction of the total chain that is occupied by a single block. When one block is longer than the other in a diblock copolymer, it is called the majority block and forms a matrix surrounding the shorter, minority block. Within a narrow range of composition for such polymers, a bicontinuous network Department of Chemical and Biological Engineering, University of Sheffield, Sheffield, UK. Email: [email protected] science.org SCIENCE

,

8. 9.

By Alisyn J. Nedoma

y

5. 6. 7.

C. Darwin, The Origin of Species (Dent, ed. 6, 1872). G. Chomicki et al., Science 383, 108 (2024). S. Stankowski et al., Science 383, 114 (2024). S. J. Gould, The Structure of Evolutionary Theory (Harvard Univ. Press, 2002). B. Riska, Evolution 43, 1172 (1989). D. G. Blackburn, J. Morphol. 276, 961 (2015). A. N. Ostrovsky et al., Biol. Rev. Camb. Philos. Soc. 91, 673 (2016). A. J. Helmstetter et al., Nat. Commun. 7, 11271 (2016). J. W. Sites Jr., T. W. Reeder, J. J. Wiens, Annu. Rev. Ecol. Evol. Syst. 42, 227 (2011). B. Crespi, C. Semeniuk, Am. Nat. 163, 635 (2004). H. Recknagel et al., Nat. Ecol. Evol. 5, 1546 (2021). B. Halliwell, T. Uller, B. R. Holland, G. M. While, Nat. Commun. 8, 2030 (2017). F. Racimo, S. Sankararaman, R. Nielsen, E. HuertaSánchez, Nat. Rev. Genet. 16, 359 (2015). M. A. B. Haase et al., Genetics 217, iyaa012 (2021).

Chemically modifying polymer ends enables tailoring of nanostructured materials

y g

1. 2. 3. 4.

Creating the “plumber’s nightmare”

y

5 JANUARY 2024 • VOL 383 ISSUE 6678

R E F E R E N C ES A N D N OT ES

CHEMISTRY

g

28

live-bearers. The diversity in these genetic regions is highly consistent with natural selection driving the evolution of live-bearing, rather than the pattern arising by chance. A key message of the paper is that the genetics of this complex trait, live-bearing, are not reflected in the history of most of the genome, even though the trait is under selection and fundamental to species’ biology. There has been a long-standing debate about whether the accumulation of small stepwise changes or big leaps (saltation) is more important in the evolution of diversity (4). Chomicki et al. and Stankowski et al. did not identify a single big evolutionary step or large-impact mutation that moved the species to a new level of phenotypic innovation. In pitcher plants, it is a case of spontaneous joining of different traits that happen to have come together in the right way. In the periwinkles, it was concluded that a gradual accumulation of the right genetic combinations over long periods of time resulted in live-bearing. This agrees with recent lessons from genomic studies of organisms as varied as humans and microbes: that changes to relatively small genetic regions, not reflected in the history of the overall genome, can have outsized impacts on phenotypes (13, 14). The genetic means by which innovative new traits evolve have been hard to pin down in ecological model species. It is notable that in the era of big genome data, neither Chomicki et al. nor Stankowski et al. pinpoint the genetic variants that cause the traits they focus on. This leaves open the next important question for both species: What types of genetic changes are responsible for these complex and composite traits? To address this will require excellent genomelevel reconstructions and functional tests of exactly which genes matter for phenotypic outcomes. The amazing breadth of plant and animal diversity across the globe has evolved by circuitous paths, and resolving the complex history of genomes and traits unlocks new depths for understanding evolution. j

p

lution can be highly flexible and variable under realistic scenarios (5). Chomicki et al. examined the evolution of the so-called springboard-trapping feature in carnivorous pitcher plants. This feature consists of structural, chemical, and mechanical aspects: a flat, horizontal lid with a waxy underside, which is triggered by rain to fling insects into the pitcher. For springboard trapping to evolve once is an elegant outcome, but the authors’ phylogenetic reconstructions indicate that the feature evolved separately in two different species of pitcher plant. What type of evolutionary path can result in the generation of such a distinctive combination twice independently? The authors used advanced models to predict how the traits evolved across the phylogeny. One option was that selection pushed the suite of traits together into their current distinctive form. Although intuitive, this was not supported by the results. Instead, the models strongly suggest that the new composition arose by “spontaneous coincidence” of the independent components coming together, facilitated by high levels of phenotypic variation. Another complex composite trait that has evolved through seemingly unlikely intermediate stages is that of females bearing live offspring (viviparity) rather than laying eggs (oviparity). This key innovation has evolved in a discrete way independently many times: in invertebrates (such as snails and insects), anamniotes (such as fish and amphibians), and amniotes (such as reptiles and mammals) (6, 7). At a macroevolutionary perspective, live-bearing tends to be associated with higher rates of lineage and species diversification, varying rates of extinction, and expansion into new geographic areas (8, 9). At a microevolutionary level, it has been linked to shifts in lifehistory traits that are important for daily existence—for example, changes in reproductive output, mating and parental care behavior, and using new habitats (10–12). Otherwise indistinguishable egg-laying and live-bearing forms of marine periwinkle snails are found on beaches across many different shores of the northeastern Atlantic Ocean. Stankowski et al. used the DNA of live-bearing and egg-laying forms to reconstruct the evolution of these populations of snails, with the aim of identifying the origin of live-bearing. By applying an advanced analytical method, they found that the genomic regions associated with egg-laying or live-bearing have a different evolutionary history than does the genome overall. These reproductive mode–associated regions involved the same genes that were found to be expressed differently in the female reproductive tissues of egg-layers and

structure forms, known as a gyroid (3, 4). In this structure, the minority block forms two interlacing networks in which three tubes intersect, whereas the majority block forms a matrix surrounding this network (see the figure). The junction between the two block constituents is located at the interface where the minority block extends into the tube formation and the majority block extends into the matrix. Each diblock exhibits a pathway of connected polymer, from one end of a sample to the other, so that the structure is continuous. Network structures can improve the mechanical properties of the material (5); simultaneously transport two quantities, such as ions and electrons (6); and give rise to specific optical and electronic properties (7). They are notoriously difficult to synthesize because of the narrow range of block

nightmare” mesophase. The former has four intersecting tubes that constitute the minority block, whereas the latter has six intersecting tubes (hence, the nightmare). Thermodynamically equilibrated mesophases are attractive because they tend to recover their equilibrium size and shape after a perturbation, which makes them more robust to different processing conditions (12). Lee et al. demonstrated thermal reversibility by ramping up the temperature until the materials undergo a phase transition and then cooling the sample down to room temperature. In all instances, the authors recovered the original mesophase, which provides very strong evidence that the room-temperature network nanostructures are thermodynamically equilibrated. Lee et al. further showed that the transformations between network

Two chemically distinct polymers that are covalently bound form a simple polymer architecture—the diblock copolymer. The diblock can be modified with chemical groups appended to the ends of constituent polymer chains. This enables the creation of structures never before realized, such as the diamond cubic and the plumber’s nightmare cubic, which have four and six intersecting tubes, respectively, embedded in a polymer matrix. Polymer 2 Linker

End-end chain interactions

Chain 2 Tube

Matrix

Tube

GRAPHIC: N. BURGESS/SCIENCE

mesophases occur epitaxially, which could lead to the application of these materials as thermally actuated switches. Lee et al. introduced two parameters to qualitatively guide the design of network structures: the strength of interactions between end groups and the degree of packing frustration, which is the compression of polymer chains at the intersection between tubes in the mesophase (and is modulated by the linking group). At the intersection between tubes, the diblock copolymer chains must compress to fit into the small volume of the intersection node. Strong end-end interactions cause the polymer chains in the majority block to stretch to accommodate the close packing of the end groups. The resulting mesophase represents the equilibrium balance between minimizing the interface between blocks, minimizing chain stretching, and constraining the system to fill space uni-

RE FE REN C ES AN D N OT ES

,

SCIENCE science.org

(Im3m) Plumber’s nightmare cubic

1. H. Lee et al., Science 383, 70 (2024). 2. F. S. Bates, G. H. Fredrickson, Annu. Rev. Phys. Chem. 41, 525 (1990). 3. D. A. Hajduk et al., Macromolecules 27, 4063 (1994). 4. M. W. Matsen, F. S. Bates, J. Chem. Phys. 106, 2436 (1997). 5. B. J. Dair et al., Macromolecules 32, 8145 (1999). 6. Y. Jung, S. Torquato, Phys. Rev. E Stat. Nonlin. Soft Matter Phys. 72, 056319 (2005). 7. K. Hur et al., Angew. Chem. Int. Ed. 50, 11985 (2011). 8. J. Park, K. I. Winey, JACS Au 2, 1769 (2022). 9. A. C. Finnefrock, R. Ulrich, G. E. S. Toombes, S. M. Gruner, U. Wiesner, J. Am. Chem. Soc. 125, 13084 (2003). 10. F. J. Martínez-Veracoechea, F. A. Escobedo, Macromolecules 42, 1775 (2009). 11. M. Stefik et al., Chem. Mater. 21, 5466 (2009). 12. F. S. Bates, G. H. Fredrickson, Phys. Today 52, 32 (1999). 13. F. J. Martínez-Veracoechea, F. A. Escobedo, Macromolecules 42, 9058 (2009). 14. P. F. W. Simon, R. Ulrich, H. W. Spiess, U. Wiesner, Chem. Mater. 13, 3464 (2001). 15. H. Y. Hsueh et al., Adv. Mater. 23, 3041 (2011). 10.1126/science.adn0168 5 JANUARY 2024 • VOL 383 ISSUE 6678

y

fractions in which they form (8). Gyroid mesophases have been formed by blending additional components into the diblock copolymer (9, 10) or by synthesizing a more complex triblock polymer (11). Lee et al. discovered a synthetically facile route for linking end groups onto an existing diblock copolymer to induce the formation of equilibrium network nanostructures. The method tolerates block fractions that vary from 38 to 48 vol % for the minority block, which extends the design space from the conventional 37 to 39 vol % window in which gyroids form (8). The ability to design a single network-forming entity removes the need for blending and simplifies the synthetic chemistry. Beyond the gyroid mesophase, Lee et al. observed two structures with diblock copolymers that had only been theorized: the diamond cubic and the so-called “plumber’s

(Pn3m) Diamond cubic

y g

(Ia3d) Double gyroid

y

Chain 1

Chain packing

End group

g

Polymer 1

p

A diversity of network structures

formly (13). Lee et al. found that the strength of end-end interactions that is required to induce network formation depends on the equilibrium curvature of the unfunctionalized diblock copolymer. The authors’ design rules provide a mechanistic basis for generating new block copolymer nanostructures. Scaling up the use of network block copolymer nanomaterials remains a challenge due to subtractive manufacturing techniques. Polymer structures are used to template inorganic materials and then the polymer is etched away or calcined (14, 15). The polymer lost in these subtractive processes is generally unrecoverable, which leads to higher manufacturing costs and greater embodied carbon within the final product. The design strategy of Lee et al. encourages a more efficient use of materials by enabling researchers to use functional polymers that possess some of the attributes of inorganic materials, such as conductivity. Creating network structures in which the polymer blocks perform the desired function reduces waste and is likely to decrease the final cost of a technology. Only a limited number of polymer pairs have been used to form network structures because self-assembly depends on the specific enthalpic interaction between the polymer species (13). Lee et al. used linker groups and end groups to introduce an additional degree of freedom that allows strong interactions between the end groups to compensate for weaker interactions between the polymer blocks. They demonstrate the versatility of their technique by applying end groups to different species of diblock copolymers, thereby showing that network structures form for both semicrystalline and amorphous polymer blocks. The tunability of end-group chemistry broadens the range of potential polymer combinations that can form network nanostructures, which will enable polymer chemists to select blocks based on desired physicochemical properties rather than their ability to self-assemble. j

29

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

IMMUNOLOGY

Restoring tolerance with antigen delivery Strategies that modulate antigen delivery are being tested to reverse autoimmunity By Bana Jabri1,2,3,4 and Valérie Abadie1,2,3

A

,

science.org SCIENCE

y

5 JANUARY 2024 • VOL 383 ISSUE 6678

y g

30

y

University of Chicago Department of Medicine, Chicago, IL, USA. 2University of Chicago Celiac Disease Center, Chicago, IL, USA. 3University of Chicago Committee on Immunology, Chicago, IL, USA. 4University of Chicago Department of Pathology, Chicago, IL, USA. Email: [email protected]

g

1Section of Gastroenterology, Hepatology and Nutrition,

Although promising results with these approaches have been reported in animal models of autoimmunity, they have not yet given rise to an efficacious antigen-specific immunotherapy that provides a significant clinical benefit. This may be for several reasons, including the degradation of antigens before they reach their target, a failure to delete pathogenic tissue-resident T cells, and the emergence of de novo disease-driving antigens. To improve these approaches by promoting the differentiation of antigen-specific Treg cells, delivery of the antigen by bacteria that are genetically manipulated to secrete regulatory cytokines, such as IL-10, has been investigated. This improved the mucosal delivery of the antigen and induced more immunosuppressive Treg cells in recent-onset nonobese diabetic (NOD) mice (9). The expansion of FOXP3+ Treg cells can also be augmented by coadministration of IL-2 muteins, which are IL-2 variants specifically designed to interact exclusively with the high-affinity IL-2 receptor. This approach demonstrated prolonged control of autoimmunity in the NOD mouse model (10). However, if IL-2 muteins are provided alone, Treg cells expand in a non–antigen-specific manner. Furthermore, they might not be able to suppress effector T cell responses because of the inflammatory nature of the tissue environment targeted by the autoimmune condition. Another approach involves bioengineering that relies on either conjugating the antigen to polymers (11) or nanotechnology (6) to target different subsets of APCs. For example, to better harness the tolerogenic environment of the liver, antigens can be conjugated to linear polymeric glycosylations, such as N-acetylgalactosamine and N-acetylglucosamine, that are recognized by C-type lectin receptors expressed by hepatic dendritic cells, Kupffer cells, liver sinusoidal endothelial cells, and hepatocytes. This results in increased uptake and presentation of these synthetically glycosylated antigens by liver APCs, which are poised to maintain tolerance. This approach provided encouraging results in the BDC2.5 T cell adoptivetransfer mouse model of type 1 diabetes (11). Intravenous administration of antigens coupled to erythrocytes, which undergo sustained cell death called eryptosis, also leads to continuous uptake of the coupled antigens by splenic dendritic cells, hepatocytes, and hepatic stellate cells. In animal models, these

p

utoimmune disorders encompass a wide range of immunological diseases that affect 4 to 10% of the population worldwide. They occur when immunological tolerance toward self-antigens is broken, resulting in immune responses against cells, tissues, or organs that lead to tissue dysfunction and/or destruction. The goal of autoimmune disease therapy is to restore tolerance to the self-antigen that causes the pathology by targeting autoreactive T cells while preserving immune competence to prevent infections and malignancies. The main challenge of this endeavor resides in choosing the mode by which the disease-driving antigen is delivered so that it can initiate the removal or reprogramming of the autoreactive T cells or the induction and/or expansion of antigen-specific regulatory T (Treg) cells to suppress autoreactive T cells. Although the potential of antigenbased immunotherapy approaches to restore tolerance have been demonstrated for the treatment of immunoglobulin E (IgE)– mediated peanut allergy (1), antigen therapies for autoimmune disorders are still being developed. There are several key hurdles that a successful autoimmune disease antigen therapy needs to overcome. For most autoimmune disorders, there is absent or incomplete knowledge of the disease-driving autoantigen. In addition, although one particular disease-causing antigen triggers and drives the autoimmune process, additional antigens are exposed as tissue destruction takes place, which causes a phenomenon called epitope spreading, whereby the pathogenic T cell response widens to a diversity of antigens (2). Furthermore, a successful therapy needs to target functionally relevant subsets of both tissue-resident and circulating pathogenic memory T cells. This is not easily achieved owing to the incomplete understanding of the regulation of tissue-resident T cells and given that there are important distinctions in their mode of regulation (3) and transcriptional program (4). Moreover, even if Treg

cells are induced, they need to be functional in an inflammatory environment. This is, for example, problematic for forkhead box P3–expressing (FOXP3+) Treg cells that fail to control pathogenic effector T cells that have been stimulated by the inflammatory cytokines interleukin-15 (IL-15), IL-21, or IL-7 that are present in the autoimmune tissue environment (3). To attempt to circumvent at least some of these issues, diverse approaches have been developed with the common goal of restoring immune tolerance by fine-tuning antigen delivery to tolerogenic or resting antigenpresenting cells (APCs) that will induce Treg cell differentiation, T cell deletion, or anergy (a state of hyporesponsiveness). The rerouting of the T cell response involves two possible mechanisms. One results in pathogenic T cells acquiring tolerogenic properties, such as the ability to produce the immunosuppressive cytokine IL-10. The other depends on the expansion and/or de novo generation of inducible FOXP3+ Treg cells or type 1 regulatory T (Tr1) cells from naïve T cells. The definition of Tr1 cells remains ambiguous because they can coproduce IL-10 and inflammatory cytokines or produce IL-10 only (5, 6). Furthermore, it is important to note that the beneficial effects of IL-10 are context dependent because it can have detrimental effects in certain autoimmune diseases by promoting extrafollicular B cell responses and the production of autoantibodies (7). Several approaches are being tested in clinical trials (see the table). One strategy is based on delivering the disease-driving antigens in their “naked” form (in the absence of any vehicle), leading to antigen presentation in a tolerogenic environment. This may be achieved by giving the antigen orally, a strategy that has proven successful in the context of peanut allergy, so that the antigen is presented by intestinal tolerogenic dendritic cells that promote the differentiation of FOXP3+ Treg cells (oral tolerance). Alternatively, the disease-driving antigen can be delivered outside of the site of the inflamed tissue to resting dendritic cells that lack costimulatory molecules by injecting immunodominant peptides subcutaneously or intramuscularly. Activating memory T cells and naïve T cells through the T cell receptor in the absence of costimulation by CD28 and CD40 ligand leads to the deletion (and anergy) of the antigen-specific T cells (8).

Modes of antigen delivery Systemic or mucosal administration of “naked” antigens targets resting or tolerogenic dendritic cells (DCs) to promote the deletion or induction of forkhead box P3–expressing (FOXP3+) regulatory T (Treg) cells, respectively. To target additional types of antigen-presenting cells (APCs), several bioengineering approaches have been developed. Antigens can be conjugated or delivered in nanoparticles. Nanoparticles can not only deliver antigens to many different types of APCs at different locations but also deliver therapeutic cargo. Bioengineering approaches have the potential to promote more wide-ranging mechanisms of tolerance by inducing regulatory type 1 and type 1–like cells and reprogramming pathogenic autoreactive T cells. Naked antigen delivery Approach

Low-dose oral antigen administration

Immune priming by resting DCs (no signal 2)

Immune priming by intestinal tolerogenic DCs

Deletion or anergy of autoreactive T cells; de novo induction of Treg cells?

De novo induction of Treg cells

Phase 1 for type 1 diabetes (Cardiff University: NCT01536431) Phase 2 for relapsing remitting multiple sclerosis (Medical University of Lodz) Phase 2 for celiac disease (ImmusanT: NCT03644069)

Phase 1/2 for type 1 diabetes (Pre-POINT Study Group: ISRCTN76104595)

Mechanism Outcome

Deletion of autoreactive T cells

De novo induction of Treg cells

Reprogramming of autoreactive T cells

Phase 1b/2a for multiple sclerosis (Cellerys AG-2022-000801-28) Phase 2 for celiac disease (Anokion SA: NCT05574010)

5 JANUARY 2024 • VOL 383 ISSUE 6678

31

,

nature of its approach, mRNA therapeutics may transform immune regulatory therapy by enabling efficient delivery of antigen as well as easier development and manufacturing. However, the inherent adjuvanticity of mRNA-based vaccines that can activate innate immune receptors may also lead to the unwanted activation of APCs and the enhancement of autoreactive immune responses. To tackle this issue, some modified mRNA vaccines have been developed by introducing methylpseudouridine in place of uridine. This nucleoside modification results in an anti-inflammatory mRNA vaccine that has reduced innate immune activation properties because it fails to activate Toll-like receptor 7 (TLR7) (15). Despite the impressive development of new technologies that offer the promise of antigen-specific tolerogenic therapies, the problems of epitope spreading and immune regulation in an inflammatory tissue environment remain to be solved. Approaches that can control both pathogenic circulating central memory and tissue-resident effector memory T cells also need to be identified. This will likely only be achieved through combination therapies that are adapted to each tissue and disease. As technologies for antigen delivery and therapeutic strategies with broader and stronger tolerogenic potential are developed, the risk of nonspecific immunosuppression also increases. Another critical issue to consider is establishing the criteria for authorizing human trials and evaluating the feasibility of acquiring pertinent preclinical data. To ensure the future success of these approaches, it will be crucial to incorporate comprehensive and unbiased immunological and multiomics investigations in clinical trials. Although animal models are valuable for

y

increase the presentation of disease-driving antigens by tolerogenic phagocytes and dendritic cells. This has been experimentally achieved in mice by using negatively charged microparticles engineered with polymeric organic materials such as polylactide-co-glycolic acid (PLGA) that were passively taken up by splenic marginal zone macrophages expressing the scavenger receptor MARCO (6). Other approaches consist of using liposomes that contain peptide antigen and phosphatidylserine to facilitate their phagocytosis by macrophages. By adding therapeutic cargo that is delivered to different cell types and in distinct anatomical locations, nanoparticles have the capacity to induce all types of immune modulation and hence be adapted to a given organ-specific autoimmune disorder. This may be critical because each organ and autoimmune disease has a particular immunological and tissue signature that requires different modes of tolerance induction. For example, KAN-101, a glycosylation signature conjugated to deamidated gliadin peptides that is specifically designed for liver targeting, is now advancing into phase 2 clinical trials for the treatment of celiac disease (13). Delivering antigens to anatomically distinct sites can also be achieved by altering the nanoparticle chemistry. Selective organ targeting has been developed, whereby manipulating multiple classes of lipid nanoparticles allows tissue-specific delivery of diverse cargos, including mRNA, Cas9 mRNA–single guide RNA (sgRNA), and Cas9 ribonucleoprotein complexes, which allows CRISPRCas–mediated gene editing in therapeutically relevant cell types (14). Nevertheless, encapsulating many antigens might be necessary to overcome epitope spreading. Although nanoparticle technology is attractive because of the flexibility and holistic

y g

SCIENCE science.org

Targeting of and priming by resting DCs, tolerogenic DCs, or nonprofessional APCs

y

approaches were shown to lead to deletion of autoreactive T cells, de novo induction of Treg cells, or reprogramming of autoreactive T cells. Although phase 1 and 2 clinical trials in patients with multiple sclerosis and celiac disease are ongoing (2022-000801-28 and NCT05574010, respectively), the efficacy of these approaches in humans remains to be demonstrated. The use of nanoparticles offers distinct ways to manipulate the immune system because they can be engineered to carry ligands that target specific cell-surface receptors (e.g., DEC-205, which is highly expressed in dendritic cells) and therapeutic cargos (e.g., tolerance-promoting drugs, such as rapamycin, or small interfering RNA) (6). Modulating the amount of antigen delivered is also more easily achieved with nanoparticles than with naked antigen delivery. Delivering higher doses of antigen may enhance tolerance induction by promoting the gradual establishment of a regulatory program in Tr1 cells that is characterized by the expression of specific negative costimulatory molecules and transcription factors—such as MAF, aryl hydrocarbon receptor (AHR), and nuclear factor interleukin-3–regulated protein (NFIL3)—in addition to IL-10 production and in the absence of inflammatory cytokine production (12). Activation of an immunosuppressive network that involves IL-10–producing Tr1 cells might be critical for establishing bystander suppression owing to the ability of IL-10 to downregulate the expression of costimulatory molecules and major histocompatibility complex class II (MHCII) molecules on the surface of APCs (5). Furthermore, the size, shape, and surface chemistry of nanoparticles can be manipulated to influence trafficking and the functional program that is activated in the targeted APCs. For instance, it is possible to

Nanoparticles

g

GRAPHIC: N. BURGESS/SCIENCE

Clinical trials

Bioengineering Antigen coupling

p

Systemic antigen administration

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

32

5 JANUARY 2024 • VOL 383 ISSUE 6678

H

ow the nervous system deals with uncertainty in the outcome of actions is a fundamental question in cognitive neuroscience. Risky decisions represent a cost-benefit trade-off, whereby the possibility of an outcome worse than a safer alternative (i.e., cost) must be weighed against the possibility of outcomes that are better than the safer alternative (i.e., benefit). This trade-off is entirely subjective and depends on the individual’s willingness to accept risk (i.e., uncertainty about the outcome). Attitude toward risk is strongly influenced by context, but very little is known about the underlying neuronal circuitry. On page 55 of this issue, Sasaki et al. (1) report that two neighboring regions in the frontal cortex of Japanese macaques (Macaca fuscata) can influence risk attitude in a competitive manner. This 1Department of Neuroscience, Johns Hopkins University

School of Medicine and Zanvyl Krieger Mind/Brain Institute, Baltimore, MD, USA. 2Department of Psychological and Brain Sciences, Johns Hopkins University, Baltimore, MD, USA. Email: [email protected]

finding opens up the possibility to uncover the neural circuit mechanism that underlies attitude toward risk. To study the influence of risk independently from the expected value (the probability-weighted sum of all possible outcome values of the gamble), the authors designed a choice task in which monkeys had to choose between sets of gambles with a constant expected value but a changing probability of winning the higher outcome. The gamble options were presented as colored circles, and the monkeys indicated their choice by shifting their eyes to one or the other. Some gambles had small probabilities of winning but a high winning amount [high risk–high return (HH)], whereas others had larger winning probabilities but smaller winning amounts [low risk–low return (LL)]. Because the expected value was the same for HH and LL options, any preference was the result of risk attitude only. Six monkeys were tested, and all showed a preference for HH options. Thus, in line with previous findings, monkeys were overall risk seeking (2–4). Sasaki et al. used muscimol injections to bilaterally inactivate either the orbitofrontal science.org SCIENCE

,

10.1126/science.adg7505

By Veit Stuphorn1,2

y

The authors thank C. Khosla and J. A. Hubbell for insightful comments. The work of B.J. and V.A. is supported by grants from the National Institutes of Health (NIH): R01 DK067180 to B.J., R01 DK128352 to V.A., and Digestive Diseases Research Core Center P30 DK42086 and R01 DK063158 to B.J. and V.A.

Specific brain pathways can lower or raise the willingness of monkeys to take risks

y g

AC K N OW L E D G M E N TS

Dopamine regulates attitude toward risk

y

1. G. Du Toit et al., N. Engl. J. Med. 372, 803 (2015). 2. C. J. Vanderlugt, S. D. Miller, Curr. Opin. Immunol. 8, 831 (1996). 3. B. Jabri, V. Abadie, Nat. Rev. Immunol. 15, 771 (2015). 4. L. K. Mackay et al., Science 352, 459 (2016). 5. M. G. Roncarolo, S. Gregori, R. Bacchetta, M. Battaglia, N. Gagliani, Immunity 49, 1004 (2018). 6. P. Serra, P. Santamaria, Nat. Biotechnol. 37, 238 (2019). 7. S. Biswas, K. Bieber, R. A. Manz, Front. Immunol. 13, 970906 (2022). 8. M. K. Jenkins, R. H. Schwartz, J. Exp. Med. 165, 302 (1987). 9. S. Robert et al., Diabetes 63, 2876 (2014). 10. C. D. Moorman, S. J. Sohn, H. Phee, Front. Immunol. 12, 657768 (2021). 11. D. S. Wilson et al., Nat. Biomed. Eng. 3, 817 (2019). 12. B. R. Burton et al., Nat. Commun. 5, 4741 (2014). 13. J. A. Murray et al., Lancet Gastroenterol. Hepatol. 8, 735 (2023). 14. M. M. Żak, L. Zangi, Pharmaceutics 13, 1675 (2021). 15. K. Karikó, M. Buckstein, H. Ni, D. Weissman, Immunity 23, 165 (2005).

NEUROSCIENCE

g

R E F E R E N C ES A N D N OT ES

p

uncovering and defining general immune concepts, pathways, and molecules, disparities exist between humans and animals in terms of immune cell subsets, innate receptors, and exposure to environmental factors, such as diet, that might affect autoimmune pathogenesis. Consequently, reliably predicting the response to antigen-delivery therapy remains a challenge, even when using animal models that offer a higher degree of physiological relevance to humans than laboratory mice. Successfully navigating the complexities of an individual’s immune system requires a comprehensive approach, with the effectiveness of strategies being contingent upon their alignment with the disease stage. This challenge intensifies as tissue destruction advances, which introduces intricacies to tolerance restoration. Amid the myriad of technical and knowledge advances, it is crucial to recognize the complexity of reinstating tolerance through antigen delivery. Consequently, celiac disease, an autoimmune disorder that targets the intestine, emerges as an exemplary human disease model. It provides a vital platform for exploring the mechanisms that underlie tolerance reinstatement through antigen delivery and understanding the factors that influence therapeutic success or failure. Notably, the strict human leukocyte antigen (HLA) restriction of celiac disease coupled with the knowledge of the disease-driving antigen (gluten) and accessibility of the tissue targeted by the pathogenic immune response make it a valuable focal point for advancing therapeutic interventions aimed at tolerance induction through antigen delivery. j

High risk, high return

Go for it! Take the risk!

Low risk, low return

Better play it safe...

Ventral BA 6V Dorsal BA 6V VTA

dorsomedial frontal cortex involved in eye movement control and evaluation—has a selective effect on risk attitude (4). Neurons in BA 6V could therefore influence the oculomotor system through the supplementary eye field. However, there are multiple other frontal and parietal regions that also influence the selection of eye movements. It will be important to establish whether there are specific and necessary pathways for expressing risk attitude or whether BA 6V has a large influence over a distributed network. Finally, it is not yet clear how effector-specific the signals in BA 6V are. Are they also active when choices are expressed through arm movements instead of eye movements? If yes, how are the abstract risk attitude signals directed to the appropriate motor circuit? j

y g

RE FE RE NC ES AN D N OT ES

AC KN OW LED G M E N TS

V.S. was supported by National Institutes of Health grant R01DA049147. 10.1126/science.adm8641 5 JANUARY 2024 • VOL 383 ISSUE 6678

33

,

1. R. Sasaki et al., Science 383, 55 (2024). 2. A. N. McCoy, J. C. Crowley, G. Haghighian, H. L. Dean, M. L. Platt, Neuron 40, 1031 (2003). 3. W. R. Stauffer, A. Lak, P. Bossaerts, W. Schultz, J. Neurosci. 35, 3146 (2015). 4. X. Chen, V. Stuphorn, Curr. Biol. 28, 3114 (2018). 5. P. H. Rudebeck, R. C. Saunders, D. A. Lundgren, E. A. Murray, Neuron 95, 1208 (2017). 6. M. O’Neill, W. Schultz, Neuron 68, 789 (2010). 7. I. E. Monosov, Nat. Commun. 8, 134 (2017). 8. E. Y. Walker et al., Nat. Neurosci. 26, 1857 (2023). 9. T. Sawaguchi, P. S. Goldman-Rakic, Science 251, 947 (1991). 10. M. Wang, S. Vijayraghavan, P. S. Goldman-Rakic, Science 303, 853 (2004). 11. T. Ott, A. M. Stein, A. Nieder, Nat. Commun. 14, 7537 (2023).

y

rior cingulate cortex (7). Alternatively, risk attitude could be influenced by modifying the nonlinear response of value-sensitive neurons to reward magnitude. For example, if neurons in BA 6V become much more active to very high reward magnitudes compared with low ones, the monkeys would be more sensitive to the high winning amounts of the HH option. This could make them more risk seeking without any direct change in probability representation. In that case, many BA 6V neurons should represent reward amount but not risk. A mixture of both of these effects is also possible. Alternatively, uncertainty could be indirectly encoded in the distribution of value-prediction signals across the population of neurons (8). In that case, changes in risk attitude could be caused by changes in the value distribution. Dopamine has both enhancing and suppressing effects on prefrontal activity (9–11). Therefore, it will be important to understand the exact effects of dopamine on neuronal activity in BA 6V. Likewise, differential modulation of risk attitude requires that only the VTA neurons projecting to the required area are active. Therefore, looking for separate VTA projections to the ventral and dorsal BA 6V is another important next step. It is also not yet clear through which pathways BA 6V influences behavior. The monkeys in this study expressed their choices through eye movements. Therefore, a specific risk attitude must express itself ultimately in the increased activity of neurons promoting eye movements to targets that are most in line with the current risk preference. Inactivation of the supplementary eye field—a region in the

y

SCIENCE science.org

Activation of pathways from the ventral tegmental area (VTA) to two different parts of the ventral section of Brodmann area 6 (BA 6V) has opposite effects on risk attitude. Activation of VTA projections to dorsal BA 6V increases preference for low risk–low return options. With this option, monkeys have a high (90%) probability of receiving a small (111 µl) amount of fluid reward. Activation of projections to the ventral BA 6V increases preference for high risk–high return options. With this option, monkeys have a small (10%) probability of receiving a large (1000 µl) amount of fluid reward. The expected value of both options is equal (100 µl).

g

cortex, the dorsal anterior cingulate cortex, or the ventral section of Brodmann area 6 (BA 6V) while the monkeys performed the task. These are all frontal brain regions thought to be candidates for a causal role in influencing risk attitude. Only inactivation of BA 6V had an effect on task outcome. This is in good agreement with a previous lesion study of a similar section of the frontal cortex, which was referred to as the ventrolateral prefrontal cortex (5). To investigate the role of reward feedback on the risk attitude control network, Sasaki et al. studied the dopamine-expressing neurons that project to BA 6V from the ventral tegmental area (VTA). To do this, they combined optogenetic manipulation of VTA projections with a chronically implanted set of electrocorticogram electrodes on the lateral surface of BA 6V. This allowed them to selectively activate the dopaminergic VTA input to specific areas of BA 6V and to observe the effects on behavior and neuronal activity. Notably, optogenetic activation of the VTA–BA 6V pathway had opposite effects on risk attitude in dorsal and ventral sections of BA 6V. Activation of VTA input to the ventral part of BA 6V led to a stronger preference for HH options—i.e., the monkeys became more risk seeking. However, activation of the VTA input to the dorsal aspect resulted in the opposite effect: The monkeys preferred LL options more often—i.e., they became more risk averse (see the figure). This unexpected finding implies that two neighboring regions in the frontal cortex together regulate risk attitude in a competitive push-pull–like fashion and can both increase and decrease risk seeking. This is important because it opens up the possibility to identify the neuronal mechanisms in the circuit underlying this ability. The extent of the functional area that regulates risk attitude is not yet clear. Combining the findings of Sasaki et al. with those of the previous lesion study (5), it appears that an extensive region of the cortex, ranging from Walker’s areas 12, 45, and ventral 46 on the orbital surface to BA 6V on the lateral surface, is potentially involved in setting risk attitude. The functional boundaries of this extended area and further functional subdivisions need to be mapped out more carefully. In principle, attitude toward risk could be modified in two different ways. Changing preference to risk directly would influence risk attitude. In that case, BA 6V neurons should be sensitive to risk but not to expected value. Such neurons have been found in the orbitofrontal cortex (6) and the ante-

Assessing risk

p

GRAPHIC: A. MASTIN/SCIENCE

A Japanese macaque (Macaca fuscata) eats wild flowers. Attitudes toward reward, such as food, are regulated by the ventral section of Brodmann area 6.

I NS I GHTS

P OLICY FORUM BIODIVERSITY

Biodiversity monitoring for a just planetary future Data that influence policy and major investment decisions risk entrenching social and political inequities

5 JANUARY 2024 • VOL 383 ISSUE 6678

science.org SCIENCE

,

34

y

1National Center for Ecological Analysis and Synthesis, University of California, Santa Barbara, Santa Barbara, CA, USA. 2Department of Environment Science, Policy, and Management, University of California, Berkeley, Berkeley, CA, USA. 3School for Environment and Sustainability, University of Michigan, Ann Arbor, MI, USA. 4Department of Ecology and Evolutionary Biology, Yale University, New Haven, CT, USA. 5Department of Society and Conservation, University of Montana, Missoula, MT, USA. 6Department of Global Development, Cornell University, Ithaca, NY, USA. 7Department of Computer Science, University of Cambridge, Cambridge, UK. 8Département de sciences biologiques, Universite de Montreal, Montreal, Canada. 9Department of Sociology, Tufts University, Medford, MA, USA.10School of Engineering and Applied Sciences, Harvard University, Cambridge, MA, USA. Email: [email protected]

y g

FROM DATA TO DECISIONS Although the impact of data disparities on decisions is central to discussions on data governance throughout society—from policing to finance to health care—the environmental domain has skirted many of these critiques under the guise that its data reflect and affect the natural world, not people, politics, and histories. The ecological community agrees that data disparities exist but has yet to assess how those disparities propagate through derived ecosystem indicators and policy and management decisions. There are several ways in which data disparities might be reflected in scienceinformed decisions in the context of global biodiversity targets. For example, extensive data collected within government-managed parks compared to community-managed and Indigenous lands (11) might lead to systematic underestimates of biodiversity

y

E

A GLIMPSE INTO GLOBAL BIODIVERSITY DATA The systems that generate biodiversity data are complex, uneven, and ultimately human. Species observations reflect human processes across space and time: from the decadal impacts of colonialism to the weekly sway of work schedules in modern society, from geopolitical strife to neighborhood-scale disparities. Take, for example, the Global Biodiversity Information Facility (GBIF). GBIF is a data repository that synthesizes billions of species observations across the globe (see the figure) and specifically aims to provide global-scale biodiversity data to underpin policy and inform decisions from invasive species management to priorities for conservation investment. Even at first glance, GBIF data do not reflect latitudinal gradients of biodiversity, but more closely trace macroeconomic patterns (see the figure). These data disparities are unsurprising to most ecologists and, like the overrepresentation of population centers, roads, and protected areas in global species observations (3), are increasingly adjusted for, even if imperfectly, within existing modeling frameworks (4). But digging deeper into these data, strikingly uneven patterns of data availability reveal signatures of armed conflict (see the figure) (5), the legacy effects of racist policies (see the figure) (6), and changes in political regimes (5). Although descriptions of how biodiver-

cologists and conservation scientists have long acknowledged that biodiversity data reflect legacies of social inequity (see the figure). Although the ramifications of these disparities were easy to dismiss when the application of large-scale biodiversity data was limited to academic biogeography and theoretical conservation prioritizations, the stakes have changed. Biodiversity data carry more influence than ever before (1), guiding the implementation of massive multilateral commitments and global investments that will affect nature and people for decades to come—from informing priorities for more than doubling the global area under conservation management to creating international biodiversity offset markets. We examine two contentious questions that arise as we consider the disparities in biodiversity data and their consequences in the wake of contemporary biodiversity policy: Are the best available data really a suitable standard? Can more data and better statistical methods ensure that inequities aren’t entrenched when implementing data-driven solutions? With hundreds of billions of dollars being invested in conserving, restoring, and sustainably managing ecosystems in the wake of the post-2020 Kunming-Montreal Global Biodiversity Framework (GBF) (2), an understanding of the ways in which data biases propagate through decision-making is critical for the effective creation and communication of data-driven solutions to

g

global biodiversity loss. The path forward will require more than technocratic fixes. Interdisciplinary collaboration and inclusive, bottom-up processes will be critical for leveraging past, present, and future biodiversity data in a way that aligns with the equity goals of global biodiversity policy.

p

By Melissa Chapman1,2, Benjamin R. Goldstein2, Christopher J. Schell2, Justin S. Brashares2, Neil H. Carter3, Diego Ellis-Soto4, Hilary Oliva Faxon5, Jenny E. Goldstein6, Benjamin S. Halpern1, Joycelyn Longdon7, Kari E. A. Norman8, Dara O’Rourke2, Caleb Scoville9, Lily Xu10, Carl Boettiger2

sity data disparities trace social and political inequity are notable (3, 5, 6), they rarely provide the insights necessary to causally attribute mechanisms of those disparities. Human patterns captured by biodiversity data undoubtedly include observational biases but also reflect a reality of the Anthropocene: People—and our politics, economies, and histories—are major drivers of ecosystem composition.  European colonial history is still detectable in the true distribution of alien floras worldwide (7). Armed conflict affects underlying ecological processes in a variety of complex ways (8). The legacy of residential segregation has influenced greenspace and tree cover across neighborhoods, which in turn affect habitat for and distribution of urban wildlife (9). To add complexity, environments most degraded by extractive infrastructure are often the most monitored—extractive infrastructures are often also (biodiversity) knowledge infrastructures. For example, the Sacramento–San Joaquin Delta in California is subject to a tremendous amount of ecological monitoring, established as a political compromise to assess the effects of building California’s complex water infrastructure (10). But will disparities in biodiversity data really translate to ineffective and inequitable decisions for nature and people? And if so, what can be done about it given the urgency of the biodiversity crisis and the immediacy of informing global policy implementation?

Biodiversity data reflect legacies of social inequity (1) The >2.6 billion species observations in the Global Biodiversity Information Facility (GBIF) database are disproportionately from high-income countries. (2) These macroeconomic disparities in data density have become more pronounced through time. (3) There are fewer species observations in places and times of conflict (5). For example, biodiversity observations notably declined during the Cambodian Civil War, which began in 1970, and especially the Cambodian genocide (1975–1979), and remained low during the following decade of armed conflict. (4) In the United States, biodiversity observations unearth the legacy of the effects of racial and ethnic discrimination in housing policy in the 1930s (“redlining”) (6). Neighborhoods that were redlined, or deemed “hazardous,” have approximately half the bird observation density today of those neighborhoods that were deemed “safe” investments (6). (5) Human histories are reflected not only in where and when data are recorded but also who collects, published, and owns data. In Nigeria, shifts can be seen in the country of data-publishing organizations following independence from Great Britain in 1960 (5). See https://github.com/milliechapman/humanDim-gbif for data, code, and further information about each panel. 1

100

101

Number of observations 102 103 104

2 105

106

0.30 0.25 0.20

p

Cumulative observations per hectare

0.35

High income Upper middle income Lower middle income Low income

0.15 0.10 0.05

g

0.00 1960

1980

2000

2020

y

Battle-related deaths in Cambodia 1000 Cambodian civil war

104

Khmer Rouge regime

Cambodian-Vietnamese war

103 10

2

101

1950

1960

1970

1980

1990

2000

2010

2020

Number of bird observations per hectare

105

United States Urban areas

4 20

Not redlined

5

Country of data publisher For observations recorded in Nigeria Nigeria

Great Britain

Other

15

10

Redlined

5

0

Colonial period

Post independence

y g

Number of observations in Cambodia

3

Organizations based in Nigeria published 28% of the country’s observations while Nigeria was under colonial rule. Since independence, they have published 70%.

y

GRAPHIC: M. HERSHER/SCIENCE

,

presence in the latter, misguiding ongoing dialogues about the impact of different land tenure, property rights, and management regimes on biodiversity outcomes. Invasive species might be detected earlier in more intensely surveyed areas, driving investment toward removal and restoration in areas most thoroughly monitored. Without directly addressing and correcting for social and political disparities in data, the conservation community will likely fall into the same traps that other domains do—entrenching the inequities of the past and present in future decision-making through data.  Quantitatively and qualitatively assessing data-derived decision biases SCIENCE science.org

and the typologies of their impacts on people and communities is an important first step to effectively mitigating the potential negative impacts of these disparities. MORE BIODIVERSITY DATA AND BETTER MODELS MIGHT NOT SOLVE PROBLEMS The past decade has marked a shift away from labor- and resource-intensive specimen collection and field surveys and toward a new generation of decentralized monitoring tools. Participatory science programs, artificial intelligence–supported sensors, and eDNA promise to substantially increase the number of records per research dollar and person-hour. More automated digitiza-

tion of natural history collections around the world is increasing the capacity to understand long-term changes in ecosystems. These technologies and their resultant data streams will undoubtedly provide critical information to fill gaps in our knowledge about global biodiversity and inform more robust global policy strategies. But as biodiversity data become easier and cheaper to collect, will sampling become widespread enough that biases are an artifact of the past, buried under the massive amounts of new information? Although new monitoring technologies continue collecting information about global biodiversity and its degradation at finer 5 JANUARY 2024 • VOL 383 ISSUE 6678

35

I NS I GHTS | P O L I C Y F O RU M

,

R E F E R E N C ES A N D N OT ES

y

DATA AS SOCIAL INFRASTRUCTURE: BIODIVERSITY MONITORING FOR EQUITABLE DECISIONS The realization that more data or better models will never fully solve systemic bias does not mean there are no solutions. It means there are no shortcuts—no getting around the need for local engagement, context-specific knowledge, and case-by-case considerations when using this data. Investments in future monitoring should not only prioritize new technologies that ease the collection of massive amounts of biodiversity data, but also ensure that those data include information about the local context and social infrastructures. Moving beyond quick technical fixes will require connecting strategically to

y g

ing of those systems than global synthesis data can provide. Further, some human drivers of observational (and ecological) processes are not digitizable or easily reduced into quantitative metrics. Although it might be possible to investigate the impact of past residential segregation policies in the United States because there is geospatial information on its history, dimensions that cannot be reduced to polygons on a map, such as human preferences, scientific funding patterns, and industrial priorities, may continue to be reflected in downstream data products and decisions. “Datafication” can thus create another layer of bias: between the social, political, and cultural dimensions of data that are easy to digitize and those that are not (14). 

y

5 JANUARY 2024 • VOL 383 ISSUE 6678

“Moving beyond quick technical fixes will require connecting strategically to community-based partners…”

community-based partners and leveraging expertise in social, ethical, cultural, and political processes underlying data infrastructures and their histories. Communitybased monitoring and information systems (CBMIS) provide a compelling framework for locally engaged monitoring and are highlighted in the GBF as one means of filling data gaps (2). Established networks of CBMIS are already operational in several countries and have proven effective at contributing to national and global-scale monitoring of ecosystems (15). Initiatives such as the International Forestry Resources and Institutions (IFRI) collect information on institutional and social variables, alongside ecosystem data, through a network of locally led Collaborating Research Centers to understand the interrelationships among social and ecological processes and outcomes in forest systems around the world. There is no technocratic solution for incorporating all relevant information about ecosystems and their social contexts into formal frameworks for assessing biodiversity or devising policy strategies at global scales. However, complementing global frameworks and synthesis databases with decentralized knowledge collected as part of CBMIS (and programs like IFRI) might help expose and ameliorate data disparities that underpin biodiversity monitoring and mitigate the implications of these disparities on the distributional equity of downstream conservation decision-making. The success of the GBF, and the meaningfulness of its proposed indicators, requires that policy-makers and scientists resist technocratic shortcuts and instead assess the equity implications of data disparities, support local knowledge generation, and work toward governance systems and monitoring frameworks that engage with biodiversity data as social infrastructure. j

g

36

izing unintended sociopolitical patterns in data is an important step toward developing analytical methodologies that more accurately reflect true biodiversity patterns. Although careful statistical models can help identify and control for data disparities that can be quantified, they are not a panacea. Quantitatively correcting socially determined bias across spatial and temporal scales from the top down would require a near-complete census of these multiscale and interacting biases—an infeasible trap. Even when such quantification reveals statistically clear associations, conducting inference on the multidirectional and interacting causal mechanisms that link social infrastructure, monitoring, and biodiversity is impossible without a deeper understand-

p

resolutions and with a broader scope, this increasing amount of information has yet to yield more representative data coverage of biodiversity distributions. Instead, new waves of biodiversity data have entrenched the long-known overrepresentation of certain regions, taxa, and time periods in global biodiversity data repositories (12). Regardless of the volume or velocity of data collection, where, when, how, and by whom species are observed will always be shaped by social, political, and economic processes (13). Collecting perfectly uniform global biodiversity data isn’t the only possible solution for addressing the gaps and disparities in existing data. Ecologists and statisticians have worked extensively on methods for bias correction of existing biodiversity data to infer how species distributions and populations vary in time and space despite imperfect data (4). Nonrandom sampling effort can be addressed in two ways: One is by assuming that unobserved variation in sampling (e.g., geopolitical conflict– associated differences in sampling effort) is not confounded with the natural process of interest (e.g., biodiversity distributions and their change); another is by “correcting” the bias in the process of interest with data preprocessing or model-based inference. In the case of social drivers of biodiversity sampling at continental and global scales, neither of these technical fixes is likely adequate to remove biases. The first option—assuming that the sampling process in question is not related to the natural process of interest—is baseless in most cases. As outlined above, the drivers of data collection are often deeply intertwined with the natural processes that scientists often seek to assess. The second option—correcting for biases—is only as effective as the capacity of quantifiable variables to explain the biases in the data. In ecology, bias correction tends to focus on bioclimatic conditions, latitudinal disparities, and simple accessibility metrics (e.g., population density, proximity to roads) (4), meaning that the other social infrastructures underlying these data likely remain reflected in ecological insights (e.g., species distribution models, metrics of community change) and downstream decisions (e.g., conservation priorities). Archiving and digitizing human societies’ darkest hours—from war to colonialism to systemic racism—may allow researchers to start to disentangle the past, present, and future signatures of humans on both biodiversity and the data capturing its distribution and change. Character-

1. A. Gonzalez, M. C. Londoño, Science 378, 1147 (2022). 2. Convention on Biological Diversity, Kunming-Montreal Global Biodiversity Framework (2022). 3. A. C. Hughes et al., Ecography 44, 1259 (2021). 4. D. I. Warton, I. W. Renner, D. Ramp, PLOS ONE 8, e79168 (2013). 5. A. Zizka et al., J. Biogeogr. 48, 2715 (2021). 6. D. Ellis-Soto, M. Chapman, D. H. Locke, Nat. Hum. Behav. 7, 1869 (2023). 7. B. Lenzner et al., Nat. Ecol. Evol. 6, 1723 (2022). 8. K. M. Gaynor et al., Front. Ecol. Environ. 14, 533 (2016). 9. C. J. Schell et al., Science 369, eaay4497 (2020). 10. C. Scoville, Theory Soc. 48, 1 (2019). 11. S. T. Garnett et al., Nat. Sustain. 1, 369 (2018). 12. B. H. Daru, J. Rodriguez, Nat. Ecol. Evol. 7, 816 (2023). 13. J. E. Goldstein, H. O. Faxon, Environ. Plan. E Nat. Space 5, 39 (2022). 14. M. Chapman et al., One Earth 4, 790 (2021). 15. M. Ferrari et al., Biodiversity (Nepean) 2, 57 (2015). 10.1126/science. adh8874

science.org SCIENCE

B O OKS et al .

CLIMATE

By Erle C. Ellis

AC KN OW LED G M E N TS

The reviewer is a visiting fellow at the Oxford Martin School, which provides funding for Our World in Data. 10.1126/science.adl5466 5 JANUARY 2023 • VOL 383 ISSUE 6678

37

,

1. Our World in Data; https://ourworldindata.org/.

y

SCIENCE science.org

RE FE REN C ES AN D N OT ES

y g

PHOTO: KATRIN SAUERWEIN/ISTOCK PHOTO

I

y

falsely, as a surprising number of educated people do—that more people are starving, n a time of daily environmental crisis suffering, living in poverty, being exposed to headlines and growing eco-anxiety, how natural hazards, and dying young than ever could anyone claim that a much better before. The turning point for her came in world is not only possible but that—in the form of the late Swedish physician Hans many ways—it already exists? The anRosling’s masterful, data-rich presentations swer, according to data scientist Hannah showing that the human world is almost uniRitchie in her first book, Not the End of the versally changing for the better. Indeed, in World, is that this is exactly what the data tell many ways, Ritchie is building on Rosling’s us. As the science outreach lead legacy, although she goes much of the widely cited project Our further: Where Rosling focused World in Data, she is certainly one almost entirely on the successes to know (1). and challenges of human develNot the End of the World is built opment, Ritchie takes on “the on the same solid, data-driven other side” of the sustainable fufoundations that characterize the ture—the environmental damage crisp, concise visuals produced by that looms like a dark cloud over Our World in Data, but it is a book everything achieved in improving Not the End with a purpose beyond presenting the human condition. of the World: facts about our planet: Ritchie’s While the human world has How We Can Be goal is to convince young people generally been getting better in the First Generation to shake off climate doom and get critical ways for decades, the opto Build a Sustainable Planet motivated to address the unprecposite has largely been true for the Hannah Ritchie edented environmental challenges rest of the planet. Yet, as Ritchie Little, Brown Spark, 2024. of our times. This mission is pershows in data-rich chapters 352 pp. sonal, and she wraps the narrative packed with illustrative charts, of the book around her journey from econot all of the trends are going in the wrong anxiety to “urgent optimism.” direction, and we already have most, if not Convinced that she “didn’t have a fuall, of the capabilities needed to address seture left to live for,” Ritchie long believed— rious environmental challenges such as air pollution, climate change, deforestation, inThe reviewer is at the Department of Geography and sufficient food supply, biodiversity loss, ocean Environmental Systems, University of Maryland, Baltimore plastics, and overfishing. County, Baltimore, MD 21250, USA, and is a visiting fellow For each challenge, she describes how at the Oxford Martin School, University of Oxford, Oxford it was first identified and understood by OX1 3BD, UK. Email: [email protected]

g

A data scientist offers an optimistic reality check for the Anthropocene

scientists and how it has been addressed to date. Most importantly, she highlights and prioritizes a suite of specific solutions, revealing how many seemingly intractable threats to people and planet are, in fact, actionable and resolvable through strategies available right now. However, she is careful to show that even with solutions in hand, many of the challenges we face are not being dealt with adequately and that, in many cases, the damage is accelerating. Ritchie also debunks two strategies that, by her assessment, “won’t fix our problems”: depopulation and degrowth. She argues that the former is unnecessary—“peak child,” she writes, has already happened—and more rapid shrinkage is unachievable without subjecting people to unconscionable harm and that the latter would leave most people on Earth unacceptably impoverished. True believers in such propositions will likely remain unswayed, facts or no facts. This brings us to the book’s greatest provocation—and subtitle—about how our generation can be the first to build a sustainable planet. Recognizing the successes and future potential of contemporary societies is certainly an important accomplishment of the book. Nevertheless, making such a strong and general claim about the superiority of today’s societies over all that came before seems premature, problematic, and unnecessary and exposes the absence of a deeper and more nuanced theoretical understanding of human– nature relations in the book. Fortunately, this and other oversimplifications do not detract from the main aims of the work, which are overwhelmingly pragmatic. That this book represents “a synthesis of nearly a decade of research and data” is clear from its remarkable coverage of environmental evidence. Yet its main contribution may be its ability to produce what Ritchie calls a “perspective that helped me dig myself out of a very dark place.” There is real peril in our widespread failure to understand just how much human lives have been improved through societal efforts and in the level of doom and helplessness that seems to be spreading among young people. As Ritchie demonstrates in Not the End of the World, a better future for both people and planet is possible and even achievable. j

p

Data-driven hope for the planet

Excessive plastic food packaging is not ideal, argues Ritchie, but a little can help prevent food waste.

I N SI G H TS | B O O K S

SUSTAINABILITY

Water recycling goes mainstream

Purified: How Recycled Sewage Is Transforming Our Water Peter Annin Island Press, 2023. 248 pp.

Public health and public relations are key to successful potable water reuse programs By Sasha Harris-Lovett

y g y ,

science.org SCIENCE

y

5 JANUARY 2023 • VOL 383 ISSUE 6678

g

38

p

contaminants (Orange County), do extra monitoring in the sewer network (El Paso), solicit additional surveillance from the otable water reuse, also known as wapublic health department (Virginia), and ter recycling or by the moniker “toilet require specific training for potable water to tap,” is the practice of deliberately reuse treatment plant operators (El Paso). incorporating treated wastewater But regulations vary widely in ways that into the drinking water supply. In Puhave the potential to compromise public rified, journalist Peter Annin’s vivid health and sour citizens on water reuse projand engaging new book, readers join Annin ects. When Annin interviews potable water on a tour of potable water reuse systems reuse system engineers in Deep Spring, across the United States, from California to Texas, for example, he learns that the city Texas to Florida. Annin’s meticulous reportdoes not have a monitoring system to deing reveals the technologies and individutect toxins, such as acetone, that can slip als responsible for this new water supply, untreated from sewage into drinking water as well as the histories of the communities through their reuse system. In much of the that are—or soon will be—drinking it. book, Annin seems to accept at Annin sets the stage with a face value the claims about water portrait of water scarcity along reuse technology made by the wathe Colorado River, which has ter engineers he interviews, but experienced record low water levI gave a silent cheer when he diels. Water levels in Lake Mead, we rectly challenges one Deep Spring learn, are nearing “dead pool”— water engineer, asking “if it might a state in which a reservoir conbe worthwhile for Big Spring watains so little water that it can no ter officials to consider adding longer flow downstream. Without real-time monitoring for acetone intervention, this will spell disasto their history-making plant?” ter for the millions of people in (The engineer cryptically replies, the Southwest who depend on it. “It may be.”) “Enter purified sewage,” Annin Annin contends that water quips. “Thanks to climate change, utilities, states, and the federal never before has something so government must ensure that we foul looked so good.” This is cerhave the right regulations in place tainly the case in parts of Texas. to guarantee that potable water Annin details how extreme reuse projects maintain public drought pushed cities such as Big Spring, Wichita Falls, and El Paso health. Such policies, he asserts, to turn to potable water reuse should include protecting sewPeople taste recycled wastewater in Segundo, California, in 2015. when their reservoirs had almost ers from chemical spills; creatrun dry. But drought is not the only driver challenges. When the project’s engineers ing real-time monitoring infrastructure for for potable water reuse. In Florida, populadetected that their wastewater purification early detection of problematic compounds; tion growth is stressing freshwater supplies technologies were creating a potent carand developing industry-standard regulato a breaking point, leading water engineers cinogen, N-nitrosodimethylamine (NDMA), tions governing treatment technologies, to consider urban sewage as a source of and adding it into the water supply, their operator training, and emergency response drinking water. public relations team explained that the plans. As potable water reuse comes into Even when there are good reasons to risk of ingesting NDMA in water was less the mainstream, Annin warns that “there is pursue potable water reuse, people are not than the risk of eating hot dogs (which also simply no margin for error.” always excited about it. In the 1990s, San contain NDMA) and described their techniAs fresh water supplies become increasDiego citizens roundly rejected a proposed cal approach to ameliorating the problem. ingly scarce and technologies for water potable water reuse project, even though This transparent communication paid off, treatment improve, potable water reuse will the city had already invested more than and “no backlash occurred.” likely become widespread. Annin’s timely $15 million in it. Through interviews and Annin’s case studies reveal how utilities and important reporting empowers readanalysis of news articles, Annin forensithat have successfully implemented potable ers to understand the critical issues at hand water reuse go well beyond the regulatory and offers an engaging introduction to poThe reviewer is at the San Francisco Estuary Partnership, requirements to ensure that the water they table water reuse. j San Francisco, CA 94105, USA. produce is safe. They do extra testing for 10.1126/science.adl2392 Email: [email protected]

P

cally investigates why that project failed. The engineers were clear: The technologies used to treat sewage to drinking water standards were robust and would protect public health. The problem, Annin concludes, is that potable water reuse proponents in San Diego did not have a strong enough communications and outreach strategy. The importance of proactive outreach and transparent communication to support potable water reuse projects is a theme that recurs throughout the book. In Orange County, for example, Annin reveals how the Groundwater Replenishment System’s “sweeping public relations offensive” proved critical to its ability to weather

INSIGHTS LET LE T TTERS ERS

p g y

NEXTGEN VOICES

Changing outdated expectations

y g

In the 3 November 2023 Working Life (https://scim.ag/WL_locationmatters), Shanta Hejmadi questions the expectation that scientists should change institutions between their PhD and postdoc. We asked young scientists: What traditional career expectation have you come across that seems outdated, counterproductive, or exclusionary? How should that expectation be adjusted? Read a selection of the responses here. Follow NextGen Voices on social media with hashtag #NextGenSci. —Jennifer Sills

y

apply their skills in the innovation divisions of large companies, at industry-led think tanks, or as teaching faculty, and may later choose another path or return to research. These experiences deepen understanding and generate new ideas. Academia can improve inclusivity by encouraging innovation, reducing stigma around “alternative” careers, and empowering individuals to make career decisions that suit them. Providing mentorship programs, career workshops, and fellowships and grants for diverse career stages and trajectories could facilitate this flexibility. Dequn Teng

Everyone who completes a PhD and postdoc is expected to continue in academia. This robs scientists of the flexibility to pursue less conventional career paths such as funding, science writing, science ethics, science journalism, or management. Information about these options is not readily available, and in some cases, there is a stigma attached to changing careers. Choosing a career outside of academia should be easy and accessible at any point in the educational journey, and people should be able to change course without feeling like they have failed. Career counseling should always be on offer, and

Garima Singh

24

5 JANUARY 2024 • VOL 383 ISSUE 6678

Fleming Fund South Asia, Mott MacDonald, Noida, Uttar Pradesh, India. Email: [email protected]

Scientists are expected to advance sequentially from PhD to postdoc to tenure-track faculty. This expectation benefits prestigious academic institutions, which can choose from many applicants, but marginalizes people who follow unconventional trajectories, such as transitions between academia, industry, and teaching. Hiring teams in academia and beyond should embrace diverse, nonlinear career paths. New PhDs may

Department of Engineering, Institute for Manufacturing, University of Cambridge, Cambridge, UK. Email: [email protected]

A PhD is expected to have skills in grant writing, teaching, industry, and entrepreneurship in addition to their subject area. Universities benefit from this assumption because it justifies the lack of help available to specialists trying to accomplish tasks outside of their area of expertise. Expecting postdocs and faculty to immerse themselves in tangential tasks slows down research progress. Acknowledging that a PhD indicates scientific but not administrative, pedagogical, or business expertise would free scientific experts to focus on research science.org SCIENCE

,

Support realistic career paths

programs and scholarships at the university level should help support career changes.

and justify hiring a specialist workforce to conduct other tasks. Houcemeddine Turki Data Engineering and Semantics Research Unit, University of Sfax, Sfax, Tunisia. Email: [email protected]

Assess experience equitably

Mayank Chugh Department of Systems Biology, Harvard Medical School, Boston, MA, USA. Email: [email protected]

In China, students are expected to complete their PhDs by age 35, after which they are no longer given preference for positions at universities. The age limit aims to encourage younger candidates to apply for positions. However, this requirement is unfair to those who start their PhD late. When applying to university positions, PhD graduates should be judged by their achievement rather than age. Awards, published papers, and patents are more important and relevant indicators of ability than age at the time of PhD completion.

exclusive commitment limits faculty’s ability to provide job market insights to students and lowers income potential compared with private sector roles. Without exclusive commitment requirements, professors could help students better navigate the job market, set up science-related businesses, and engage in other complementary activities that would improve undergraduate teaching. Paulo Bezerra Department of Engineering and Technology, Universidade Federal Rural do Semi-Árido, Pau dos Ferros, RN, Brazil. Email: [email protected]

Doctors are expected to work long hours to benefit patients and medical institutions, but the pressure leads to physical and mental exhaustion. If institutions allowed doctors to work reasonable schedules and provide appropriate opportunities for vacation and self-care, doctors would be able to better manage and balance their work and personal lives, thereby enhancing their work efficiency and quality.

Ju Wen

LiangBin Wu

Towards AI, Montreal, QC, Canada. Email: [email protected]

School of Liberal Education, Chengdu Jincheng College, Chengdu, Sichuan, China. Email: [email protected]

Peking University Shenzhen Hospital, Shenzhen City, Guangdong, China. Email: [email protected]

Standardize fair contracts

Incentivize quality research

Marine scientists who are just starting their career are expected to accept unpaid work opportunities. Although the positions often require graduate-level education, numerous certifications (such as SCUBA), and previous experience, participants are rarely given the option of advancing to a paid role. This framework is exclusionary to those without the financial means to support themselves for weeks or months to gain what organizations tout as valuable experience. Instead, marine conservation organizations should adopt ethical programs that compensate participants for their time and skills. Budget allocations for stipends would be ideal, but an alternative could be the creation of partnered internships with structured learning objectives and benefits such as credits.

The “publish or perish” adage in academia pressures postdoctoral researchers to prioritize quantity of publications over quality. This paradigm, deeply ingrained in academic culture, benefits those who produce a high volume of work, potentially at the expense of thoroughness or innovation. The expectation not only stifles creativity and patience but also discourages investment in high-risk and time-consuming but highreward and high-impact research. Academia should pivot to valuing the quality and impact of research rather than the number of publications. This would cultivate an environment that encourages innovative and transformative science. Institutions can promote this shift by redefining metrics of success to include broader impacts such as community engagement, teaching, and policy influence, ensuring a holistic evaluation of medical researchers’ contributions.

SCIENCE science.org

Emily Yi-Shyuan Chen Department of Marine Ecology, Institute of Oceanology, Polish Academy of Sciences, Sopot, Pomerania, Poland. Email: [email protected]

Jingtao Huang

In Brazil, many full-time faculty at public universities are expected to agree to exclusive commitment contracts, which prohibit external employment. For example, as a civil engineering professor, I’m barred from authoring building designs. These contracts are intended to reduce conflicts of interest, ensure institutional loyalty, and provide quality in teaching and research. However,

Shantou University Medical College, Shantou, Guangdong Province, China. Email: [email protected]

Scientific researchers are expected to produce novel results. This focus, intended to drive innovation, inadvertently fosters an unhealthy culture that deters essential replication studies. Excessively prioritizing originality, and neglecting or even 5 JANUARY 2024 • VOL 383 ISSUE 6678

25

,

Scientists are expected to train abroad to build a competitive career. Working in foreign countries undoubtedly provides better international networks and collaborations; exposure to new methods, skills, funding, and publishing landscapes; and a diversity of ideas, people, and experiences. However, expecting mobility is based on a false assumption that global access is equal to all. Some passports (mostly in the Global North) provide more opportunities than others. The unwritten requirement of working abroad, especially for postdocs, particularly harms and excludes scholars from the Global South. To provide equitable access for everyone, institutions should reform

y

Shantou University Medical College, Shantou, Guangdong, China. Email: [email protected]

y g

Yan Chen

y

In the field of artificial intelligence, there is an expectation that applicants with the most prestigious institutional pedigrees will have the most impressive skills. However, this assumption overlooks individuals who have gained valuable practical experience, pursued alternative educational paths, or come from underprivileged backgrounds without access to top-tier education. To remedy this inequity, institutions should revise hiring criteria to focus on practical experience, demonstrated expertise, and portfolio quality rather than prioritizing prestigious degrees.

g

Louis-François Bouchard

p

Artificial intelligence researchers are expected to have PhDs. However, innovation in artificial intelligence often occurs in industry settings, and industry experience can rival academic credentials in driving artificial intelligence advancements, thus rendering the lengthy PhD process unnecessary. Therefore, industry experience should be considered equivalent to a PhD. This change would empower professionals to contribute without committing to the extended time and narrow focus of a doctoral program. For those who do pursue a PhD, thesis graduation criteria should include substantial industry achievements as an alternative to academic certifications.

evaluation practices and metrics for early career researchers and faculty during hiring, tenure, promotions, funding, and potential awards to prioritize a candidate’s research, skills, background, and lived experiences over their ability to secure international opportunities.

I NS I GHTS | L E T T E R S

discouraging replication efforts, undermines the goal of establishing confirmed and reliable knowledge. The scientific community should acknowledge the value of replication studies. To facilitate this shift, funding bodies and academic journals need to adjust their policies and reward systems to recognize and support replication efforts. Qianjun Wen The Affiliated Hospital, Guizhou Medical University, Guiyang City, Guizhou, China. Email: [email protected]

Respect community ties

Many PIs spent their time as graduate students and postdocs battling harsh (even abusive) mentors in isolated, unsupportive environments. They survived in academia by accepting the circumstances and succeeding despite obstacles, and they rightly wear their struggle as a badge of honor. However, perpetuating the expectation that succeeding in academia requires struggling limits the pool of future capable scientists. Instead, institutions should strive to support each scientist as the multifaceted individual they are. Academic departments should normalize mentor-mentee discussions about work styles, expectations, career plans, and project management; provide training for careers outside of academia; proactively address the needs of scientists with disabilities, chronic illnesses, and mental health challenges; and allow for mid-career pauses. To see diversity and equity in science, academia needs to be a place where excellent scientists are encouraged, rather than challenged. Madeline Klinger Helen Wills Neuroscience Institute, University of California, Berkeley, Berkeley, CA, USA. Email: [email protected]

Scientists are expected to seek individual achievement. Success is tied to earning grants, authorships, and positions of power such as PI. Overemphasizing these individualistic practices can stifle collaboration and exclude or marginalize people whose cultures and communities prioritize collective pursuits. I grew tired of the individualistic environment I experienced while obtaining a PhD, partly because it was a poor match to my values as a Latina woman. Providing more collective structures and practices, such as mentoring the next generation and encouraging individuals to help team members, could prevent members of underrepresented groups from leaving science. Emma Claudia Perez Pasadena, TX, USA. Email: [email protected]

Encourage interdisciplinary work Academic research conducted by business students is expected to provide insight into managerial challenges. Business schools should expand their research objectives to address global operational challenges such as climate change and food insecurity. Moving beyond the topic of management will increase theoretical and interdisciplinary research and facilitate innovative approaches that can tackle global problems. Samuel Nathan Kirshner University of New South Wales Business School, University of New South Wales, Sydney, NSW, Australia. Email: [email protected]

Applicants for lab positions are expected to pursue projects closely aligned with the group’s current research. Institutions benefit from this strategy because new lab members can integrate quickly and make immediate contributions. However, basing hiring on narrow interests and goals hampers the advancement of interdisciplinary studies. Instead, institutions should recruit researchers with diverse disciplinary backgrounds and interdisciplinary aspirations. Diverse disciplinary teams and broad perspectives are vital in emerging fields. Chao Xu School of Resource and Environmental Sciences, Wuhan University, Wuhan, Hubei, China. Email: [email protected]

Immunologists are expected to focus exclusively on their own discipline, but an interdisciplinary approach could be more effective. For example, to decipher the pathological changes responsible for autoimmune bone marrow failure, focusing only on immunological aspects (lymphocytes) might not lead to a comprehensive understanding. Instead, immunologists need to collaborate and look beyond immunology for the putative agent. The fields of genetics, molecular genomics, and environmental and health research could hold the answers. Vandana Sharma

Scientists are expected to devote all of their energy to their research to succeed, and those raising children as they launch their career are assumed to be less productive.

Department of Hematology, All India Institute of Medical Sciences, New Delhi, India. Email: [email protected] 10.1126/science.adn4211

science.org SCIENCE

,

5 JANUARY 2024 • VOL 383 ISSUE 6678

Create supportive environments

y

26

Climate-Energy Lab, Carnegie Institution for Science, Stanford, CA, USA. Email: [email protected]

y g

As Shanta Hejmadi explains in her Working Life, scientists are expected to change institutions between a PhD and postdoc, which usually requires a long-distance move. The goal is to expose scientists to different experiences, which can later inform their work as faculty. I was so passionate about becoming faculty that I moved from Wisconsin to Texas for a postdoc. My husband quit his job to move with me. We missed our families and the stronger support system we had in our home state. Unlike Hejmadi’s “happy ending,” the sacrifices we made ultimately cost me my academic career. Instead of a long-distance move, postdocs should be given the option of working in a different department, with a different faculty mentor, at their PhD institution. Faculty who stay at the same institution for a PhD and postdoc could be an asset given their long-standing knowledge of the institution’s history and strong ties to the community. Not penalizing

Edgar Virguez

Boston Consulting Group, Minneapolis, MN, USA. Email: [email protected]

y

Department of Aerospace Engineering, University of Maryland, College Park, MD, USA. Email: [email protected]

Bridget Stroup

g

Christine Hartzell

Early-career scientists can be productive researchers and responsible parents if they can access supportive policies, such as parental leave, flexible schedules, familysustaining wages, and advice on managing time and setting boundaries.

p

Faculty are expected to take a sabbatical at an institution in another state or country about every 6 years. In theory, these experiences give faculty time to develop new ideas and build collaborations with new people. However, this expectation is unrealistic for many families, including those with dual incomes (if the spouse can’t work remotely for months), health problems, children with special needs, or local family obligations. Institutions could address these challenges by providing prestigious external funding opportunities for multiple 1-week-long institutional visits during a sabbatical instead of one months-long visit. I am fortunate to live in a region with many nearby institutions, so I can take a sabbatical without relocating my family. However, any travel to institutions in other states or countries currently would be funded by my regular research grants, which would not be as prestigious as a named sabbatical fellowship.

postdocs for staying at the same institution as their PhD would help postdocs with spouses and families pursue their academic dreams. This solution would especially benefit women and could even help increase the rate of women in academic leadership.

RESEARCH

Photoluminescent inorganic compounds embedded in a polymer can be used to create dual-color three-dimensional structures.

IN S CIENCE JOU R NA L S Edited by Michael Funk

p g

PEROVSKITES

y

A supramolecular boost to emission

M

any inorganic perovskites can show strong photoluminescence, including ones based on hafnium and zirconium halides, but these compounds are often air and water sensitive. Zhu et al. extended a crown ether–assisted supramolecular assembly approach to guide and stabilize the assembly of octahedral [HfBr6]2– octahedral centers. Solid

SCIENCE science.org

Science p. 62, 10.1126/science.adh8342

ORGANIC CHEMISTRY

Science p. 77, 10.1126/science.adk2312

Lighting the way to amides

2D MATERIALS

Cobalt carbonyl complexes catalyze the coupling of alkenes, hydrogen, and carbon monoxide to produce aldehydes through

Two-dimensional (2D) van der Waals materials that exhibit a moiré effect typically have

Mapping out moirés

complicated electronic band structures. To study them, researchers can apply magnetic fields and look for oscillations in transport properties. Thermodynamic properties such as magnetization should also exhibit quantum oscillations, but these are tricky to detect in the very small 2D material samples typically available. Bocarsly et al. used a very sensitive magnetometer— a superconducting quantum interference device—placed on the apex of a sharp pipette to scan a sample of moiré graphene. The authors detected patterns of magnetization oscillations and used the data to reconstruct an intricate band structure. —JSt Science p. 42, 10.1126/science.adh3499

5 JANUARY 2024 • VOL 383 ISSUE 6678

39

,

Cancer immunotherapy with immune checkpoint inhibitors blocks negative signals of T cell activation to mount an immune response against tumors but can also lead to immunopathologies. Colitis is a frequent and severe adverse event in patients treated with antibodies targeting the checkpoint inhibitor cytotoxic T-lymphocyte protein 4 (CTLA-4), but the underlying mechanisms leading to this reaction remain unclear. Lo et al. demonstrate that CTLA-4 blockade–induced colitis in mice is dependent on gut microbiota composition and is driven by the unrestrained activation of T cells and the

hydroformylation. Faculak et al. report that by adding light to the mix, they can swap out the hydrogen for primary or secondary amines and produce amides with the same catalyst. Adding a silane afterward can reduce these products to more highly substituted amines if desired. The authors propose that the photoexcitation expels one carbon monoxide ligand from the catalyst, opening a critical coordination site under mild conditions. —JSY

y

PHOTO: ZHU ET AL.

Microbiota-reactive T cells trigger colitis

concurrent depletion of a subset of regulatory T cells in the gut by receptors recognizing the Fc domain of the anti-CTLA-4 antibodies. Anti-CTLA-4 nanobodies lacking the Fc domain were found to stimulate antitumor immunity without inducing colitis. These findings may support the development of next-generation CTLA-4 inhibitors with reduced inflammatory toxicities. —STS

Science p. 86, 10.1126/science.adi4196

y g

IMMUNOLOGY

powders of (18C6@K)2HfBr6 exhibited 96% photoluminescence quantum yield for blue emission, and a zirconium analog had a photoluminescence quantum yield of 83% for green emission. Addition of a polymer created solution-processable inks for printing luminescent thin films and structures. —PDS

R ES E ARCH

ALSO IN SCIENCE JOURNALS NEUROSCIENCE

Brain circuits handle risky decisions

The origins of trait innovation Traits are constantly changing in populations, but the emergence of a trait that provides a new function is a rarer occurrence. Two studies examined how such traits arise. Stankowski et al. identified the multiple genome regions associated with the transition from egg-laying to live-bearing reproduction in a clade of snails, demonstrating a polygenic basis of live-bearing reproduction with alleles that accumulated in response to selection over time. Chomicki et al. found that the spontaneous occurrence of three variable traits created a novel mechanism for trapping insects in pitcher plants, which arose separately in two different species. Together, these studies provide insight into the emergence of complex, transformative traits (see the Perspective by Elmer). —BEL and CNS Science p. 114, 10.1126/science.adi2982, p. 108, 10.1126/science.ade0529; see also p. 27, 10.1126/science.adm9239

Science p. 94, 10.1126/science.adj1962

IMMUNOLOGY

Double pi

The TAO(K3) of T cell signaling

Carbenes contain a carbon atom in which only two of the four valence electrons are bonded to other atoms. In general, at least one of the two remaining electrons resides in the same plane as those bonds. Hu et al. report an unusual carbene with flanking phosphorus centers in which both of the nonbonding electrons occupy the π orbital perpendicular to the bonding plane. The carbene, isolated and characterized in a complex with each phosphorus coordinated to

T cell receptor signaling is inhibited by the phosphatase SHP-1, which dephosphorylates and inactivates several effectors downstream of the receptor. Poirier et al. found that the kinase TAOK3 promotes T cell receptor signaling by phosphorylating SHP-1, leading to the ubiquitylation and degradation of SHP-1. The loss of TAOK3 reduced both tonic and liganddependent T cell receptor signaling in mouse CD4+ T cells and in a human T cell line. Thus,

CARBENE CHEMISTRY

5 JANUARY 2024 • VOL 383 ISSUE 6678

,

41-B

Metal nanoparticles on oxide surfaces used as catalysts for industrial reactions tend to become larger and less active after prolonged exposure to reaction conditions. Liu et al. have shown that after exposure to methanol vapor at 200ºC, copper nanoparticles in dealuminated beta zeolite become smaller. The average size decreased from ~5.6 to ~2.4 nanometers because copper migrated and became trapped at new surface sites. This catalyst maintained nearly 100% conversion and selectivity for ambient-pressure hydrogenation of ethylene glycol to dimethyl oxalate at 400ºC for 200 hours. —PDS

EVOLUTION

y

Many block copolymers will phase separate into distinct regions, where the minority component might form spheres, cylinders, or layers inside the majority component depending on the strength of the interactions between the two components and their relative proportions. Some metastable phases are also possible, including the bicontinuous “plumber’s nightmare.” Lee et al. show that it is possible to make this phase a stable one in a diblock copolymer by adjusting the interactions of the chain ends (see the Perspective by

Maintaining nanoparticles with methanol

Sci. Signal. (2024) 10.1126/scisignal.adg4422

y g

The chemistry of the chain ends matters

CATALYSIS

Science p. 81, 10.1126/science.adk6533

TAOK3 sets the threshold for T cell activation by controlling SHP-1 abundance. —AMV

y

POLYMERS

Science p. 70, 10.1126/science.adh0483; see also p. 28, 10.1126/science.adn0168

rhodium, reacted as a Lewis acid in the plane but as a Lewis base above or below it. —JSY

g

Science p. 55, 10.1126/science.adj6645; see also p. 32, 10.1126/science.adm8641

Nedoma). The strength of endgroup interactions was shown to modulate the curvature of the adjacent polymer, imposing significant frustration when the adjacent polymer is crystalline. By developing a series of design rules, the authors have found a controlled way to reversibly make complex structures. —MSL

p

Decision-making to select between high risk–high return and low risk–low return choices is critical not only in economic activities such as investing, but also in many daily life situations. However, the underlying neural circuits remain largely elusive. Sasaki et al. identified two subsystems of the brain’s mesofrontal pathways that control risk-dependent decision-making in macaques (see the Perspective by Stuphorn). Stimulation of the pathway from the ventral tegmental area to the ventral aspect of the ventral Brodmann area 6 (area 6V) resulted in stronger preference for high risk–high return choices. By contrast, activation of the circuit from the ventral tegmental area to the dorsal aspect of area 6V led to low risk–low return inclination. Repetitive stimulation of these distinct pathways over time resulted in stimulus-independent long-term enhancement or reduction of preferences. —PRS

Edited by Michael Funk

science.org SCIENCE

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

CANCER

Killing cohesin-mutant MDS

Edited by Caroline Ash and Jesse Smith ELECTROCATALYSIS

Balancing Pd and Pt for epoxidation Propylene oxide is produced at massive scale using dangerous and corrosive oxidants. A potentially safer method uses electrochemistry to oxidize propylene with water, but catalyst stability is a persistent challenge. Chung et al. report that by alloying palladium with platinum, they could generate an electrocatalyst that delivers >60% Faradaic efficiency toward propylene epoxidation by water oxidation. Initially focusing on acetonitrile as a cosolvent, the authors went on to explore purely aqueous media and discuss the prospects of improving propylene throughput in the face of low solubility. —JSY

y

Science p. 49, 10.1126/science.adh4355

Atypical models of arrestin binding

Tissue-resident CD8+ T (TRM) cells are constantly surveilling organs and tissues for the presence of uninvited microbes. Previous studies have shown that TRM cell migration is triggered by chemoattractant and adhesion molecule signaling, which facilitates the rapid detection of infected cells. More recent evidence has indicated that TRM cells within submandibular salivary glands display different motility patterns exclusive of chemosensing. Ruef et al. now show that submandibular salivary gland TRM cells from virally infected mice display spontaneous retrograde F-actin flow as a means of force-generated translocation. Similar patterns of locomotion were detected in TRM cells from other exocrine glands and were dependent on the sensing of changes in mechanical loads

Interactions between G protein– coupled receptors and their downstream signaling partners are often weak, transient, or mediated by short sequences or structural motifs and are thus difficult to study in detail. Maharana et al. determined cryo–electron microscopy structures of the signaling and regulatory protein b-arrestin in complex with full receptors or a phosphopeptide corresponding to the receptor C terminus. Using refinement focused on the arrestin component, the authors visualized the peptidebound conformation. In one of the structures, the protein’s C terminus adopted an a-helical conformation that the authors propose is important for dynamic recognition of different receptors and phosphorylation patterns. —MAF

5 JANUARY 2024 • VOL 383 ISSUE 6678

Science p. 101, 10.1126/science.adj3347

IMMUNODEFICIENCY

Immune imbalance in rare disease Combined immunodeficiencies result from genetic mutations that cause defective immune responses. Lui et al. found that human siblings homozygous for a variant of lymphocyte-specific protein tyrosine kinase (LCK), a protein essential for T cell receptor (TCR) signaling, exhibited an altered T cell compartment, suffered from recurrent infections, and failed to thrive. The authors created cell lines that expressed the mutated LCK protein, which had a single amino acid change, and found that it disrupted TCR signaling. In mice genetically modified to carry the missense mutation, some conventional T cells could still mature, and the development of intestinal inflammation was attributed to

a deficiency in regulatory T cells. —SHR J. Exp. Med. (2024) 10.1084/jem.20230927

FLORAL EVOLUTION

Flower diversity peaked early The emergence of flowers created new reproductive modes, mutualistic interactions with pollinating animals, and fodder for diversification in angiosperms. Flower shape, arrangement, color, and anatomy vary widely between and within some lineages. López-Martínez et al. quantified disparities in flower structure across lineages and over time using morphological data from more than 1000 extant species and more than 100 fossil species and found that floral disparity peaked in the science.org SCIENCE

,

Mechanosensing by T cells

y

SIGNALING COMPLEXES

y g

CELL MIGRATION

40

IN OTHER JOURNALS

g

Sci. Transl. Med. (2024) 10.1126/scitranslmed.ade2774

Sci. Immunol. (2023) 10.1126/sciimmunol.add5724

p

Splicing modulators have been developed to treat cancers with splicing factor mutations, such as myelodysplastic syndromes (MDS). Mutations in the cohesin complex are also very common in MDS, but whether these cohesin-mutant cancers would respond to splicing modulation is unknown. Wheeler et al. show that cohesin-mutant cells are particularly sensitive to splicing modulators, which induce missplicing and down-regulation of DNA damage repair genes, leading to the accumulation of DNA damage and sensitization to talazoparib or chemotherapy in vitro. In patient-derived xenograft mouse models, splicing modulators reduced cohesin-mutant cancer burden, and sequential treatment with talazoparib or doxorubicin and cytarabine significantly improved survival. —MLN

through signals triggered by nuclear deformation. —CNF

PARTICLE PHYSICS

Coming to terms with the standard model

QUANTUM OPTICS

Straining diamond for quantum control

T

y

Phys. Rev. X (2023) 10.1103/PhysRevX.13.041037

g

he optical and electronic properties of color centers in diamonds make them a promising platform for quantum technologies, including enhanced sensing and quantum information processing applications. However, coupling to vibrations in the diamond crystal lattice can induce decoherence and typically requires operation at millikelvin temperatures to mitigate for the phonon-induced performance loss. Guo et al. show that application of uniform strain can enhance the properties of the color center for tin vacancy centers in a diamond membrane. The authors also show that the center can be addressed with microwaves and the operation temperature pushed up to 4 kelvin, conditions reachable with widely available cryogenic systems. This work demonstrates that strain-tuning provides a mechanism for the stable control and coherence protection of color centers in diamond membranes. —ISO

p

The standard model (SM) of particle physics is frequently tested experimentally by measuring some of its predictions. One such prediction is the production rate of the W boson along with a top and antitop quark pair in protonproton collisions. Measurements of this rate by the CMS and ATLAS collaborations at the Large Hadron Collider have yielded values slightly higher than, but still consistent with, the SM prediction. However, the SM prediction itself has an uncertainty associated with the complexities of the calculations used to derive it. Buonocore et al. improved the precision of the SM value, using two different methods to estimate a particularly difficult to calculate term. The two approaches agree with each other, giving confidence in the result and enabling more stringent tests of the SM. —JS Phys. Rev. Lett. (2023) 10.1103/PhysRevLett.131.231901

Strain-tuning helps control the color centers in diamond.

GLOBAL WARMING

Risky known unknowns

Inside a spider’s brain Spiders are known to weave webs to catch flying prey. However, ancestors of spiders might have been silk-lined burrow dwellers that adopted web building during the Jurassic–Cretaceous period, when flowering plants and flying insects emerged. To understand the evolution of their distinctive hunting strategy, Jin et al. looked into the brains of spiders for clues. They took SCIENCE science.org

Nat. Ecol. Evol. (2023) 10.1038/s41559-023-02238-y

VIROLOGY

Bat–virus arms race Possibly because of the physiological demands of being a flying mammal, bats have evolved to trade off fever-inducing immune responses for tolerance to a wide

As the severity of the global warming crisis grows, a large number of potential geoengineering responses have been suggested. However, dangerous unintended consequences of approaches such as the injection of light-reflecting particles (sulfate, for example) into the stratosphere still are not understood. Vattioni et al. investigated how stratospheric injections of alumina particles might perturb atmospheric composition, finding that the effects on heterogeneous chemistry could result in the destruction of more than twice as much stratospheric ozone as that eroded by chlorofluorocarbons in the late 1990s. Once again, it is clear that there is no silver bullet for curbing global warming other than reducing the carbon dioxide content of the atmosphere. —HJS

PLOS Biol. (2023), 10.1371/journal.pbio.3002398

Geophys. Res. Lett. (2023) 10.1029/2023GL105889

5 JANUARY 2024 • VOL 383 ISSUE 6678

41

,

EVOLUTION

range of viruses. Bat viruses, however, rarely cause overt disease in their hosts. Lytras et al. have looked at the molecular evolution of bat tolerance and found that in horseshoe bats, a prenylated residue is missing on a potent mammalian antiviral protein called 2′-5′-oligoadenylate synthetase 1 (OAS1). The authors reconstructed in vitro the most likely sequence of an ancestral bat OAS1 and found that it blocks severe acute respiratory syndrome coronavirus 2 replication in A549-ACE2-TMPRSS2 cells. Protein structure predictions showed that the sites most under selection in OAS1 lie near the RNA-binding region at the C terminus where the prenylation signal should be. What the evolved OAS1 does for modern horseshoe bats is a mystery. —CA

y

New Phytol. (2023) 10.1111/nph.19389

a high-throughput single-cell transcriptomic approach and built a comprehensive atlas of cell types for the adult spider brain. Aerial web building and spatial niches expose spiders to more environmental challenges and risks. The authors found that genes that have been positively selected in the ancestral web-building spiders are preferentially expressed in the integration center of the brain and that these genes are involved in learning and memory. —DJ

y g

Early Cretaceous. This finding may become even more apparent as more fossil flowers are discovered and described. These results provide quantitative support for several hypothesized and qualitatively described patterns in plant trait diversity. —BEL

RES EARCH

RESEARCH ARTICLE

◥

2D MATERIALS

De Haas–van Alphen spectroscopy and magnetic breakdown in moiré graphene Matan Bocarsly1†, Matan Uzan1†, Indranil Roy1†, Sameer Grover1, Jiewen Xiao1, Zhiyu Dong2, Mikhail Labendik1, Aviram Uri2, Martin E. Huber3, Yuri Myasoedov1, Kenji Watanabe4, Takashi Taniguchi5, Binghai Yan1, Leonid S. Levitov2*, Eli Zeldov1*

5 January 2024

1 of 7

,

Bocarsly et al., Science 383, 42–48 (2024)

y

*Corresponding author. Email: [email protected] (E.Z.); [email protected] (L.S.L.) †These authors contributed equally to this work.

y g

Department of Condensed Matter Physics, Weizmann Institute of Science, Rehovot 7610001, Israel. 2Department of Physics, Massachusetts Institute of Technology, Cambridge, MA 02139, USA. 3Departments of Physics and Electrical Engineering, University of Colorado Denver, Denver, CO 80217, USA. 4Research Center for Electronic and Optical Materials, National Institute for Materials Science, 1-1 Namiki, Tsukuba 305-0044, Japan. 5Research Center for Materials Nanoarchitectonics, National Institute for Materials Science, 1-1 Namiki, Tsukuba 305-0044, Japan.

y

1

the studies of 2D electron systems mostly to the nonthermodynamic Shubnikov–de Haas (SdH) oscillations in transport coefficients. Nevertheless, several studies have succeeded in resolving magnetization oscillations in 2D electron gas (2DEG) in gallium arsenide (GaAs) heterostructures by using millimeter-sized samples (13–16), as well as in magnetically doped zinc selenide (ZnSe) (17). By contrast, exfoliated clean van der Waals structures are typically limited to tens of micrometers in size, which makes observation of the dHvA effect in atomically thin systems quite challenging. Furthermore, previous dHvA studies in 2DEG and in bulk materials have been global, providing no spatial information on the local BS and thermodynamic electronic properties. We report the observation of dHvA effect in a van der Waals structure and spatial mapping of the thermodynamic QOs with resolution as high as 170 nm. We observed very large magnetization oscillations in moiré flat bands in Bernal-stacked bilayer graphene (BLG) aligned to hexagonal boron nitride (hBN) (2, 18, 19). The oscillations appear at low magnetic fields and at carrier densities of a few electrons per superlattice unit cell. In the integer quantum Hall effect (QHE), the periodicity of the QOs is tied in a universal manner to the carrier density through the LL degeneracy. By contrast, the observed oscillations display characteristic frequencies that are an order of magnitude lower than the fourfold-degeneracy frequency and form complex spectra, revealing the coexistence of multiple FSs and allowing accurate moiré BS reconstruction. When a number of FSs coexist, the QOs display several frequencies, reflecting the relative

Transport measurements of longitudinal resistivity (rxx) and Hall resistivity (ryx) (Fig. 1, A and B) of the BLG sample (Fig. 2A and fig. S1) were performed at a temperature T = 300 mK as a function of applied out-of-plane magnetic field Ba and carrier density n. Secondary peaks in rxx (Fig. 1C) at n = 4n0 ≅ ±3.48 × 1012 cm–2 indicate that the BLG is aligned to the hBN substrate with a twist of q ≅ 0.70°, forming a moiré superlattice with unit cell size l ≅ 11.5 nm (35), where n0 corresponds to one electron per moiré unit cell. The weak rxx peaks at filling factor n = n/n0 = ±4 reflect minima in the density of states (DOS) and the absence of a full gap between the low-energy flat bands and the dispersive remote bands. At low fields, ryx shows several sign reversals in Fig. 1B. The Hall carrier density nH extracted for Ba = 300 mT (Fig. 1D, green line) reveals a van Hove singularity at n ≅ 3.5, accompanied by a change in carrier type from electrons to holes. Similar behavior is found at n ≅ –3.5. Several additional nH sign reversals occur at higher jnj, which is consistent with the presence of several remote moiré bands. At higher fields, the sign reversals in ryx in Fig. 1B disappear (Fig. 1D, blue line), and nH follows nH = n (Fig. 1D, dashed red line), a behavior characteristic of a single band with no moiré potential. At elevated Ba, a Landau fan originating from the charge neutrality point (CNP) is visible at all fillings (Fig. 1A), corresponding to fourfold degenerate LLs stemming from spin and valley degeneracies in graphene (fig. S4B). Additionally, Hofstadter patterns are visible as horizontal lines periodic with f0/Ba (Fig. 1A, dashed black lines), arising from the interference of the moiré unit cell with the area occupied by a flux quantum f0 = h/e (where h is Planck’s constant and e is the elementary charge), as reported previously (1–3, 18, 36, 37). More complicated Landau fans originating from close to n = ±4 are discerned at intermediate fields. In the following, we describe local studies at fields

g

O

scillations in the thermodynamic and transport properties of metals subject to an external magnetic field are a fundamental quantum effect that originates from the quantization of cyclotron orbit areas. In two-dimensional (2D) systems, the periodicity of quantum oscillations (QOs), explained by discrete Landau levels (LLs), has a universal relation to the applied field and Fermi surface (FS) geometry. The oscillations are indispensable for resolving the band structure (BS) of moiré materials, in which the presence of a superlattice potential and enhanced electron-electron interactions lead to the formation of narrow minibands with multiple FSs and symmetry broken states (1–7). QOs can also reveal the band topology (8, 9) and straininduced pseudomagnetic fields in graphene (10–12). In bulk materials, QOs are detected by measuring magnetization oscillations caused by the de Haas–van Alphen (dHvA) effect. These oscillations in thermodynamic properties, however, are usually experimentally inaccessible in 2D electron systems because the signal scales with the sample volume and is therefore extremely weak in two dimensions. This limits

Transport measurements p

Quantum oscillations originating from the quantization of electron cyclotron orbits provide sensitive diagnostics of electron bands and interactions. We report on nanoscale imaging of the thermodynamic magnetization oscillations caused by the de Haas–van Alphen effect in moiré graphene. Scanning by means of superconducting quantum interference device (SQUID)–on-tip in Bernal bilayer graphene crystal axis-aligned to hexagonal boron nitride reveals large magnetization oscillations with amplitudes reaching 500 Bohr magneton per electron in weak magnetic fields, unexpectedly low frequencies, and high sensitivity to superlattice filling fraction. The oscillations allow us to reconstruct the complex band structure, revealing narrow moiré bands with multiple overlapping Fermi surfaces separated by unusually small momentum gaps. We identified sets of oscillations that violate the textbook Onsager Fermi surface sum rule, signaling formation of broad-band particle-hole superposition states induced by coherent magnetic breakdown.

size of the different Fermi pockets (FPs) encircled by the cyclotron orbits (17, 20–24). In addition to these fundamental orbits, electron orbits delocalized in k-space and supporting coherently entangled states in different bands can arise because of interpocket tunneling. Such tunneling, which is hindered by momentum conservation at zero magnetic field, is made possible at elevated fields through the coherent magnetic breakdown (CMB) mechanism (25–32) that is expected to occur in the vicinity of saddle points or Lifshitz transitions, across which the topology of the FS changes (23, 33, 34). Unlike conventional CMB, we found a breakdown that occurs at low applied fields, spans a wide range of energies, and displays QO frequencies that are described by fractional Onsager quantization relations.

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

p g y y g y

below 350 mT, for which QOs in transport measurements are barely resolved. Imaging QOs

To study the magnetization oscillations, we used a scanning superconducting quantum interference device (SQUID) fabricated on the apex of a sharp pipette [SQUID-on-tip (SOT)] (38). An indium SOT (39) of about 170 nm in Bocarsly et al., Science 383, 42–48 (2024)

above 500 Ω are saturated for clarity. (C) rxx versus n at Ba = 0 T showing secondary peaks caused by hBN-BLG moiré. (D) Hall carrier density nH = Ba/(eryx) versus n derived from ryx at Ba = 0.3 T (green), displaying several sharp sign reversals that stem from the moiré BS. At Ba = 4.2 T (blue), these sign reversals are absent. The dashed red line is drawn at a slope of 1 corresponding to nH = n in absence of moiré flat bands that are washed out by the full CMB.

diameter (fig. S2) was scanned at a height of h ≈ 200 nm above the sample surface (Fig. 2A) dc at T = 300 mK (35). The dc voltages Vtgdc and Vbg applied to the top and bottom Pt gates, respectively, were used to control n. A small ac voltage ac of 5 to 20 mV root mean square was applied Vbg to the backgate, modulating the carrier density by nac and the corresponding nac by 0.004 to ac 0.016, and the resulting Bac z ðx; y Þ ¼ n ðdBz =dnÞ

5 January 2024

was imaged across the sample. This signal reflects the induced ac modulation in the local magnetization, mz(x, y) = dMz(x, y)/dn, which can be reconstructed directly from the measured Bac z ðx; yÞ (fig. S3) by a numerical inversion (40) [Mz is the magnetization per unit area, and mz is the magnetization per excess electron; both are dominated by orbital effects (35)]. A map is shown in Fig. 2B of the 2 of 7

,

Fig. 1. Transport measurements in BLG aligned to hBN. (A) rxx versus carrier density n and magnetic field Ba at T = 300 mK shown on a logarithmic scale. The oscillations at high Ba reflect the existence of fourfold degenerate LLs (fig. S4) and display a Hofstadter’s butterfly (black dashed lines). (B) ryx versus n and Ba featuring several sign reversals at low field that disappear at higher Ba as the system enters the magnetic breakdown regime. Resistivity values

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

p g y

Fig. 2. Imaging the dHvA effect. (A) (Top) Optical image of the BLG-hBN sample with indicated contacts for rxx and ryx measurements. The dashed yellow rectangle indicates the area imaged in (B), and the dotted black line indicates the line cut presented in Fig. 3. (Bottom) Schematic sample structure indicating the applied top-gate and back-gate voltages, dc dc ac Vtg and Vbg þ Vbg , and the corresponding ac magnetic field Bac z imaged

BS of moiré bilayer graphene

To gain further insight, we performed continuummodel single-particle BS calculations of the BLGhBN moiré system (35, 41). The mismatch of graphene and hBN’s lattice constants creates a superlattice, causing band folding into the moiré mini-Brillouin zone (mBz) (1–3, 41–45). The calculated conduction C1 and valence V1 flat bands along with highly overlapping remote valence bands V2 and V3 are shown in Fig. 4A. Generally, the inequivalent lattice sites in the hBN substrate break inversion symmetry and open a gap at CNP. However, remote bands such as V2 and V3 are highly overlapping, creating complex FPs. Depending on BS parameters, 3 of 7

,

5 January 2024

than being restricted to integer fractions, f varies continuously as a function of n. (iii) At higher jnj, multiple characteristic frequencies are present simultaneously. (iv) There are substantial spatial variations in the relative intensity of the QOs (Fig. 3A and movie S2). These features reveal the presence of narrow moiré bands with overlapping FSs.

y

Bocarsly et al., Science 383, 42–48 (2024)

was markedly different, characterized by strong oscillations with substantially larger and variable periods Dn (Fig. 3, B and D). To quantify the periodicity, we performed a fast Fourier transform (FFT) over a narrow window of dn = 1.11 around a given n (35), for Bac z ðnÞ as indicated with the white dashed line in Fig. 3A (Fig. 3E; FFT at other positions are provided in movie S2). In the integer QHE, the frequency of the QOs as a function of n is given by f ¼ N1 Bf0a , where N is the spin-valley degeneracy. For a given N, f is determined solely by Ba and should thus be independent of the position and moiré band filling n. For jnj ≲ 3:5, the FFT reveals f0 , which is rather indepena peak at f ≅ 4B a dent of n and position (fig. S7 and movie S2). Therefore, the QOs at these fillings originate from the standard QHE, with fourfold degenerate LLs. At jnj ≳ 3:5, the QOs show a very rich behavior (Fig. 3E), departing from the standard QHE in a number of ways: (i) f is up to one f0 . (ii) Rather order of magnitude less than 4B a

y g

mz(x, y) at n = –7.65 and Ba = 334 mT, displaying extremely large values reaching ±500 Bohr magneton (mB)/electron and forming patches of positive and negative mz with a characteristic size of about 1 mm. Upon varying n, the patches move across the sample, and mz(x, y) reveals quasi-periodic oscillations (Fig. 2C and movie S1). The shape of the QOs and mz amplitude vary substantially with position (Fig. 2, D and E). We observed these oscillations from applied fields as low as 116 mT and up to Ba = 334 mT, at which our SOT has sufficient sensitivity (fig. S5). Similar behavior was found at various values of displacement field D (fig. S6) and in two additional samples (figs. S6 and S7). To investigate the origin of the QOs, we measured the evolution of Bac z ðx Þ with n over an extended range (Fig. 3A) by repeated scanning along the path indicated with the black dotted line in Fig. 2A, while incrementing n in 0.006 steps at Ba = 300 mT. For jnj ≲ 3:5, weak periodic oscillations in Bac z ðxÞ were discerned (Fig. 3C). For jnj ≳ 3:5, the behavior

with the scanning SOT. (B) Map of orbital magnetization at Ba = 334 mT, T = 300 mK, and n = –7.65, showing domains of positive and negative local magnetization mz(x, y) with amplitude of up to ±500 mB/electron. The color bar applies to (B) to (E). (C) Tomographic rendering of mz(x, y, z) (movie S1). (D) Slice of the tomographic data mz(y, n) at x = 1.36 mm. (E) mz(x, n) slice along y = 1.22 mm.

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

p g y y g

Bocarsly et al., Science 383, 42–48 (2024)

independent LLs, leading to QOs with multiple fundamental frequencies. Traditionally, in bulk materials QOs are measured versus 1/Ba, in which case, each FP has fi proportional to the FP area Si (20, 22). This conjecture, however, does not hold for QOs measured versus n, where the oscillation frequencies are given f0 1 @Si , where NX by fi ¼ NNiððeeFFÞÞ 4B i ðeF Þ ¼ 4p2 @e is the a DOS of pocket i, N ðeF Þ ¼ i N ðeF Þ is the total DOS, and eF is the Fermi energy, and we consider fourfold degenerate bands (35). This stems

5 January 2024

from the fact that varying Ba affects the cyclotron motion of the entire Fermi sea, whereas varying n affects behavior only near eF. Hence, upon increasing eF by De, one LL is added to a pocket when its Si has increased by DSi ¼ a 4p2 N i ðeF ÞDe ¼ 4B X f0 , leading to the Onsager sum rule of fundamental frequencies, i fi ¼ f0 f0 ¼ 4Ba . The experimental FFT data from Fig. 3E is shown in Fig. 4D overlaid with color-coded lines that indicate the fi (n) of each pocket calculated 4 of 7

,

either a full gap or a minimum in DOS occurs between V1 and V2 at n = –4 (35). For jnj ≲ 3:5, a single FP exists around the K point (Fig. 4B). At jnj ≅ 3:5, a Lifshitz transition occurs [which is consistent with the van Hove singularity at Ba = 0.3 T (Fig. 1D)], forming two FPs centered around X and Y points (movie S3). With increasing jnj, the FS topology becomes more complicated, consisting of three or more FPs resulting from overlapping bands (Fig. 4C). Each FP accommodates

(D) Bac z ðnÞ along the green segment in (A), revealing large-amplitude low-frequency oscillations comprising multiple frequencies. (E) FFT of Bac z ðnÞ at x = 4.97 mm indicated with the white dotted line in (A) performed over a narrow window of dn = 1.11 around n (FFT at different locations is shown in movie S2). The frequency is in units of f0/4Ba. At jnj ≲ 3:5, the QOs arise from conventional fourfold degenerate f0 . For jnj ≲ 3:5, the low-frequency oscillations are governed QHE LLs with f ¼ 4B a by multiple overlapping FSs. Positive and negative FFT frequencies are redundant and are shown for clarity.

y

Fig. 3. Evolution of the dHvA QOs with n and position. (A) Bac z ðx; nÞ measured along the dotted black line in Fig. 1A at Ba = 300 mT (additional field values are available in fig. S5). At jnj ≲ 3:5, the orbital magnetization and the corresponding ac Bac z are weak, whereas for jnj ≳ 3:5, large Bz accompanied by pronounced lowfrequency QOs are present. (B) Zoomed-in cross-section of Bac z ðnÞ along the magenta segment in (A), showing low-frequency QOs with a period that gradually varies with n. (C) Bac z ðnÞ cross-section along the red segment in (A), showing high-frequency periodic oscillations caused by conventional fourfold-degenerate LLs.

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

p g y

Fig. 4. Calculation of BLG-hBN BS and of QOs. (A) Single-particle BS calculation for a single valley (K) showing (top) the conduction flat band (C1), (middle) the valence flat band (V1), and (bottom) two partially overlapping remote valence bands (V2 and V3) in the moiré mBz of BLG aligned to hBN with q = 0.75°. The bands are fourfold degenerate, with K′ bands rotated by 180°. Tight-binding parameters were chosen to best fit (D) with rNB = 0.5 and w = 0.5

Magnetic breakdown

The BS calculations with enhanced lattice relaxation provide a good description of the observed fundamental QO frequencies. Yet there are a number of prominent lines in Fig. 4D that are not accounted for by the calculated fi(n). Moreover, these lines Xdo not obey the Onsager band area sum rule i fi ¼ f0 and cannot be explained by simple harmonics of fi or by sample disorder (35). These unaccounted-for lines indicate the presence of electron orbits that encompass areas outside the closed FS contours. Such trajectories are facilitated by interband electron tunneling caused by the CMB mechanism, which has been widely investigated in bulk metals (27) but has not been identified in 2D vdW materials. When two FPs are separated by a small momentumgap Dk, the magnetic field–induced interpocket tunneling occurs with probability !1=2 BMB f Dk3 f ≅ 0 Dk2 P ≅ e Ba ; BMB ¼ 0 1 2 R1 þ R12 2 5 of 7

,

5 January 2024

tion mechanisms and its effect on hBN-aligned moiré heterostructures.

y

Bocarsly et al., Science 383, 42–48 (2024)

tunneling strengths between overlapping boron and carbon atoms (tBC) and between overlapping nitrogen and carbon (tNC) to be equal, or rNB ≡ tNC/tBC = 1. In (45), rNB ≅ 0.67, according to ab initio density functional theory calculations. We found that both of these parameter sets fail to fit our experimental data (fig. S8). Experiments have revealed substantial lattice relaxation in magic-angle twisted bilayer graphene (46), which has a substantial impact on the BS (4, 47–49). Lattice relaxation is captured with the phenomenological parameter w = tAA/tAB ≈ 0.8 (50). Recently, simulations on aligned BLG-hBN heterostructures have also shown substantial lattice relaxation, which was proposed to affect band topology (19). By correlating the high-resolution QOs with simulations, we found rNB ≅ 0.5 and w ≅ 0.5 throughout the sample (movie S2). These values are substantially lower than estimated previously, leading to larger band overlaps with smaller energy gaps (fig. S8) and to an enhancement of magnetic breakdown. This finding of strong lattice relaxation in aligned BLG-hBN calls for further exploration of lattice relaxa-

y g

from BS. For hole doping at 0 > n > –3.5, one FP around K point (V1Κ) is present, resulting in a single frequency fV1K = f0 (Fig. 4D, green). At the Lifshitz transition at n ≅ –3.5, the FS breaks into two pockets around the mBz corners, V1X and V1Y (movie S3). As a result, two QO frequencies coexist for a small region of n, fV1X and fV1Y, until V1 (Fig. 4D, green) overlaps V2 (Fig. 4D, light brown). At n < –5, V3 band (Fig. 4D, pink) starts to be occupied, forming two FPs, V3X and V3Y, with increasing DOS with jnj that coexist with the V2Κ FP with decreasing DOS (Fig. 4C). As a result, the two V3 frequencies, fV3X and fV3Y, grow with jnj (Fig. 4D, pink), whereas the fV2K frequency decreases (Fig. 4D, light brown). The calculated behavior for –5 > n > –10.5 closely follows the experimentally derived frequencies and their evolution with n. We observed similar behavior for electron doping at n > 3.5. With the above insight, the high sensitivity to oscillation frequencies at fillings where multiple FPs coexist makes nanoscale magnetization imaging a particularly sensitive tool for mapping the local BS. The work in (41) takes the

(fig. S8) (35). (B and C) Example of (B) simple and (C) complex FSs (solid contours) at n = 2.054 and –10.14, respectively. The dashed contours indicate the change in the areas of the FPs with small increase in n, reflecting the DOS and the QO frequencies of each pocket. (D) The FFT of QOs from Fig. 3E, overlaid with fundamental frequencies (lines coded by band colors) calculated by the f0 relative DOS of each pocket, fi ðnÞ ¼ NNiððeeFFÞÞ 4B . a

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

p g y

5 January 2024

gives rise to extended equi-energy electron orbits. The green dashed line in Fig. 5A indicates the shortest orbit that traces two-thirds of the circumference of the two electron pockets C3X and C3Y and one-third of the circumference of the hole pocket C2K. This results in a QO frequency given by a fractional Onsager relation 1 fMB ¼ ð2fC3X þ 2fC3Y þ fC2K Þ 3 where fi are the frequencies of the individual FPs (35), in contrast to common CMB behavior with integer Onsager relations (27, 32). Shown 6 of 7

,

Bocarsly et al., Science 383, 42–48 (2024)

(Fig. 5A, dark pink). The sharp touching points between the C3Y and C2K pockets are characterized by a very small momentum gap Dk ≅ 0.010 nm–1, and the gaps between the C3X and C2K pockets have an even smaller Dk ≅ 0.005 nm–1 (Fig. 5B, inset), leading to CMB at low fields. Moreover, in contrast to the common situations in which close proximity of FPs is limited to the vicinity of Lifshitz transitions, the distinct highly overlapping BS of relaxed BLG-hBN leads to small gaps extending over almost the entire energy range of the remote bands, with sharp ridges that closely follow each other (Fig. 5B). The tunneling between the FPs

y

where BMB is the breakdown field and R1 and R2 are the k-space radii of curvature of the two FSs in the gap region (27). For BMB ≲ Ba ¼ 0:3 T, this requires Dk ≲ 0:012 nm–1. Our BS calculations show very small gaps between FPs in remote bands (Fig. 5B). To derive the MB orbits and their corresponding QO frequencies fMB, we analyzed two prominent unaccounted-for lines in the experiment at 5 < n < 10.5 and –12.5 < n < –10 (Fig. 5, C and E, bright green). The FS structure at n = 8.4 is shown in Fig. 5A, displaying electron pockets at X and Y originating from the C3 band (Fig. 5A, red) and a hole pocket at K from the C2 band

of Bac z ðnÞ at x = 2.04 mm, showing a pronounced frequency line for 4.5 < n < 10 that does not obey the sum rule. This frequency is accounted for by the CMB orbit in (A), resulting in fMB ¼ 31 ð2C3K þ 2fC3Y þ fC2K Þ (green line). (D) Calculated constant energy BS cut at n = –12.28 showing three degenerate electron pockets from the V4 band (brown) and one hole pocket from the V3 band (light purple). Scale bar applies to both (A) and (D). The dashed green trajectory indicates the shortest CMB orbit. (E) FFT of Bac z ðnÞ at x = 4.44 mm with overlaid calculated fMB (green line) (35). Positive and negative FFT frequencies are redundant and are shown for clarity.

y g

Fig. 5. Coherent magnetic breakdown. (A) Calculated constant energy BS cut at n = 8.4 showing the occupied C3X and C3Y electron pockets (red, clockwise black arrows) and the C2K hole pocket (dark pink, counterclockwise arrows). Scale bar is shown in (D). The dashed green trajectory indicates the shortest magnetic breakdown orbit. (B) Cut of the calculated BS along the dashed black line in (A). Dashed line at e = 154.2 meV corresponds to the energy value in (A). The gap between C2K and C3X pockets remains very small over a large range of energies. (Inset) Zoom-in showing a gap of Dk ≅ 0.005 nm–1. (C) Expanded view of FFT

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

1. 2. 3. 4. 5. 6. 7. 8. 9. 10.

14. 15.

16.

20. 21.

24. 25. 26. 27. 28. 29. 30. 31. 32. 33. 34.

5 January 2024

45. 46. 47. 48. 49. 50. 51.

AC KNOWLED GME NTS

The authors thank E. Berg, A. Stern, I. Rozhansky, and N. Avraham Tayar for useful discussions. Funding: This work was supported by the European Union (ERC, MoireMultiProbe – 101089714): Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or the European Research Council. Neither the European Union nor the granting authority can be held responsible for them (to E.Z.); the Minerva Foundation, with funding from the Federal German Ministry of Education and Research Grant 140687 (to E.Z.); United States–Israel Binational Science Foundation (BSF) Grant 2022013 (to E.Z. and L.S.L.); Andre Deloro Prize for Scientific Research and Leona M. and Harry B. Helmsley Charitable Trust grant 2112-04911 (to E.Z.); Goldfield Family Charitable Trust (to E.Z.); Science and Technology Center for Integrated Quantum Materials, NSF grant DMR1231319, and Army Research Office Grant W911NF-181-0116 (to L.S.L.); Sagol Weizmann-MIT Bridge Program (to E.Z. and L.S.L.); European Research Council (ERC Consolidator Grant “NonlinearTopo,” 815869) (to B.Y.); Israel Science Foundation ISF grant 2932/21 (to B.Y.); JSPS KAKENHI grants 21H05233 and 23H02052 (to K.W. and T.T.); World Premier International Research Center Initiative (WPI), MEXT, Japan (to K.W. and T.T.); MIT Pappalardo Fellowship (to A.U.); VATAT Outstanding Postdoctoral Fellowship in Quantum Science and Technology (to A.U.); and a VATAT Outstanding PhD Fellowship in Quantum Science and Technology (to M.B.) Author contributions: Local magnetization measurements: M.B., I.R., and S.G. Sample fabrication and transport measurements: M.U. SOT fabrication and tuning fork feedback: I.R., M.L., and Y.M. Development of experimental setup: A.U. Design and building of the SOT readout system: M.E.H. BS calculations: J.X., B.Y., and M.B. Development of theoretical model (CMB): M.B., E.Z., L.S.L., and Z.D. Data analysis: M.B., I.R., M.U., and S.G. hBN crystals: K.W. and T.T. Writing of original manuscript: M.B., E.Z., L.S.L., J.X., M.U., I.R., and Z.D. Editing and review of manuscript: all authors. Competing interests: The authors declare no competing interests. Data and materials availability: The data from this study and code for BS calculations are available at the Weizmann Institute of Science repository (51). License information: Copyright © 2024 the authors, some rights reserved; exclusive licensee American Association for the Advancement of Science. No claim to original US government works. https://www.science.org/about/sciencelicenses-journal-article-reuse SUPPLEMENTARY MATERIALS

science.org/doi/10.1126/science.adh3499 Materials and Methods Supplementary Text Figs. S1 to S9 References (52–72) Movies S1 to S3 Submitted 26 February 2023; accepted 23 November 2023 10.1126/science.adh3499

7 of 7

,

23.

44.

y

22.

42. 43.

y g

18. 19.

L. A. Ponomarenko et al., Nature 497, 594–597 (2013). C. R. Dean et al., Nature 497, 598–602 (2013). B. Hunt et al., Science 340, 1427–1430 (2013). Y. Cao et al., Nature 556, 80–84 (2018). H. S. Arora et al., Nature 583, 379–384 (2020). P. Stepanov et al., Phys. Rev. Lett. 127, 197701 (2021). J. M. Park et al., Nat. Mater. 21, 877–883 (2022). A. Alexandradinata, C. Wang, W. Duan, L. Glazman, Phys. Rev. X 8, 011027 (2018). C. Guo et al., Nat. Commun. 12, 6213 (2021). F. Guinea, M. I. Katsnelson, A. K. Geim, Nat. Phys. 6, 30–33 (2010). N. Levy et al., Science 329, 544–547 (2010). J. Mao et al., Nature 584, 215–220 (2020). S. A. J. Wiegers et al., Phys. Rev. Lett. 79, 3238–3241 (1997). I. Meinel et al., Phys. Rev. Lett. 82, 819–822 (1999). N. Ruhe, J. I. Springborn, C. Heyn, M. A. Wilde, D. Grundler, Phys. Rev. B Condens. Matter Mater. Phys. 74, 235326 (2006). T. Windisch et al., Phys. Rev. B Condens. Matter Mater. Phys. 80, 205306 (2009). J. G. E. Harris et al., Phys. Rev. Lett. 86, 4644–4647 (2001). T. Iwasaki et al., Nano Lett. 20, 2551–2557 (2020). E. Arrighi et al., Non-identical moiré twins in bilayer graphene. arXiv:2205.01760 (2022). A. S. Alexandrov, A. M. Bratkovsky, Phys. Rev. Lett. 76, 1308–1311 (1996). R. A. Shepherd et al., Phys. Rev. B Condens. Matter 60, R11277–R11280 (1999). J.-Y. Fortin, E. Perez, A. Audouard, Phys. Rev. B Condens. Matter Mater. Phys. 71, 155101 (2005). K. Kishigi, Y. Hasegawa, Phys. Rev. B 104, 085412 (2021). H. Zhou et al., Science 375, 774–778 (2022). C. E. T. G. da Silva, L. M. Falicov, Phys. Rev., B, Solid State 8, 527–535 (1973). M. I. Kaganov, A. A. Slutskin, Phys. Rep. 98, 189–271 (1983). D. Shoenberg, Magnetic Oscillations in Metals (Cambridge Univ. Press, 1984). P. Steda, A. H. MacDonald, Phys. Rev. B Condens. Matter 41, 11892–11898 (1990). P. H. Beton et al., Phys. Rev. B Condens. Matter 43, 9980–9983 (1991). V. M. Gvozdikov, Phys. Rev. B Condens. Matter Mater. Phys. 75, 115106 (2007). P. Walmsley et al., Phys. Rev. B 102, 045150 (2020). J. F. Linnartz et al., Phys. Rev. Res. 4, L012005 (2022). C.-K. Lu, H. A. Fertig, Phys. Rev. B Condens. Matter Mater. Phys. 89, 085408 (2014). A. Alexandradinata, L. Glazman, Phys. Rev. Lett. 119, 256601 (2017).

41.

Materials and methods are available as supplementary materials. R. Krishna Kumar et al., Science 357, 181–184 (2017). J. Barrier et al., Nat. Commun. 11, 5756 (2020). D. Vasyukov et al., Nat. Nanotechnol. 8, 639–644 (2013). Y. Anahory et al., Nanoscale 12, 3174–3182 (2020). A. Y. Meltzer, E. Levin, E. Zeldov, Phys. Rev. Appl. 8, 064030 (2017). P. Moon, M. Koshino, Phys. Rev. B Condens. Matter Mater. Phys. 90, 155406 (2014). M. Yankowitz et al., Nat. Phys. 8, 382–386 (2012). J. R. Wallbank, A. A. Patel, M. Mucha-Kruczyński, A. K. Geim, V. I. Fal’ko, Phys. Rev. B Condens. Matter Mater. Phys. 87, 245408 (2013). P. San-Jose, A. Gutiérrez-Rubio, M. Sturla, F. Guinea, Phys. Rev. B Condens. Matter Mater. Phys. 90, 075428 (2014). J. Jung, A. Raoux, Z. Qiao, A. H. MacDonald, Phys. Rev. B Condens. Matter Mater. Phys. 89, 205414 (2014). Y. Jiang et al., Nature 573, 91–95 (2019). P. Lucignano, D. Alfè, V. Cataudella, D. Ninno, G. Cantele, Phys. Rev. B 99, 195419 (2019). M. Angeli et al., Phys. Rev. B 98, 235137 (2018). X. Lu et al., Nature 574, 653–657 (2019). M. Koshino et al., Phys. Rev. X 8, 031087 (2018). M. Bocarsly et al., Data and code for: de Haas–van Alphen spectroscopy and magnetic breakdown in moiré graphene. WIS Library (2023); https://doi.org/10.34933/e6c6e1ae-7c9c45ec-a607-fbbdb00a9616.

y

11. 12. 13.

17.

Owing to the extreme sensitivity of the QOs to the BS details, our measurement provides a distinctive probe of multiband moiré FSs and their low-energy electronic properties. Crucially, the ability to detect thermodynamic QOs at low fields allows probing of the BS with high energy resolution and without perturbing it with high magnetic fields. In particular, our results show that the hBN-graphene coupling is substantially weaker than previously estimated values, giving rNB = tNC/tBC ≅ 0.5 and a large lattice relaxation with w = tAA/tAB ≅ 0.5, a direct demonstration of a weak moiré potential with strongly overlapping bands and small energy gaps. This fine potential modulation is readily washed out at elevated magnetic fields, leading to a full breakdown in which the carriers orbit mostly along the original BLG FS unperturbed by the moiré potential (fig. S9) (35). The nH measured at Ba = 4.2 T shown in Fig. 1D shows essentially no signs of the moiré multi-pocket BS resolved at low fields. Our findings open up a previously uncharted regime of exotic CMB physics at unusually low Bocarsly et al., Science 383, 42–48 (2024)

RE FERENCES AND NOTES

35. 36. 37. 38. 39. 40.

g

Discussion and outlook

Ba, arising from cyclotron orbits delocalized in k-space and supporting states coherently entangled among different subbands. This regime is manifested through QOs that do not obey Onsager’s FS area sum rule. Instead, we observed a fractional Onsager quantization relation that indicates the occurrence of particlehole superposition states shared by adjacent bands and exhibiting a high degree of interband phase coherence. The real-space cyclotron radius of the observed CMB orbits in Fig. 5A  f0 S 1=2 ≅ 500 nm, a value is as large as Rc ¼ 2pB a p comparable with the characteristic scales of disorder and sample dimensions (35), which calls for further theoretical and experimental studies for understanding the QOs in the presence of BS disorder. The particle-hole coherence induced by CMB is an appealing and not-yet-explored direction for band engineering in moiré materials.

p

in Fig. 5C is a good fit between the calculated fMB (Fig. 5C, green) and the experimentally unaccounted for QO line. The fMB line is derived from the fundamental fi, with no adjustable parameters and over their entire range. A gap of Dk ≅ 0.01 nm–1 corresponds to tunneling probability P ≈ 0.5, allowing the carriers to orbit along both the closed FPs and along the CMB trajectories. As a result, both the fundamental (Fig. 5C, red and pink) and the CMB frequency lines (Fig. 5C, green) were observed concurrently. Similar CMB QOs were observed in device 2 with a larger hBN twist angle q = 1.15° (fig. S7). The BS at n = –12.28 is shown in Fig. 5D, displaying three degenerate hole pockets in the V4 band and an electron pocket at K in the V3 band. CMB creates an electron trajectory that flows along the inner and outer edges of the V3 and V4 pockets (Fig. 5D, green dashed line), which explains well the unaccounted QO line in Fig. 5E as indicated with the calculated green fMB line. The fundamental fV3 line (Fig. 5E, purple) is essentially invisible in the experiment. This can be understood in view of the extremely small gap between the V3 and V4 pockets with Dk ≅ 0.02 nm–1 and very small radii of curvature R. As a result, the electrons tunnel between the pockets with P ≅ 1, leaving essentially no carriers that circulate exclusively in the V3 pocket and hence no detectable fV3. Moreover, in contrast to the usual CMB behavior (32) and unlike the fMB line in the conduction bands (Fig. 5C, green), the fMB line in Fig. 5E cannot be expressed as either integer or fractional Onsager combination of the fundamental frequencies because of nontrivial evolution of the FS with doping.

RES EARCH

ELECTROCATALYSIS

Direct propylene epoxidation via water activation over Pd-Pt electrocatalysts Minju Chung1, Joseph H. Maalouf1, Jason S. Adams2, Chenyu Jiang2, Yuriy Román-Leshkov1, Karthish Manthiram2* Direct electrochemical propylene epoxidation by means of water-oxidation intermediates presents a sustainable alternative to existing routes that involve hazardous chlorine or peroxide reagents. We report an oxidized palladium-platinum alloy catalyst (PdPtOx/C), which reaches a Faradaic efficiency of 66 ± 5% toward propylene epoxidation at 50 milliamperes per square centimeter at ambient temperature and pressure. Embedding platinum into the palladium oxide crystal structure stabilized oxidized platinum species, resulting in improved catalyst performance. The reaction kinetics suggest that epoxidation on PdPtOx/C proceeds through electrophilic attack by metal-bound peroxo intermediates. This work demonstrates an effective strategy for selective electrochemical oxygen-atom transfer from water, without mediators, for diverse oxygenation reactions.

1 of 7

,

5 January 2024

We examined the bulk and surface properties of our annealed PdPt/C, Pd/C, and Pt/C catalysts (PdPtOx/C, PdOx/C, and PtOx/C, respectively)

y

Chung et al., Science 383, 49–55 (2024)

Catalyst characterizations

y g

*Corresponding author. Email: [email protected]

Previously studied catalysts for electrochemical propylene epoxidation can be clustered into palladium-derived (Pd) (9–12), platinumderived (Pt) (13–15), and silver-derived (Ag) (16, 17) groups (Fig. 1B and table S1), and the best catalytic performance for electrochemical epoxidation has been observed with oxidized Pd and Pt catalysts. Studies on Pd catalysts have shown that epoxidation occurs on the surface oxide layer at anodic potentials >1.1 V versus reversible hydrogen electrode (RHE), whereas allylic oxidation producing acrolein and acrylic acid occurs on the metallic surface at lower potentials (9, 12). The performance enhancement of oxidized Pt catalysts was correlated with an increase in surface presence of Pt(II) and Pt(IV) species, whereas pure PtO2 was less active (14). Moreover, a theoretical study has identified weak oxygen-binding metal oxides as promising catalyst candidates for propylene epoxidation and cautioned that the metallic phase of strong-binding metals, such as Pt and Pd, may facilitate the production of unwanted side products through dehydrogenation (18). These findings suggest that the catalytic performance may be further optimized by adjusting the oxidation states of the metal. However, preparing Pt oxide catalysts with high concentrations of oxidized Pt species is challenging because of their instability. By contrast, stable Pd oxide catalysts can be obtained through high-temperature annealing, but they generally exhibit lower rates of epoxidation.

y

Department of Chemical Engineering, Massachusetts Institute of Technology, Cambridge, MA 02139, USA. 2 Division of Chemistry and Chemical Engineering, California Institute of Technology, Pasadena, CA 91125, USA.

Catalytic performance of PdPtOx/C and design principle

g

1

possible to directly transfer reactive intermediates typically seen in water oxidation to the alkene before the intermediates are fully oxidized into O2 (Fig. 1A, top) (6–8). This direct electrochemical alkene-epoxidation approach does not involve hazardous oxidizing agents, such as Cl2, making it an attractive avenue for high-demand propylene epoxidation.

p

P

ropylene oxide (PO) is an important chemical building block used in the production of many commodity chemicals, such as polyurethanes and polyesters (1). Currently, PO production relies on highly reactive, corrosive, or explosive chemical oxidants such as molecular chlorine (Cl2) or peroxides (2). Although direct oxidation of propylene with molecular oxygen (O2) has been explored as an alternative solution, achieving high selectivity for PO is challenging because of the allylic hydrogen stripping in propylene and consequent further oxidation (2). In this regard, a selective direct propylene-epoxidation pathway that uses water as a cleaner and an easily accessible oxygen source would present a compelling alternative to the current epoxidation methods. Electrochemistry offers a promising approach for using water as a sustainable source of oxygen atoms at ambient temperature and pressure. In water electrolysis, typically oxygen and hydrogen are generated, but alternatively, commodity chemicals can be synthesized through oxidative processes at the anode of an electrochemical cell combined with hydrogen evolution at the cathode. For example, chloride oxidation can be conducted at the anode to generate Cl2, which acts as a redox mediator in alkene epoxidation (Fig. 1A, bottom) (3–5). This electrified version of the chlorohydrin process was demonstrated to be effective at industrially relevant current densities and selectivity (4). However, this method produces byproducts such as hypochlorite, chlorinated organic compounds, and brine streams that pose serious environmental challenges unless 100% selectivity and recycling of the chloride electrolytes can be achieved. Rather than relying on Cl2 as a redox mediator, it should be

Alloying Pt with Pd can be a strategy that leverages complementary properties of the two elements, enabling the attainment of optimal oxidation states and structures for an epoxidation catalyst (Fig. 1C). Under aerobic conditions, Pt nanoparticles tend to sinter into a metallic form when annealed above 400°C (19) because volatile PtO2 species mediate the sintering process by facilitating Ostwald ripening (20, 21). By contrast, stable PdO nanoparticles can be produced through annealing of Pd nanoparticles without substantial sintering (19, 22). PdO can trap mobile PtO2 species, which results in less sintering of Pd-Pt alloy nanoparticles than of Pt nanoparticles (20, 23). In addition, Pd and Pt share very similar crystal structures and atomic sizes, allowing homogeneous distributions of Pd and Pt in their alloys. In this study, we designed a Pd-Pt alloy catalyst containing both Pd oxide and Pt oxide by embedding and stabilizing Pt oxide species in Pd oxide. A series of alloys with varying ratios of Pd and Pt were synthesized using a coreduction method (24) on an amorphous carbon substrate and subsequently annealed under static air before being tested for their activity in propylene epoxidation (materials and methods and figs. S1 to S3). Our catalyst is compatible with a blended electrolyte composed of acetonitrile and water (Fig. 1D) that can accommodate a broad spectrum of alkene substrates, as well as aqueous electrolytes (Fig. 1E) that can minimize overall cell voltage for gaseousalkene epoxidation (fig. S4). The alloy catalyst with an equimolar composition of Pd and Pt annealed at 500°C (PdPtOx/C) was most active toward propylene epoxidation and outperformed previously reported catalysts for direct anodic epoxidation under ambient conditions (Fig. 1B and table S1). Furthermore, alloys with metal compositions of Pd3Pt1 and Pd1Pt3 exhibited epoxidation activities on par with those of PdPtOx/C; when Pd and Pt are mixed at the nanometer scale (figs. S2 and S3), Pd-Pt alloy catalysts can achieve enhanced epoxidation rates and Faradaic efficiencies (FEs) relative to either Pd or Pt alone (Fig. 1, D and E, and fig. S5). In an electrolytic cell equipped with circulating liquid electrolytes (fig. S4), we demonstrated continuous PO production with high FE averaging 66 ± 5% [propylene glycol (PG); 1.3 ± 0.3%] for 3 hours at a constant current density of 50 mA/cm2. Additionally, PdPtOx/C catalyzed cyclooctene epoxidation with an FE of ~90%, a notable improvement over the results observed in our previous studies, in which we used Ir-MnOx (~50%) and MnOx (~30%) catalysts (table S2) (7, 8).

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

p g y y g

Chung et al., Science 383, 49–55 (2024)

5 January 2024

to 40.0° in PdPtOx/C. Furthermore, the broader peaks in the PdPtOx/C XRD pattern indicate that the PdPtOx/C has smaller crystallites than those of PtOx/C (table S3), which is consistent with our hypothesis that the sintering of PdPt/C is less severe than that of Pt/C upon annealing. Extended x-ray absorption fine structure (EXAFS) measurements revealed that, within fitting errors, Pd and Pt coordinate with the same number of oxygens in PdPtOx/C; the Pd-O and Pt-O coordination numbers were estimated to be 3.7 ± 0.4 and 2.8 ± 0.6, respectively (Fig. 2,

B and C, and tables S4 to S6). The local coordination environments of Pd in PdPtOx/C and PdOx/C were similar to that of PdO, whereas PtOx/C resembled the Pt foil standard and did not show any contribution from the Pt-O scattering path. By contrast, the local environment of Pt in PdPtOx/C was different from that in the Pt foil or PtO2 standard, and the first-shell scattering path of PdPtOx/C was fit with a combination of Pt-O and Pt-Pd/Pt scattering paths. These EXAFS data suggest that blending Pt in the PdO structure induced a local coordination 2 of 7

,

to shed light on the enhanced performance of PdPtOx/C compared with PdOx/C and PtOx/C. X-ray diffraction (XRD) patterns showed that although PdOx/C and PtOx/C only exhibited oxide and metallic phases, respectively, the PdPtOx/C contained both phases (Fig. 2A and fig. S6). The oxide phase of PdPtOx/C resembles that of PdOx/C because the (101) peak in PdPtOx/C at 33.7° occurs at a similar angle as the PdOx/C peak at 33.9°. Similarly, the metallic phase of PdPtOx/C parallels that of PtOx/C, with the (111) peak shifted from 39.8° in PtOx/C

performance with different compositions of Pd-Pt-on-carbon catalysts in (D) water-acetonitrile electrolyte and (E) aqueous electrolyte. Potentials in this figure were 100% iR-compensated (i, current; R, resistance). The vertical error bars represent standard deviations from the mean of multiple replicates (n = 3) of the same experiment.

y

Fig. 1. Background and demonstration of direct electrochemical propylene epoxidation. (A) Comparison of chlorine-mediated and direct electrochemical propylene epoxidation. (B) Compilation of previously reported direct electrochemical propylene epoxidation and our work. (C) Effect of annealing on Pt, Pd, and Pd-Pt nanoparticles. (D and E) Direct electrochemical propylene-epoxidation

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

p g

Electrochemical kinetic study and proposed reaction mechanism

Understanding the mechanism of direct propylene epoxidation on PdPtOx/C can help us develop more efficient catalysts and process conditions for producing PO. The reaction mechanism was investigated with a multifaceted approach, including kinetic rate measurements, deuterium kinetic isotope effect measurements, 3 of 7

,

5 January 2024

Furthermore, EXAFS fitting of catalysts under operating conditions showed that coordination numbers and scattering-path lengths of Pd and Pt did not change during electrolysis (tables S5 and S6). Applying a cathodic potential followed by an anodic potential demonstrated that the edge energies and the coordination environment of Pd and Pt respond to applied potentials in a partially, albeit not fully, reversible fashion (Fig. 3 and figs. S10 and S11). The results from XRD, CV, and the elemental analysis collectively suggest that Pd and Pt are mixed in both the bulk and surface of the PdPtOx/C catalyst. XANES and XPS results revealed that a larger quantity of oxidized Pt species is stabilized in PdPtOx/C than in PtOx/C. XAS data suggest that Pd stabilizes Pt at a similar oxidation state (around +2) and crystal structure as those of PdO, and these combined factors improve catalytic performance.

y

Chung et al., Science 383, 49–55 (2024)

S8 and S9), indicating that the higher ratio of oxidized Pd and Pt species is contributing to the enhancement in the catalyst performance. Operando x-ray absorption spectroscopy (XAS) experiments probing the Pd K-edge and Pt L3edges during epoxidation highlighted again that achieving more oxidized Pt may have a critical role in improving catalytic activity toward propylene epoxidation. Progressively more oxidative potentials were applied from 0.45 to 1.15 V versus Fc/Fc+, followed by a reductive step to −1.35 V and then two oxidative steps back to 1.15 V. The average metal valency at each condition was estimated from linear combination fitting of x-ray absorption near edge structure (XANES) spectra by using reference spectra from standard materials. The results indicate that PdO stabilizes oxidized Pt species in Pd-Pt alloys, leading to a higher average oxidation state of Pt in PdPtOx/C (+2) compared with that in PtOx/C (0). This difference is maintained under anodic potentials (0.45 to 1.15 V versus Fc/Fc+, Fig. 3A). Unlike the stark contrast in Pt valency between PdPtOx/C and PtOx/C, the Pd valency stayed the same (+2) in both PdPtOx/C and PdOx/C (Fig. 3B). These observations suggest that the more-oxidized Pt species contribute to the enhancement in the production rate and selectivity of PO.

y g

environment of Pt resembling that of PtO, which is unstable alone. The mixing of Pd and Pt on the surface, matching the bulk composition, was confirmed through cyclic voltammetry (CV) and x-ray photoelectron spectroscopy (XPS) analyses. A single peak of surface-oxide reduction [at ~0 V versus ferrocene/ferrocenium (Fc/Fc+)] was observed in the CV of the PdPtOx/C electrode, suggesting that Pd and Pt coexist on the catalyst surface (Fig. 2D and fig. S7). Furthermore, near-surface compositions derived from XPS survey spectra also indicate that Pd and Pt are present in an equimolar ratio near the surface (table S7), matching the bulk composition determined with inductively coupled plasmaoptical emission spectrometry (table S8). XPS analysis showed that alloying Pt with Pd stabilizes oxidized Pt on the catalyst surface. Pt(0) species were dominant in the PtOx/C, whereas a higher content of Pt(II) species was observed in PdPtOx/C (Fig. 2F). The ratio of Pt(II):Pt(0) increased from 1.0 to 6.4 as the catalyst treatment temperature was increased from 400° to 500°C (table S9). A small amount of Pd(0) was present in PdPtOx/C annealed at 400°C, whereas Pd(II) was exclusively observed in the catalyst annealed at 500°C (Fig. 2E). The epoxidation rate and selectivity also improved with the treatment temperature for PdPt/C (figs.

voltammetry traces of the catalysts. The scans were recorded at 50 mV/sec scan rate with 85% automatic iR compensation (i, current; R, resistance). (E) Highresolution Pd 3d XPS spectra of annealed Pd/C and PdPt/C catalysts. (F) Highresolution Pt 4f XPS spectra of annealed Pt/C and PdPt/C catalysts.

y

Fig. 2. Ex situ characterizations of annealed PdPt/C, Pd/C, and Pt/C catalysts. (A) Magnified XRD patterns around the peak associated with the oxide and metallic phases of annealed catalysts at 500°C. (B and C) EXAFS spectra of standards and catalysts at (B) Pd K-edge and (C) Pt L3-edge. (D) Cyclic

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

y ,

4 of 7

y g

5 January 2024

0.48 mA/cm2. Diverse derivatives of this mechanism involving the OO* epoxidation intermediate are plausible, depending on whether a dual or single site is presumed for OO* and O* intermediates. We selected the model with the lowest RMSE as our representative mechanism (Fig. 4D), and the model provides a quantitative explanation of the data illustrated in Fig. 4, A to C. We also simulated surface coverage with the model (Fig. 4E and supplementary text S1). At higher propylene pressures where O* and OH* are abundant, the overall epoxidation rate would be largely determined by the rate of water oxidation to form OOH*. However, the comparable prevalence of different surface-oxygen intermediates underscores the necessity for a comprehensive kinetic model that does not assume a single dominant reactive intermediate and a rate-limiting step. Kinetic isotope labeling and stilbene control experiments further support our hypothesis asserting OO* as the key intermediate in epoxidation. The use of deuterated water lowered the epoxidation rate (kinetic isotope effect value kH/kD = 1.8 ± 0.5; fig. S15 and table S10), and the FE remained similar. The difference observed in the epoxidation rate is consistent with the kinetically relevant water-activation steps involving O–H cleavages in our proposed mechanism. Moreover, our mechanism identifies OO* as the branching intermediate for both epoxide and O2 evolution (Fig. 4D), providing an explanation for the marginal selectivity difference because neither of these pathways from OO* involves O–H cleavage. In addition, cisstilbene was used as a probe substrate to help determine whether epoxidation proceeds through a concerted mechanism with OOH* or a stepwise pathway with OO* (25, 26). Cis-stilbene (96%) epoxidation with PdPtOx/C produced 23% cis-stilbene oxide and 77% trans-stilbene oxide, which suggests that alkene epoxidation over

y

Chung et al., Science 383, 49–55 (2024)

first order at a higher water ratio (Fig. 4B and fig. S12). At lower propylene pressures, the water dependence showed more linear approximately first-order dependence in the entire water concentration range. For propylene, the rate showed first-order dependence on propylene activity at lower propylene pressures, but the dependence decreased at elevated propylene pressures (Fig. 4C and fig. S13). To address the complexity arising from noninteger reaction orders and variations in orders with reactant concentrations, we implemented a multiparameter kinetic model and fitted it to the kinetic data collected over a wide range of experimental conditions. We initially hypothesized that surface-adsorbed hydroperoxo (OOH*), peroxo (OO*), or oxo (O*) species might serve as the reactive oxygen species for epoxidation. We then proposed various elementary steps and corresponding rates for each hypothesis (supplementary text S1). Given that acetonitrile is a less-effective proton acceptor than water, we suggested elementary steps in which two distinct water molecules play explicit roles in the reaction, one as a reactant and the other as a proton acceptor. Then, we employed kinetic modeling to deduce which intermediate would most likely react with propylene. This approach was viable because the reaction of different species (OOH*, OO*, and O*) with propylene generates distinct surface intermediates after epoxidation (OH*, O*, and *, respectively). By contrast, the evolution of O2 invariably leaves vacant (*) surface sites. The goodness-of-fit of each model was evaluated by the root mean square error (RMSE) between the experimental data and the simulated data from the model, with the lowest value indicating the best-fit model. Our experimental data best matched the mechanism suggesting OO* as the key intermediate for epoxidation, with an RMSE of

g

probe substrate analysis, and substrate-based descriptor assessment. Electrochemical kinetic rate measurements were conducted to examine electrolyte species dependences for direct anodic epoxidation on PdPtOx/C (Fig. 4 and figs. S12 and S13). The reaction rates, represented by the partial current density toward epoxidation, were measured with chronoamperometry in conjunction with product quantification performed with proton nuclear magnetic resonance (1H-NMR) (materials and methods). We hypothesized that the direct alkeneepoxidation pathway at the anode may involve multiple coupled proton-electron transfer steps from water to the electrode, followed by oxygenatom transfer from intermediates typically seen in water oxidation to propylene. O2 was observed to be the major byproduct with the PdPtOx/C catalyst, although a small amount of CO2 was also observed (fig. S14). It is worth noting that the selectivity of PO over other propylene-derived products (PG, acetone, and CO2) reaches 92%. Therefore, in our proposed mechanism, only the competition between epoxidation and oxygen evolution was considered. In cases in which the surface coverage is primarily determined by a single reactive intermediate and a rate-limiting step is evident, the measured Tafel slope of 119 ± 5 mV/decade implies that the rate-limiting step involves a single electron transfer and that no pre-equilibrated electron transfer steps occur between the abundant reactive intermediate and the rate-limiting step (Fig. 4A). Nevertheless, noninteger and nonlinear reaction orders were observed in the water and propylene dependences, signaling the presence of more intricate kinetic control. At 1 atm of propylene, the observed dependence of rates on water molarity exhibited approximately second-order behavior at lower water ratio, which decreased to approximately

p

Fig. 3. Operando x-ray absorption spectroscopy. (A) Changes of Pt average valency in PdPtOx/C and PtOx/C in responses to applied potentials. (B) Changes of Pd average valency in PdPtOx/C and PdOx/C in responses to applied potentials. Potentials in this figure were 100% iR-compensated (i, current; R, resistance).

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

p g y

Additional insights into direct anodic epoxidation on PdPtOx/C were obtained from evaluating the correlation between the electrophilicity of alkene substrates and the epoxidation rate. Previous studies on thermochemical epoxidation suggest that either electrophilic (27–29) or nucleophilic (30, 31) epoxidation can be a reasonable reaction pathway for electrochemical propylene epoxidation on late-transition metal catalysts (supplementary text S2). By understanding the electronic nature of the oxygen transfer during the reaction, we can enhance our understanding of the mechanism and guide future catalyst development. Chung et al., Science 383, 49–55 (2024)

5 January 2024

kenes with various structures and functional groups were selected, with w− values ranging from 12 to 44 kJ/mol. To ensure consistent reactant chemical potential across different substrates, the gaseous substrates were fed until saturation, and the liquid substrates were added to the electrolyte near the phase-separation point, corresponding to ~0.2 M substrate for a 0.11 M tetrabutylammonium tetrafluoroborate (TBABF4), 10 M water in acetonitrile (table S11). The substrates were tested at the same potential referenced to the ferrocene redox couple (1.05 V versus Fc/Fc+). A negative correlation between the logarithm of rate and the alkene electrophilicity was observed, with more nucleophilic alkenes displaying higher reactivity toward epoxidation (Fig. 5). The results suggest that the epoxidation may proceed through an electrophilic 5 of 7

,

Correlation between alkene electrophilicity and epoxidation rate

The global electrophilicity index (w) can be calculated by taking the square of alkene chemical potential divided by chemical hardness and is expressed in units of energy (32, 33). A higher value of w indicates that the molecule is more electrophilic. Furthermore, the Fukui functions can be used to determine the local reactivity of each atom in a molecule by predicting the redistribution of electron density during electrophilic, nucleophilic, and radical attacks. The local electrophilicity (w−) at each atom center can be obtained by multiplying the global electrophilicity by the value of the Fukui function for the removal of an electron. The average w− of the vinyl carbons as a substrate-based descriptor was assessed against the epoxidation rates of a wide range of alkene substrates. A total of 11 distinct aliphatic al-

y

PdPtOx/C occurs through a stepwise mechanism (fig. S16) with OO* reactive intermediates.

including the direct anodic-epoxidation and oxygen-evolution pathways. (E) Simulated propylene partial-pressure dependences of intermediate species coverage at 10 M water, 1.1 V versus Fc/Fc+ condition. Potentials in this figure were 100% iRcompensated (i, current; R, resistance). Dotted lines in (A) through (C) represent best fit from kinetic model of proposed mechanism in (D) (more details are shown in supplementary text S1, Case 2d). Vertical error bars represent standard deviations from the mean of multiple replicates (n = 3) of the same experiment.

y g

Fig. 4. Kinetic data on direct anodic epoxidation with PdPtOx/C catalysts. (A) Anodic potential dependence of current density at 1 atm propylene in a 10 M water, 0.4 M tetrabutylammonium tetrafluoroborate (TBABF4) acetonitrile solution. ipo+pg, partial current density toward PO and PG. (B) Water molar concentration dependences of propylene-epoxidation current density at different propylene partial pressures. (C) Propylene partial-pressure dependences of propylene-epoxidation current density at different water concentrations. (D) A putative reaction mechanism

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

Fig. 5. Correlation between the alkene electrophilicity index and the direct anodicepoxidation rate. Pearson correlation coefficient (r) value = −0.85.

,

6 of 7

y

5 January 2024

y g

Chung et al., Science 383, 49–55 (2024)

y

The successful operation of direct epoxidation through water activation hinges on whether the electrolyte system is compatible with a particular combination of catalyst and alkene substrate. Water-acetonitrile blended electrolytes were used to investigate the epoxidation of various alkene substrates because the acetonitrile cosolvent enhances the solubility of hydrophobic alkene substrates in the water-rich phase. However, the use of an organic solvent can lead to an increase in the ohmic resistance throughout the electrolyte (fig. S4). The use of water alone as the solvent for the electrolyte, without an organic cosolvent, would be ideal because water is required as a reactant and can effectively solvate electrolyte ions, minimizing the ohmic resistance of the overall electrolyte. For gaseous substrates that are sparingly soluble in water, such as propylene, gas-diffusion electrodes (GDEs) could be used to reduce the distance that reactant gases must diffuse through the electrolyte to reach the catalyst, and therefore, aqueous electrolytes would be sufficient for epoxide production. Indeed, the PdPtOx/C catalyst showed ~60% FE toward propylene epoxidation at 1.65 V versus standard hydrogen electrode (SHE) in a sodium phosphate aqueous electrolyte when incorporated into a carbon-paper GDE with Nafion binder (Fig. 1E and fig. S4). Adding the ionomer onto the electrode was critical to achieve epoxidation performance commensurate with

g

Electrolyte choice and the effect of solvent medium on the reaction kinetics

water-acetonitrile electrolytes because the hydrophobic backbone of the ionomer generates partially wetted channels that are effective for transport of the reactant gases to the catalyst. The microenvironment of electrolytes also affects the conversion of epoxides into glycols. A noticeable amount of PG was observed in the aqueous propylene epoxidation (~10% FE), higher than that observed in the blended wateracetonitrile electrolyte. Furthermore, using 0.1 M HClO4 or any acidic electrolyte results in the exclusive formation of PG, with no detected PO (table S1). Because epoxides are known to undergo acid-catalyzed hydrolysis to afford glycols, it is imperative to find a pH regime in which hydrolysis is minimized to ensure high selectivity toward the product of interest in aqueous electrolytes. The reaction rates of PO to PG in aqueous electrolytes with pH 4 to 7 were measured with 1H-NMR: The hydrolysis of PO to generate PG was found to be faster in the pH 4 buffer solution, with almost 40% consumption of PO after 7 hours, whereas the hydrolysis in less-acidic solutions, at pH 5 to 7, was largely suppressed, with less than around 10% PO to PG conversion (fig. S18). Thus, maintaining a near-neutral pH in the aqueous electrolyte is essential to optimize PO production. The practical industrial operation of this system will likely require the use of a membraneelectrode assembly (MEA) resembling a water electrolyzer with humidified propylene gas fed to the anode. The MEA is advantageous for minimizing ohmic loss and separating PO from liquid electrolytes. However, our analysis on the influence of pH on the PO hydrolysis rate suggests that minimizing the interfacial pH gradient at the anode would be the key challenge for maximizing PO production over PG in MEAs operated at high current densities. As long as the issues of mass-transport limitation and hydrolysis of PO are addressed, mechanistic insights on propylene epoxidation

p

attack by oxygen species and that the energy barriers to this attack are lower for more nucleophilic alkenes. Furthermore, the correlation suggests that a frontier-orbital interaction between the lowest unoccupied molecular orbital (LUMO) of the oxygen species and the highest occupied molecular orbital (HOMO) of the alkene modulates the reactivity.

obtained in water-acetonitrile electrolyte may be extended to aqueous systems. The investigation on water activity in water-acetonitrile electrolyte suggests that the aqueous electrolyte may be viewed as an extreme case of the wateracetonitrile electrolyte with the water molar ratio approaching unity. In the water-acetonitrile electrolyte system, positive deviation from Raoult’s law (fig. S17) indicates that the water in this electrolyte prefers self-interactions over mixing, meaning that the activity coefficient of water in this case is greater than unity (gwater > 1) and that it approaches unity with increasing water concentration. The comparison between the rate order with respect to water activity and water concentration provides information about the activity coefficient of the transition state (g‡) as well (supplementary text S4). At lower water concentrations (1 to 5 M), the propylene-epoxidation rate showed approximately fourth-order dependence on water activity, whereas approximately second-order dependence on water molar ratio was observed. The second-order dependence is consistent with our hypothesis that the predominant rate-determining step leading to the formation of OOH* involves a first water molecule that is a reactant and a second water molecule that is a proton acceptor. The abnormally high apparent fourth-order dependence on water activity suggests that the stabilization of the transition state may be achieved by augmenting the water content. However, approximately second-order dependences in both water activity and concentration were observed at higher water concentrations (5 to 12 M). This can be rationalized if g‡ scales with gwater at low water concentrations but remains relatively unchanged at higher water concentrations. This hypothesis aligns with our chemical intuition: Once a sufficient quantity of water is present around the transition state, adding more water does not substantially alter the near–transition state chemical environment. Therefore, once the water concentration exceeds 5 M in acetonitrile, the chemical environment at the active site should be akin to that of an aqueous electrolyte. In this case, the transition from wateracetonitrile electrolyte to aqueous electrolyte may be a simple extrapolation of water molar ratio to one, without a discontinuous change in the solvent properties. This interpretation is supported by the measured reaction rates in aqueous electrolytes that is consistent with the predictions from the mechanistic model derived from kinetic studies in water-acetonitrile electrolytes (fig. S12). Collectively, the results from multiple orthogonal analyses provide a comprehensive understanding of the mechanism of electrochemical epoxidation on PdPtOx/C. Our findings are consistent with a mechanistic hypothesis that OO* is the key species involved in the electrophilic epoxidation of propylene. The PdO

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

RE FE RENCES AND N OT ES

33. R. Pal, P. K. Chattaraj, J. Comput. Chem. 44, 278–297 (2023). 34. M. Chung et al., Direct propylene epoxidation via water activation over Pd-Pt electrocatalysts, Version 3, Dryad (2024); https://doi.org/10.5061/dryad.s7h44j1dd. AC KNOWLED GME NTS

We are grateful to S. Ehrlich, L. Ma, E. Stavitski, and D. Leshchev for their help in operating XAS experiments. We also appreciate A. Penn for taking high-resolution energy dispersive spectroscopy mapping of catalysts. Funding: This research was supported by the US Department of Energy (DOE) Office of Science, Office of Basic Energy Sciences, Catalysis Science Program, award no. DE-SC0023207. K.M. gratefully acknowledges the support of the Sloan Foundation. This research used resources of the National Synchrotron Light Source II, a US DOE Office of Science User Facility operated by Brookhaven National Laboratory under contract no. DE-SC0012704. Author contributions: Conceptualization: M.C. and K.M. Methodology: M.C. Investigation: M.C., J.H.M., and J.S.A. Validation: M.C. and J.S.A. Visualization: M.C. Funding acquisition: K.M. Project administration: M.C. and K.M. Supervision: K.M. and Y.R.-L. Writing – original draft: M.C. Writing – review and editing: M.C., J.H.M., J.S.A., C.J., Y.R.-L., and K.M. Competing interests: M.C. and K.M. have filed a provisional patent application (no. 63/544,664) regarding the design of catalysts and electrodes. Data and materials availability: There are no restrictions on materials. All data needed to evaluate the conclusions in the paper are present in the paper or the supplementary materials. Tabular data underlying the figures are deposited in the Dryad data repository (34). License information: Copyright © 2024 the authors, some rights reserved; exclusive licensee American Association for the Advancement of Science. No claim to original US government works. https://www. science.org/about/science-licenses-journal-article-reuse

g

SUPPLEMENTARY MATERIALS

science.org/doi/10.1126/science.adh4355 Materials and Methods Supplementary Text Figs. S1 to S25 Tables S1 to S12 References (35–39)

y

1. M. W. Allsopp, G. Vianello, Ullmann’s Encyclopedia of Industrial Chemistry – Wiley Online Library (Wiley, 2012); https:// onlinelibrary.wiley.com/doi/book/10.1002/14356007. 2. S. J. Khatib, S. T. Oyama, Catal. Rev. Sci. Eng. 57, 306–344 (2015). 3. J. A. M. Leduc, “Electrochemical Process for the Production of Organic Oxides,” US Patent 3,288,692 (1966). 4. W. R. Leow et al., Science 368, 1228–1233 (2020). 5. M. Chung, K. Jin, J. S. Zeng, K. Manthiram, ACS Catal. 10, 14015–14023 (2020). 6. J. A. M. Le Duc, “Electrolytic Production of Olefine Oxides,” US Patent 3,427,235 (1969).

7. M. Chung, K. Jin, J. S. Zeng, T. N. Ton, K. Manthiram, J. Am. Chem. Soc. 144, 17416–17422 (2022). 8. K. Jin et al., J. Am. Chem. Soc. 141, 6413–6418 (2019). 9. X. C. Liu et al., J. Am. Chem. Soc. 144, 20895–20902 (2022). 10. A. Winiwarter et al., Energy Environ. Sci. 12, 1055–1067 (2019). 11. R. P. H. Jong, E. Dubbelman, G. Mul, J. Catal. 416, 18–28 (2022). 12. S. Koroidov et al., Catal. Sci. Technol. 11, 3347–3352 (2021). 13. K. Otsuka, T. Ushiyama, I. Yamanaka, K. Ebitani, J. Catal. 157, 450–460 (1995). 14. S. Iguchi, M. Kataoka, R. Hoshino, I. Yamanaka, Catal. Sci. Technol. 12, 469–473 (2022). 15. I. Yamanaka, K. Sato, K. Otsuka, Electrochem. Solid-State Lett. 2, 131–132 (1999). 16. L. L. Holbrook, H. Wise, J. Catal. 38, 294–298 (1975). 17. J. Ke et al., Nat. Commun. 13, 932 (2022). 18. H. Li, C. S. Abraham, M. Anand, A. Cao, J. K. Nørskov, J. Phys. Chem. Lett. 13, 2057–2063 (2022). 19. M. Chen, L. D. Schmidt, J. Catal. 56, 198–218 (1979). 20. C. Carrillo et al., J. Phys. Chem. Lett. 5, 2089–2093 (2014). 21. P. N. Plessow, F. Abild-Pedersen, ACS Catal. 6, 7098–7108 (2016). 22. E. D. Goodman et al., J. Chem. Phys. 151, 154703 (2019). 23. A. Aitbekova et al., Nat. Mater. 21, 1290–1297 (2022). 24. D. Gao et al., J. Am. Chem. Soc. 137, 4288–4291 (2015). 25. D. T. Bregante, P. Priyadarshini, D. W. Flaherty, J. Catal. 348, 75–89 (2017). 26. C. K. Sams, K. A. Jørgensen, S. V. Lindeman, J. Songstad, Acta Chem. Scand. 49, 839–847 (1995). 27. C. Kim, T. G. Traylor, C. L. Perrin, J. Am. Chem. Soc. 120, 9513–9516 (1998). 28. X. Yang et al., Angew. Chem. Int. Ed. 54, 11946–11951 (2015). 29. Y. Lei et al., Science 328, 224–228 (2010). 30. A. R. Chianese, S. J. Lee, M. R. Gagné, Angew. Chem. Int. Ed. 46, 4042–4059 (2007). 31. A. Zanardo, F. Pinna, R. A. Michelin, G. Strukul, Inorg. Chem. 27, 1966–1973 (1988). 32. F. De Vleeschouwer, V. Van Speybroeck, M. Waroquier, P. Geerlings, F. De Proft, Org. Lett. 9, 2721–2724 (2007).

p

crystal structure embedding Pt(II) species was found to be effective for catalyzing direct anodic epoxidation, suggesting that other materials with similar structures and stronger inductive effects on Pt may also produce effective catalysts by promoting more electrophilic epoxidation pathways. Additionally, we discovered through fundamental kinetic analysis that the identity of reactive oxygen species may vary between catalysts. For instance, the proposed reactive oxygen species for epoxidation on PdPtOx/C and MnOx-based catalysts (7, 8) are OO* and O*, respectively. Developing approaches to stabilize OO* may enhance epoxidation on noble metals while avoiding oxygen evolution. Our work contributes toward advancing the sustainable synthesis of epoxides, which currently have substantial energy and environmental footprints. The molecular-level understanding of direct anodic epoxidation obtained from this study may have implications for other electrocatalytic oxygen-atom transfer reactions driven by water-activation intermediates.

Submitted 4 March 2023; resubmitted 12 September 2023 Accepted 29 November 2023 10.1126/science.adh4355

y g y ,

Chung et al., Science 383, 49–55 (2024)

5 January 2024

7 of 7

RES EARCH

NEUROSCIENCE

Balancing risk-return decisions by manipulating the mesofrontal circuits in primates Ryo Sasaki1*, Yasumi Ohta2, Hirotaka Onoe3, Reona Yamaguchi4, Takeshi Miyamoto1,5, Takashi Tokuda6, Yuki Tamaki1†, Kaoru Isa1, Jun Takahashi7, Kenta Kobayashi8, Jun Ohta2, Tadashi Isa1,3,4* Decision-making is always coupled with some level of risk, with more pathological forms of risk-taking decisions manifesting as gambling disorders. In macaque monkeys trained in a high risk–high return (HH) versus low risk–low return (LL) choice task, we found that the reversible pharmacological inactivation of ventral Brodmann area 6 (area 6V) impaired the risk dependency of decision-making. Selective optogenetic activation of the mesofrontal pathway from the ventral tegmental area (VTA) to the ventral aspect of 6V resulted in stronger preference for HH, whereas activation of the pathway from the VTA to the dorsal aspect of 6V led to LL preference. Finally, computational decoding captured the modulations of behavioral preference. Our results suggest that VTA inputs to area 6V determine the decision balance between HH and LL.

1 of 7

,

5 January 2024

Because reward information is critical for riskdependent decision-making, we next investigated the role of input from the ventral tegmental area (VTA)—a core area of the midbrain dopaminergic system (26), which projects to the superficial and deeper layers of area 6VV (Fig. 3E,

y

Sasaki et al., Science 383, 55–61 (2024)

Optogenetic activation of the mesofrontal pathway in risk-return decision-making

y g

*Corresponding author. Email: [email protected] (R.S.); [email protected] (T.I.) †Present address: Japanese Red Cross Otsu Hospital, Otsu-shi, Shiga 520-0046, Japan.

Six macaque monkeys were trained to perform a cue-target choice task (Fig. 1A). While a monkey fixated on a central point, we presented two colored cues (selected from either 16 or 25 possible colors, respectively) in the left and right hemifields. We assigned different probabilities and EVs of water reward delivery to the different colors (Fig. 1B). The colors corresponded to different reward probabilities with the same EV along the horizontal axis and changed in EVs systematically along the vertical axis [four (n = 2 monkeys) or five (n = 4 monkeys) levels between 100 and 250 ml]. As an example, when a red cue is selected, the reward would be given in only 10% of the trials with a reward size of 1000 ml, corresponding to HH choice, whereas the blue cue would be rewarded in 90% of the trials with a reward size of 111.1 ml, representing a LL choice. In both cases, the resulting net EV of the reward would be the same (100 ml). Thus, this task allows us to clarify the decisional balance between HH and LL preference with fixed net EV. In a two-dimensional (2D) heatmap representing the proportion chosen, which is computed based on the frequency of choosing a given option against all other options in the matrix, all six monkeys (monkeys Y, S, M, J, C, and H) showed a preference for the left-uppermost corner of the matrix, corresponding to the HH choice with the largest EV (Fig. 1C). The line plot

y

Division of Physiology and Neurobiology, Department of Neuroscience, Graduate School of Medicine, Kyoto University, Kyoto-shi, Kyoto 606-8501, Japan. 2Division of Materials Science, Graduate School of Science and Technology, Nara Institute of Science and Technology, Ikoma-shi, Nara 630-0192, Japan. 3Human Brain Research Center, Graduate School of Medicine, Kyoto University, Kyoto-shi, Kyoto 606-8507, Japan. 4Institute for the Advanced Study of Human Biology (WPI-ASHBi), Kyoto University, Kyoto-shi, Kyoto 606-8501, Japan. 5Japan Society for the Promotion of Science, Chiyoda-Ku, Tokyo 102-0083, Japan. 6Institute of Innovative Research, Tokyo Institute of Technology, Meguro-Ku, Tokyo 152-8550, Japan. 7 Department of Clinical Application, Center for iPS Cell Research and Application, Kyoto University, Kyoto-shi, Kyoto 606-8507, Japan. 8Section of Viral Vector Development, National Institute for Physiological Sciences, Okazaki-shi, Aichi 444-8585, Japan.

Identification of frontal brain areas regulating risk-dependent decision-making

g

1

have also been reported in nonhuman primates (16, 17), this again could not clearly dissociate risk levels and EVs. Because decision-making is usually linked to more variable combinations of risk levels and EVs, we investigated the brain regions involved in risk-dependent decision-making using macaque monkeys trained to perform a HH versus LL decision task with a variable combination of reward probabilities and EVs.

p

T

he real world is full of uncertainty, and our decisions are always accompanied by some level of risk. Decision-making between high risk–high return (HH) and low risk–low return (LL) choices is regulated by the subject’s utility function (1, 2), which varies substantially from person to person. Pathological risk-taking decisions are known to underlie problematic behaviors, such as gambling disorders (3–6). Dysregulation of the reward circuitry, including the ventral striatum and the orbitofrontal and medial prefrontal cortices, is implicated in gambling disorders (7–12); however, their causal role is still elusive owing to technical limitations of human studies. Risky choice behavior has also been investigated in psychopharmacological studies in rats (13, 14). In humans and rodents, riskdependent decisions are frequently assessed using variants of the Iowa Gambling Task (15). However, this task cannot uncouple HH and LL choices from the negative and positive expected values (EVs), respectively. Furthermore, although preferences for risky choices

above the matrices indicates the risk-dependent choices of the monkeys at consistent EVs with a linear regression, showing that the monkeys prefer HH options without any EV dependence. The HH preference was also statistically not significantly different depending on the EV blocks [paired t test, t(5) = 2.021, P = 0.099]. We also estimated each monkey’s utility and probability weighting functions from their choice using standard parameters described in the literature (18–20), which suggested the risk attitude in decision-making (fig. S1). This was common in all of the monkeys, which fits with the framework of utility function in monkeys (2) and prospect theory of decision-making in humans (21). To identify the frontal brain areas involved in processing risk information, we injected a g-aminobutyric acid type A (GABAA) receptor agonist, muscimol (0.2 to 0.5 ml, 5 mg/ml), into a variety of bilateral symmetric prefrontal regions, including ventral Brodmann area 6 (area 6V) and area 12 (Fig. 2, A to C), which has previously been suggested to play a role in coding mental operation (22). Neurons in this area have also been associated with probability of reward in decision-making (23) [described as the ventrolateral prefrontal cortex (vlPFC)], which suggests that area 6V might be a potential candidate for the representation of complex risk-return computation. Muscimol injection into the ventral aspect of area 6V (6VV) (Fig. 3C) eliminated the risk dependency of decisionmaking, which recovered the next day (Fig. 2E) [compare the difference of slope (“post pre”) to HH-LL between muscimol and saline injection in six experiments in two animals: monkey S, Wilcoxon rank-sum test, P = 0.026, and monkey M, Wilcoxon rank-sum test, P = 0.016]. Area 6VV is known for its roles in representing motor actions in many studies. Additionally, a few studies (22, 24, 25) have suggested its role in coding for human decision-making. The EV dependency of decision-making was not affected by area 6VV inactivation (Fig. 2F; Wilcoxon rank-sum test, P > 0.05). We observed no effect from other candidate areas—muscimol injections into the orbitofrontal cortex [OFC (area 14)] (four sessions) and the dorsal anterior cingulate cortex (dACC) (four sessions) had no effect (Fig. 2, E and F; Wilcoxon ranksum test, P > 0.05). These results suggest that the area 6VV plays an important role for regulating HH-LL decisions, whereas the OFC and dACC play no critical role in such complex computations.

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

C

A

1.0

Monkey Y

Monkey C

0.5

1.0 0.5

Fixation point Target Reward cue (w/ or wo/) -1000

-500 0 500 1000 Time from target onset (ms)

LL HH

Probability x Reward = Expected size value : 90% 111.1 μL 100 μL : 10% 1000.0 μL 100 μL

0.0

250.0

212.5

212.5

175.0

175.0

137.5

137.5

100.0 1.0

0.5

0.0

100.0 10

Proportion chosen

Monkey H

Reward

0.0

250.0

No-reward

30

50

70

90

1.0

0.5

0.0

1.0

0.0

0.0

250.0

250.0

212.5

212.5

175.0

175.0

137.5

137.5

2500

Monkey M

1500

212.5

30

50

70

90

Probability (%) 1.0

1.0

0.5

0.0

30

50

70

90

Probability (%) 1.0

Monkey S

0.5

0.0

0.0

250.0

250.0

200.0

200.0

150.0

150.0

g

1000

10

Proportion chosen

0.5

2000 175.0

90

100.0 10

Proportion chosen 250.0

70

p

B

0.0

50

0.5

100.0 0.5

30

Probability (%) 1.0

Monkey J

0.5

1.0

10

Proportion chosen

Probability (%)

137.5 500 100.0

10

30

50

70

90

LL

HH-LL based on Probability (%)

10

30

50

70

90

y

HH

0

Probability (%) 100.0

100.0 1.0

0.5

0.0

Proportion chosen

20

40 60 80 Probability (%)

1.0

0.5

0.0

20

Proportion chosen 0

40 60 80 Probability (%)

0.2

0.4

0.6

0.8

1

Proportion chosen

5 January 2024

P < 0.05; Fig. 3J). We stimulated the VTA-6VV, 6VD, and/or vlPFC pathways optogenetically with red LEDs right after target onset (Fig. 3B; three repetitions of 20 ms ON and 80 ms OFF, 625-nm wavelength; intensity: 0.782 mW at 2 mA, 1.824 mW at 4 mA). Saccades were initiated at 200 to 500 ms after target presentation. In these experiments, one of the cues was always fixed at 50% reward probability and 175 ml EV (Fig. 3A, right), and another cue was selected from the 25 color possibilities (Fig. 3A, left). During these sessions, we divided the trials randomly into 50% stimulated (LED-ON) and 50% unstimulated (LED-OFF) trials. In monkey J (Fig. 3M, top), photostimulation of the VTA-6VV pathway, delivered at 2 mA during epoch II (0 to 220 ms after cue presentation) led to a shift in decision-making to the HH-preferring mode, without affecting EV dependency (Fig. 2 of 7

,

Sasaki et al., Science 383, 55–61 (2024)

the injection site in the VTA of two monkeys (J and H), we observed that tdTomato was expressed in the majority of VTA neurons around the injection site, with 78% (776/1000) and 66% (570/860) in monkeys J and H, respectively (Fig. 3, E to H). These neurons were tyrosine hydroxylase (TH)–positive, which suggests that most of the activated neurons were dopaminergic. Next, to determine whether the neural activities in 6VV, 6VD, and/or vlPFC were modulated on the basis of HH or LL choices, we recorded their activity with the ECoG electrodes during the HH-LL choice task. Cue-related a-band (8 to 13 Hz) activity in all channels at 100 to 500 ms after cue presentation (Fig. 3I), when the monkey presumably made its decision, was higher for HH choices (10% reward probability) compared with LL choices (90% reward probability) (Wilcoxon rank-sum test,

y

right) (27). We expressed a red light–activatable opsin, ChrimsonR (28), in the bilateral VTA using an adeno-associated viral vector (AAV2.1-SynChrimsonR-tdTomato) (fig. S2) and selectively stimulated the combination of light-emitting diode (LED) channels belonging to the VTA6VV pathway, the VTA–dorsal aspect of area 6V (6VD) pathway, or the VTA-vlPFC pathway coupled with electrocorticogram (ECoG) electrodes (29) that were chronically implanted on the surface of the bilateral target areas (Fig. 3, C and D, and fig. S3) in monkeys J, C, Y, and H. We used a nonselective promotor (Syn) to express ChrimsonR because we could not see the axon terminals in the frontal cortex using immunohistochemistry staining in a pilot experiment using a dopamine neuron–specific promotor, which likely suggests weak expression with the specific vector. Upon confirming expression at

computed by all comparison against other options, are shown as a color contour map for each of the six monkeys. The scatter plots show the proportion of choices for reward probabilities (top) computed by the consistent EV, and the proportion of choices for EVs (left) computed by the consistent probability. The lines in the scatter plots represent their linear regressions.

y g

Fig. 1. Task design and behavior. (A) The monkeys were required to choose one of two cues by saccade to receive a certain amount of water reward. (B) Assignment of the reward probabilities and EVs for each cue color (left) and the relationship between the reward probabilities and the reward sizes (right). (C) Proportions of choosing each combination of reward probability and EV,

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

A

B

I II III IV

6VV

C

I

II

III

IV

dACC

Area 6VV

D

1.0

OFC

Pre

30 min

60

130

200

24h

0.5

10 30 50 70 90

Probability (%)

E 1.0

Monkey S Monkey M

0.5

1.0

0.5

Accumulation of stimulus effects on risk-dependent behavioral choices

0.0

OFC Muscimol

dACC Muscimol

-1.0

-0.5

Area 6VV Muscimol Saline

OFC Muscimol

dACC Muscimol

-1.0

5 January 2024

3 of 7

,

Sasaki et al., Science 383, 55–61 (2024)

epochs I, III, and IV): the period before cue onset (epoch I, −500 to −280 ms), after cue onset (epoch III, 500 to 720 ms), and at reward delivery (epoch IV, 1000 to 1220 ms). We observed no change in the risk- and EV dependency of decision-making when photostimulation was applied during epochs I, III, and IV (fig. S4, A to G; Wilcoxon signed-rank test, P > 0.05). Thus, optogenetic stimulation of VTA terminals in the 6VV resulted in modification of riskdependent decisions (but not EV dependent decisions) to HH-preferring only during epoch II, when the monkeys were presumably making decisions to select a cue. By contrast, photostimulation of area 6VD relieved the HH-preferring mode (Fig. 3, N and P, top). The decision changed to LL-preferring

y

3M, bottom). Analysis of the population data (Fig. 3O, top), whereby the preference for HH was evaluated by a threshold for HH-preferring decisions in the psychometric function, showed that the preference for HH was higher during the LED-ON trials compared with the LED-OFF trials (Wilcoxon signed-rank test, z = −3.89, P = 1.01 × 10−4), whereas EV dependency was not affected (Fig. 3O, bottom; z = 0.26, P = 0.79). Furthermore, the cue-related a-band response in channels located in area 6VV was enhanced during the LED-ON trials compared with the LED-OFF trials [Fig. 3K; (LED-ON, 3213 trials; LED-OFF, 3203 trials) Wilcoxon rank-sum test, P < 0.05]. To understand the temporal dynamics of this response, we also applied the stimulation during various time epochs (fig. S4A;

y g

Fig. 2. Effects of muscimol injections into frontal cortical regions on risky choice decisions. (A) Surface view of the macaque cortex showing the rostro-caudal levels of the planes in (C) and the approximate location of area 6VV (yellow). (B) MRI images of the coronal plane from monkey S, including area 6VV. The arrowheads indicate the Gd3+ injection site. (C) Effective muscimol injection sites in area 6VV (purple-filled circles). The results from two animals are superimposed. (D) Proportion chosen for a single session (monkey S) plotted against the reward probability computed by the consistent EV at each time window from injection (preinjection, 0 to 30 min, 30 to 60 min, 60 to 130 min, 130 to 200 min, and 24 hours). (E) The difference of slope to reward probability (HH-LL) between before and after injection at each injection site [OFC (area 14) and dACC] in two monkeys. (F) The same arrangement as (E) for the difference of slope to the EV. Plots with an asterisk exhibit a statistically significant effect (Wilcoxon rank-sum test, *P < 0.05).

The photostimulation-dependent effects on choice behavior accumulated over time. In monkey J, after six control sessions without stimulation, we introduced photostimulation at 6VV or 6VD, randomly interleaved (Fig. 4A). Thereafter, the HH-preference indices for both the LED-ON and LED-OFF trials increased gradually (lower threshold of the psychometric function for HH preference). During the initial six photostimulation sessions, HH preference was higher in the LED-ON trials compared with the LED-OFF trials, but after the seventh session, HH preference became saturated, and HH preference in the LED-ON and LED-OFF trials became nearly equal. This result implies that the HH-preferring mode might accumulate more easily compared with the LL-preferring mode. To clearly distinguish the stimulation effect between 6VV and 6VD, we stimulated each area in a block task design for the second monkey (monkey H; Fig. 4B). The preference for HH became higher in the LED-ON trials compared with the LED-OFF trials after three sessions (sessions 23 to 25). The effect became saturated at session 40, and stimulation was stopped for nine sessions (sessions 41 to 49). Upon restarting stimulation, the HH preference reappeared quickly (sessions 49 to 50) before it became saturated again (Fig. 4B, magenta line). We subsequently began photostimulation at 6VD, whereby HH preference clearly dissipated over time (Fig. 4B, cyan line). Similar behavioral effects of 6VD stimulation were also observed in the third monkey (C), where HH preference was relieved over time and saturated (Fig. 4C, cyan line).

y

Area 6VV Muscimol Saline

g

*

*

0.0

-0.5

F

p

0.0

during the LED-ON trials rather than during the LED-OFF trials (Wilcoxon signed-rank test, z = 4.55, P = 5.33 × 10−6) while not affecting the EV dependency (Fig. 3P, bottom; z = −0.57, P = 0.56). Additionally, the cue-related a-band responses in area 6VD were inhibited during the LED-ON trials compared with the LED-OFF trials (Fig. 3L; Wilcoxon rank-sum test, P < 0.05), hence having an opposite effect compared with photostimulation at area 6VV. It is worth emphasizing that more-localized photostimulation still convincingly enhanced or relieved the HH-preferring mode (fig. S5). Photostimulation primarily at vlPFC caused no effect in both risk dependency (Fig. 3Q, top; z = 0.10, P = 0.91) and EV dependency (Fig. 3Q, bottom; z = 0.61, P = 0.61). Behavioral threshold and brain activities in the control animal (monkey Y), who was injected with the control vector carrying no ChrimsonR sequence (AAV2.1-SyntdTomato), were not affected (fig. S6). Notably, we could largely exclude that these effects were a result of side effects of heat generated from photostimulation because the temperature changes induced by LED stimulation were 0.05; ***, P < 0.001; ****, P < 0.0001 versus WildR ISO, Dunnett’s multiple comparisons test. Data presented as mean ± SEM. (B) Hematoxylin and eosin (H&E)-stained sections of cecal tissues 15 days after starting ICB treatment, and histology scores. Scale bars, 200 mm. White arrowheads point to epithelial damage, black arrowheads point to regenerating epithelia, the gray arrow indicates a crypt abscess, and open arrows indicate inflammatory infiltrates. Each point represents Lo et al., Science 383, 62–70 (2024)

5 January 2024

ag R

BL /

12

57

6 9 Days

C

0

6

IC

0

1 -/IC R B ag 1 -/IS O

B

N.S.

Rag1-/- ISO

,

N.S.

ild

R

Rag1-/- ICB Rag1-/- ISO

*

200

W

ild W

***

800

Rag1-/- ICB

C57BL/6 ICB

y

ng / g feces

D C57BL/6 ICB

Lipocalin-2

1000

y g

C

6

Histology Score

0

O

0

F

***

N.S.

Histology Score

***

4000

SPF JAX ICB

y

****

WildR ICB

WildR ISO WildR ICB SPF ICB

I JA CB X IC B

We then compared the respective contributions of antibodies targeting CTLA-4 or PD-1 in the development of ICB-driven gut inflammation. WildR microbiota mice treated with dual ICB or anti-CTLA-4 antibodies alone were similarly susceptible to intestinal inflammation as evidenced by matching fecal LCN-2 induction and disease severity in intestinal tissues (fig. S4, A and B). By contrast, WildR microbiota mice that received anti-PD-1 antibodies alone did not develop intestinal inflammation (fig. S4, A and B). These data indicate that CTLA-4 blockade is the principal driver of colitis in this microbiota-dependent model, which is consistent with human studies (3, 4). Furthermore, profiling of cecal immune cells by flow cytometry revealed the accumulation of IFNg+, IL-17+, and double positive IFNg+ IL-17+ CD4+ T helper (TH) cells 9 days after commence-

Lipocalin-2 5000

ng / g feces

CTLA-4 blockade elicits intestinal inflammation driven by CD4+ T cells and IFNg

ment of CTLA-4 blockade coinciding with high disease severity indicated previously by fecal LCN-2 levels (Fig. 2A). We also observed the induction of IFNg+ CD8+ T cells after treatment with anti-CTLA-4 antibodies (fig. S5). Consistent with the skewed IFNg response, we found robust Tbet but relatively mild RORgt expression by CD4+ T cells in the intestinal tissue, further highlighting a biased TH1 response during ICB-mediated inflammation (Fig. 2B and fig. S6). Moreover, antiCTLA-4 antibody–induced colitis was associated with an increase in multiple myeloid cell subsets including neutrophils, monocytes, macrophages, and dendritic cells in the gut (fig. S7, A and B). Intestinal regulatory T cells (Tregs) expressing Foxp3 are critical for immune homeostasis and can be categorized as thymic-derived Tregs (tTregs) or peripherally induced Tregs (pTregs) which are responsive to the gut microbiota (20, 21). In GF mice, most gut Foxp3+ Tregs display features suggestive of thymic origin whereas the colonization with intact microbiota or select bacterial species results in the induction of pTregs including a subset expressing RORgt (22–28). In assessing Foxp3+ Tregs

g

A

bodies (Fig. 1, C and D). Collectively, these data indicate that induction of colitis in mice by ICB requires both the microbiota of freeliving animals and an intact adaptive immune response.

p

indexes of individual mice indicates that the bacterial community structures of SPF JAX microbiota and WildR microbiota are dissimilar (fig. S3B). Alterations in the WildR microbiota were also observed in response to dual ICB coinciding with intestinal inflammation (fig. S3B). To compare bacterial taxa of the WildR microbiota and SPF JAX microbiota, we performed linear discriminant analysis effect size (LEfSe). We identified 132 OTUs at greater than 97% nucleotide sequence identity, which were differentially abundant in the WildR microbiota compared with SPF JAX microbiota (fig. S3C). These findings suggest that the bacterial communities of the SPF JAX microbiota and freeliving WildR microbiota are similarly diverse yet highly divergent. To determine the contributions of adaptive immunity in the present model of ICB-driven colitis, we colonized GF WT or GF Rag1−/− mice lacking lymphocytes with WildR microbiota and challenged them with dual ICB or isotype antibodies. As colitis was triggered in GF WT mice colonized with WildR microbiota by dual ICB, GF Rag1−/− mice colonized with WildR microbiota did not develop colitis after treatment with dual ICB or isotype control anti-

an individual mouse. n = 4 to 6 mice per group. Representative data of two independent experiments. N.S., P > 0.05; **, P < 0.01, Kruskal-Wallis with Dunn’s test. (C) GF C57BL/6 and GF Rag1−/− mice were colonized with WildR microbiota and treated with ICB or ISO. Colitis was assessed by fecal lipocalin-2. N.S., P > 0.05; *, P < 0.05; ***, P < 0.001 versus Rag1−/− ICB, Dunnett’s multiple comparisons test. Data presented as mean ± SEM. (D) H&E-stained sections of cecal tissues 12 days after starting ICB treatment and histology scores. Scale bars, 200 mm. White arrowheads point to epithelial damage, black arrowheads point to regenerating epithelia, and open arrows indicate inflammatory infiltrates. Each point represents an individual mouse. n = 3 to 5 mice per group. Results were confirmed using SPF mice. N.S., P > 0.05; *, P < 0.05, Kruskal-Wallis with Dunn’s test. 2 of 9

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

receptors (TCRs) (26, 28, 29). Moreover, RORgt+ pTregs are highly enriched in the large intestine but are mostly absent across systemic sites and in tumors (28). Consistent with previous studies, colonization of GF mice with WildR microbiota similarly induced RORgt+ pTregs which have higher expression of CLTA-4 compared with a Helios+ subset of tTregs (fig. S9, A and B). These data indicate that CTLA-4 blockade induces preferentially a TH1 response and a biased depletion of RORgt+ pTregs in the gut. The accumulation of CD8+ T cell subsets at mucosal surfaces is a common feature of lab-

2.81

55

4.84

37.3

25 20 15 10 5 0

IFN

Gated on CD4+ cells

0

0.5 0.0 isotype anti-CTLA-4

anti-CTLA-4 91.2 7.02

D

1.76

0.06

3 2 1

0 isotype anti-CTLA-4

20 10 0

**

54.6

30.2

42.3

13.2

***

5 4 3 2 1 0

isotype anti-CTLA-4

ROR t

E

F

IF N

ot

an

0

,

is

10

e

0

yp

10

20

ti-

20

**

30

y

5 January 2024

Histology Score

30

40

y g

****

anti-IFN

ot

40

Fig. 2. Anti-CTLA-4 antibodies elicit IFNg and CD4+ T cell-mediated colitis. (A) FACS analysis of lymphocytes isolated from cecal tissues of WildR microbiota mice 9 days after starting anti-CTLA-4 antibody or isotype treatment. Pseudocolor plots depict IFNg and IL-17 expression by CD4+ T cells and numbers of cytokine-expressing CD4+ T cells. Events displayed in flow plots: 2808, isotype and 6388, anti-CTLA-4 antibody-treated. (B) Numbers of Foxp3− CD4+ T cells expressing Tbet and RORgt isolated from cecum. (C) Pseudocolor plots of Foxp3 expression by cecal T cells and percentage expression of Foxp3 by T cells. Events displayed in flow plots: 5906, isotype and 14803, anti-CTLA-4 antibody-treated. (D) Pseudocolor plot of Helios and RORgt expression by Foxp3+ T cells and ratio of Helios+ Treg (tTreg) to RORgt+ Treg (pTreg) numbers. Events displayed in flow plots: 940, isotype and 1039, anti-CTLA-4 antibody-treated. Each dot represents an individual animal. n = 4 mice per group. Representative data of two independent experiments. **, P < 0.01; ***, P < 0.001; ****, P < 0.0001; unpaired two-tailed Lo et al., Science 383, 62–70 (2024)

isotype

n.s.

an

anti-CD8

yp e tiC D 4 an tiC D 8

anti-CD4

Histology Score

isotype

****

anti-CTLA-4

isotype

isotype anti-CTLA-4

Foxp3

25 20 15 10 5 0

Gated on Foxp3+ T cells

**

30

**

is

2.1

15.9

1.0

5 4 3 2 1 0

ROR t+

y

CD3

82

1.5

Foxp3+ % CD4+ T cells

isotype

***

2.0

Tbet+

g

C

****

Helios

94.6

2.9

0

B

IFN + IL-17A+

IL-17A+

p

IL-17A

2.56

CD4+ T Cells (x104)

isotype

+

IFN

anti-CTLA-4

oratory mouse strains colonized with the microbiota of free-living animals (16, 18). Moreover, tissue-resident CD8+ T cell subsets with cytotoxic effector features have been implicated in the pathogenesis of ICB-induced colitis in the clinic (6, 30), and we noted the induction of IFNg+ CD8+ T cells in the present mouse model (fig. S5). To clarify the involvement of CD4+ and CD8+ T cells in ICB-driven intestinal inflammation, we selectively depleted each T cell subset in WildR microbiota mice during ICB treatment (fig. S10, A and B). Anti-CD8 antibodies delayed LCN-2 induction but had

tTreg : pTreg

A

CD4+ T Cells (x104)

in the gut during ICB, we observed a reduction in the percentages of Foxp3+ T cells (Fig. 2C) and notably the selective depletion of RORgt+ pTregs whereas the relative percentages of a Helios+ subset of tTreg cells remained largely intact (Fig. 2D). The reduction of RORgt+ pTregs was confirmed in response to anti-CTLA-4 antibody treatment preceding peak inflammation (fig. S8, A and B). Previous reports demonstrate that RORgt+ pTregs selectively induced by the microbiota possess distinct immunosuppressive features including exaggerated expression of CTLA-4, IL-10, and bacteria-reactive T cell

t-test. (E) WildR microbiota mice undergoing CTLA-4 blockade were treated with isotype control antibodies or cell-depleting antibodies against CD4 or CD8. H&Estained cecal sections 12 days after starting ICB treatment and corresponding histology scores. Scale bars, 200 mm. White arrowheads point to epithelial damage, black arrowheads point to regenerating epithelia, and open arrows indicate inflammatory infiltrates. Each point represents an individual mouse. Data were combined from three independent experiments. N.S., P > 0.05; ****, P < 0.0001; Kruskal-Wallis with Dunn’s test. (F) WildR microbiota mice receiving anti-CTLA-4 antibodies were treated with neutralizing anti-IFNg antibodies or isotype control. H&E-stained cecal sections 12 days after starting ICB treatment, and histology scores. Scale bars, 200 mm. White arrowheads point to epithelial damage, the black arrowhead points to regenerating epithelium, and open arrows indicate inflammatory infiltrates. Each dot represents an individual mouse. Representative data of two independent experiments. **, P < 0.01; Mann-Whitney test. 3 of 9

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

no impact on reducing subsequent disease severity whereas mice that received antibodies that ablate CD4+ cells were protected from ICBinduced gut inflammation (Fig. 2E and fig. S11). Furthermore, animals treated with IFNg-

A

neutralizing antibodies displayed highly diminished inflammation in the gut during the ICB treatment course (Fig. 2F). We also found that IL-17A/F KO mice displayed severity of gut immunopathology similar to that of WT

B

CD4+ T cells Th2 Tfh

B-cells supCPM_2

6 Naïve T cells

4

1

+

−2

3

C

Tbet+ PD-1+ Th1

p

−4

Tbet+ PD-1- Th1

TCF-1 T-cells

9 10 11 12 13 14

0

11 Th17

8

Th22

13 5 12 8 9 2 10 7

5

2

7

14

4

6

2

Treg

1 2 3 4 5 6 7 8 9 10 11 12 13 14

3

1

4

mice in this disease model (fig. S12, A and B). Thus, intestinal inflammation triggered by CTLA-4 blockade in WildR mice is driven primarily by CD4+ T cell–mediated responses and IFNg.

−10

−5 0 supCPM_1

5

isotype

anti-CTLA-4

C5: Treg g

re

g

g

pTreg

tTre

tT

Tbet+ Treg

pT

1

reg

pT

0

2

y

tTreg

−2.5

Tbet +Treg

Tbet + T re g

supCPM_2

2.5

D re

3

g

1 2 3

5.0

−5.0

−4

0 2 supCPM_1

C1: Tbet+ PD-1- Th1

isotype

4

C3: Tbet+ PD-1+ Th1

C5: Treg

anti-CTLA-4

C5-1: pTreg

C5-2: tTreg T cells with unique TCR

anti-CTLA-4

Fig. 3. Gene expression and clonotype analyses of CD4+ T cells in ICB-induced colitis. (A) Sort-purified intestinal CD4+ T cells from WildR mice treated with isotype or anti-CTLA-4 antibodies underwent scRNA-seq analyses; each sample contained pooled cells from tissues of 3 mice per treatment group. CD4+ T cell clusters were visualized by supervised capacity preserving mapping (supCPM). (B) Circos plot showing relative contribution to each cluster identity by CD4+ T cells from each treatment. (C) Cluster 5 Treg cells were Lo et al., Science 383, 62–70 (2024)

5 January 2024

segregated and underwent further clustering and supCPM rendering and (D) circos plots showing relative abundance of each Treg subcluster by treatment group. (E) Doughnut plots showing relative abundance of clonotypes in each cluster classified by three levels of clonal expansion. CD4+ T cells with moderate clonal expansion include clonotypes with frequencies between two and four, and CD4+ T cells with marked clonal expansion include clonotypes with frequencies of five or greater. 4 of 9

,

Marked clonal expansion

y

Moderate clonal expansion

isotype

y g

E

−2

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

B WT (ICB)

***

FcR KO (ICB) FcR KO (ISO)

* N.S.

0

20 10

2.64

93

2.39

39.6

79.1

12

IFN

D

IFN

**** **** ****

80 60

20

FcR KO (ISO)

21.4

19.9

WT (ICB)

0

Foxp3 44.7

23.2

43.9

FcR KO (ICB)

57.1 Helios

Helios

22.3

WT (ICB)

30

10.9 ROR t

ROR t

5 January 2024

0

ICB

ISO

ICB

WT FcR KO

ISO

ICB

10 0

WT FcR KO

ISO

ICB

**** **** ****

4 2 0

WT FcR KO

ISO

ICB

isotype or anti-CTLA-4 antibodies. Events displayed in flow plots: 2949, WT (ISO); 2515, FcRg KO (ISO); 8060, WT (ICB); 1165, FcRg KO (ICB). (D) Numbers of cytokine-expressing T cells. Each point represents an individual mouse. Combined data from two independent experiments for each treatment group. **, P < 0.01; ***, P < 0.001; ****, P < 0.0001; Dunnett’s multiple comparisons test. (E) Pseudocolor plots of Foxp3 expression by CD4+ cells isolated from cecal tissues, and percentages of Foxp3 expression by T cells. Events displayed in flow plots: 5949, WT (ISO); 6487, FcRg KO (ISO); 6664, WT (ICB); 4870, FcRg KO (ICB). (F) Subsetting Helios+ tTregs and RORgt+ pTregs, and the ratios of tTregs to pTregs. Events displayed in flow plots: 1272, WT (ISO); 1293, FcRg KO (ISO); 567, WT (ICB); 1024, FcRg KO (ICB). Each point represents an individual mouse. Representative data from two independent experiments. ***, P < 0.001; ****, P < 0.0001; Dunnett’s multiple comparisons test.

ter visualization (31) and observed diverse immune cell populations similar to clusters reported by the original investigators (fig. S13A and table S1) (6). We noted a singular Treg cluster (cluster 7), defined by high expression levels of Foxp3, Ctla4, Il10, and Cd4, which were elevated in relative abundance in ICB-treated patients who developed colitis compared with ICB-treated patients who did not develop co-

litis and normal control samples, which is consistent with the previous report (6) (fig. S13, A and B, and table S1). We further segregated the Treg cluster (cluster 7) and identified three Treg subclusters: Treg subcluster 7-1 was marked by high expression of Helios, encoded by Ikzf2; Treg subcluster 7-2 was defined by lower expression of Ikzf2; and Treg subcluster 7-3 was characterized by high expression of Sell and 5 of 9

,

Lo et al., Science 383, 62–70 (2024)

0

y

To assess whether ICB alters Treg subset abundance in a clinical setting, we examined a comprehensive single cell RNA sequencing (scRNA-seq) data set of intestinal immune cells isolated from ICB-treated melanoma patients who developed colitis, ICB-treated melanoma patients who did not develop colitis, and control cases (6). We utilized supervised capacity preserving mapping (supCPM) for clus-

42.4

5

y g

Fig. 4. Colitis induced by anti-CTLA-4 antibodies requires Fcg receptors. (A) WT and Fcer1g−/− (FcRg KO) mice were colonized with WildR microbiota and treated with anti-CTLA-4 antibodies (ICB) or isotype control antibodies (ISO). Colitis was assessed by fecal lipocalin-2. Statistical significance between ICB-treated WT and FcRg KO mice was determined by Sidak’s multiple comparisons test and indicated by N.S., P > 0.05; *, P < 0.05; ***, P < 0.001. Data presented as mean ± SEM. (B) H&E-stained sections of cecal tissues 12 days after starting indicated treatment and histology scores. Scale bars, 200 mm. White arrowheads point to epithelial damage and open arrows indicate inflammatory infiltrates. Each point represents an individual mouse. N.S., P > 0.05; *, P < 0.05; Kruskal-Wallis with Dunn’s test. Representative data from two independent experiments. (C) Pseudocolor plots of cytokine expressing CD4+ T cells isolated from ceca of WildR microbiota-colonized WT and FcRg KO mice treated with

29.2

2

20

6 tTreg : pTreg

FcR KO (ISO)

10

**

y

WT (ISO)

4

** **

**** *** ***

40

CD4

CD4

F

21

15

g

Foxp3

8.51

*** ***

Foxp3+

FcR KO (ICB) % of CD4+ T cells

WT (ISO)

6

IFN + IL-17A+

***

40

ISO

E

IL-17A+

+

p

IFN

51.5

FcR KO (ICB) 0.77 8.15

CD4+ T cells (x104)

93.6

WT (ICB) 5.71 3.19

FcR KO (ISO) 3.74 0.32

IL-17A

IL-17A

WT (ISO) 3.42 0.34

N.S.

0

6 9 12 Days

WT (ICB) FcR KO (ICB) FcR KO (ISO)

C

*

30

Histology Score

Lipocalin-2 2500 2000 1500 1000 500 0

Fc WT R (I Fc KO CB) R ( KO ICB (IS ) O )

ng / g feces

A

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

300 100 0

10 5 0

9 12 15 Days

600

an

anti-CTLA-4 + anti-PD-1

200 0

0 3 6 9 12 15 18 Days

H11-HLE + anti-PD-1

* ** ******** 0 6 9 12 15 18 Days

H11-HLE + anti-PD-1

H11-HLE + anti-PD-1

an

isotype anti-CTLA-4 + anti-PD-1

** *

g

isotype anti-CTLA-4 + anti-PD-1

25 20 15 10 5 0

p

0

*** ** ****

*

Isotype

TL A I 11 -4 + sot yp -H a LE nt e i-P + an D-1 tiPD -1

**

400

F

C

1000

**** **** ***

Lipocalin-2

Histology Score

E

CT26

1500 500

6

15

isotype anti-CTLA-4 H11-HLE

ng / g feces

2000

**

** *

20

o ti- typ e C TL H A-4 11 -H LE

0 3 6 9 12 15 Days

D

H11-HLE

* **

0

0

isotype anti-CTLA-4 H11-HLE Tumor volume (mm3)

* *

200

anti-CTLA-4

Histology Score

*

500

Isotype

*

Is

***

1000

C

Lipocalin-2

400

H

B

MC38

ti-

1500

ng / g feces

Tumor volume (mm3)

A

Considering the pathogenic contributions of CD4+ T cells in the current mouse model of CTLA-4 blockade–associated colitis, we sought to further define CD4+ T cell responses by single cell transcriptomics. Sort-purified CD4+ T cells from the intestinal tissues of isotype or anti-CTLA-4 antibody-treated mice were labeled with hashtag antibodies to minimize batch effects, and pooled samples underwent droplet-based scRNA-seq for gene expression and paired TCR⍺ and TCRb clonotype analyses. After filtering out cells with mitochondrial DNA content over 25% and those with a unique molecular identifier of 7500,

6 of 9

,

5 January 2024

CTLA-4 blockade induces oligoclonal TH1 cells with cytotoxic features

we acquired 9671 high-quality transcriptomes. We clustered cells by using the top 5000 most differentially expressed genes and a panel of markers to stabilize immune lineage segregation (fig. S14 and table S2). To better preserve the accurate correlation between geographic distance and biological distance and improve the visualization of intracluster variance, we employed supCPM to project the clusters onto a two-dimensional space (31) (Fig. 3A). We identified nine CD4+ T cell functional subsets and a small cluster of B cells (Fig. 3A and fig. S14). Following anti-CTLA-4 antibody treatment, we found a selective and marked increase in IFNg+ Tbet+ TH1 cells defined by clusters 1 and 3 which were mostly absent in isotype-treated mice (Fig. 3, A and B). One of the TH1 clusters showed enhanced expression of PD-1 (cluster 3) and was further distinguished by elevated expression of transcripts related to cytotoxicity

y

Lo et al., Science 383, 62–70 (2024)

These data suggest that nonthymic Treg persistence is selectively associated with intestinal homeostasis during ICB therapy.

y g

Ccr7 (fig. S13, C and D). Although human Treg subset-defining transcripts remain imperfect, high Ikzf2 expression may aid in the identification or enrichment of human tTregs (32, 33). We found that the Ikzf2hi tTreg subcluster 7-1, the most abundant Treg subset, was similarly elevated in patients undergoing ICB therapy who developed colitis as well as ICB-treated patients who did not develop colitis compared with control cases (fig. S13E). By contrast, the relative proportion of the Ikzf2lo nonthymic Treg subcluster 7-2 was diminished in ICBtreated patients who had developed colitis compared with ICB-treated patients who did not develop colitis or control cases (fig. S13E). Moreover, the percentage of Ikzf2lo Treg subcluster 7-2 is sustained in ICB-treated patients who did not develop colitis compared with control cases (fig. S13E). Treg subcluster 7-3 abundance was unaltered across patient groups.

independent experiments. (D to F) BALB/c mice colonized with WildR microbiota were injected subcutaneously with CT26 tumor cells, then treated with isotype, anti-CTLA-4 antibodies and anti-PD-1 antibodies, or H11-HLE and anti-PD-1 antibodies. Tumor volumes were tracked (D), and intestinal inflammation was assessed by fecal lipocalin-2 (E). (F) H&E-stained cecal sections 18 days after starting ICB treatment and histology scores. Scale bars, 200 mm. White arrowheads point to epithelial damage and open arrows indicate inflammatory infiltrates. Each point represents an individual mouse. *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001; Dunnett’s multiple comparisons test (D, E), or Kruskal-Wallis with Dunn’s test (F) versus anti-CTLA-4 and anti-PD-1 group. Top symbols denote P values of comparisons with isotype treatment group, and bottom symbols denote P-values of comparisons with H11-HLE and anti-PD-1 treatment group (E). Data presented as mean ± SEM (D and E). Representative data of two independent experiments.

y

Fig. 5. Anti-CTLA-4 nanobodies stimulate anticancer immunity without inducing intestinal inflammation. (A to C) WildR microbiota-colonized WT mice were injected subcutaneously with MC38 tumor cells, then treated with isotype, anti-CTLA-4 antibodies, or anti-CTLA-4 H11 nanobodies with half-life extender (H11-HLE). Tumor volumes were tracked (A), and intestinal inflammation was assessed by fecal lipocalin-2 (B). (C) H&E-stained cecal sections 15 days after starting ICB treatment, and histology scores. Scale bars, 200 mm. White arrowheads point to epithelial damage and open arrows indicate inflammatory infiltrates. Each point represents an individual mouse. *, P < 0.05; **, P < 0.01; ***, P < 0.001; Dunnett’s multiple comparisons test (A and B), or Kruskal-Wallis with Dunn’s test (C) versus anti-CTLA-4 group. Top symbols denote P-values of comparison with isotype treatment group and bottom symbols denote P-values of comparison with H11-HLE treatment group (B). Data presented as mean ± SEM (A and B). Representative data of two

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

,

7 of 9

y

5 January 2024

Given the requirement of Fc-FcgR interactions in the induction of colitis by anti-CTLA-4 antibodies, we hypothesized that camelid heavy chain–only antibody fragments (VHHs) or nanobodies which lack an Fc domain may be used for CTLA-4 blockade therapy while reducing the risk of gut irAEs. We utilized the CTLA-4-binding H11 VHH stabilized with a half-life extender (H11-HLE) that was previously found to exhibit anticancer activity in some settings (35, 36). We sought to compare the antitumor activity of H11-HLE and anti-

y g

Lo et al., Science 383, 62–70 (2024)

Anti-CTLA-4 nanobodies stimulate antitumor immunity without inducing colitis

y

The depletion of intratumoral Tregs by antiCTLA-4 antibodies following Fcg receptor (FcgR) engagement is thought to be one of the mechanisms involved in ICB-induced tumor rejection, although antitumor effects without Treg depletion in tumor tissue have been observed in several studies (34–40). To test whether analogous Fc-dependent processes occur in the gut, we employed mice that lack the common FcRg chain which is required for the assembly and function of FcgR family members (41). FcRg-deficient mice colonized with WildR microbiota were highly resistant to intestinal inflammation following treatment with antiCTLA-4 antibodies as demonstrated by minimal fecal LCN-2 induction during the course of ICB treatment and no evidence of immunopathology in cecal tissues compared with anti-

CTLA-4 antibodies in syngeneic tumor models; mice inoculated with tumors were also colonized with WildR microbiota to permit simultaneous tracking of ICB-mediated tumor rejection and colitis. First, to examine anticancer responses in a monotherapeutic checkpoint blockade setting, we inoculated mice subcutaneously with MC38 adenocarcinoma cells, and targeted CTLA-4. Compared with isotype-treated mice in which tumors grew as expected and reached the experimental endpoint by day 15 of the treatment course, we found that mice receiving anti-CTLA-4 antibodies displayed reduced tumor sizes by days 12 and 15 after starting CTLA-4 blockade (Fig. 5A and fig. S17, A and B). Furthermore, the same animals receiving anti-CTLA-4 blockade developed colitis as indicated by fecal LCN-2 induction during the treatment course and high histological disease scores in intestinal tissues compared with animals treated with isotype antibodies (Fig. 5, B and C). Notably, tumor-inoculated mice receiving H11-HLE exhibited similar antitumor activity as mice receiving antibodies against CTLA-4 but by contrast did not show evidence of colitis (Fig. 5, A to C, and fig. S17, A and B). While assessing MC38 intratumoral T cells—in comparison to the isotype control—the percentages of Foxp3+ Tregs in CD4+ T cells were reduced in animals treated with either anti-CTLA-4 antibodies or H11-HLE, though to a lesser extent (fig. S18, A to C). Moreover, treatment with anti-CTLA-4 antibodies did not alter the percentages of Helios+ or RORgt+ intratumoral Tregs compared with isotype controls thus indicating the tissue-specific effects of antiCTLA-4 antibodies on large intestinal Treg subsets (fig. S18, D to F). In addition, the frequencies of CD4+ Foxp3− T cells among T cells were similarly elevated in tumors of mice treated with anti-CTLA-4 antibodies and H11HLE compared with isotype controls (fig. S18G). Collectively, these data suggest that Fc domain– lacking nanobodies that block CTLA-4 can stimulate antitumor responses without inducing gut irAEs. Combination CTLA-4 and PD-1 blockade can stimulate highly potent antitumor immune responses but also increase the risk and severity of irAEs (3, 4). We therefore hypothesized that the use of the Fc-lacking H11-HLE CTLA-4 inhibitor with PD-1 blockade can provide similar therapeutic benefits while overcoming the limitations of its associated gut toxicities. To test this, we inoculated WildR microbiota mice with CT26 tumor cells which require both CTLA-4 and PD-1 blockade for efficient tumor rejection (42). We found that as tumors grew in isotype-treated animals, mice treated with anti-CTLA-4 antibodies and anti-PD-1 antibodies had similarly efficient antitumor responses compared with mice treated with H11-HLE and anti-PD-1 antibodies, as evidenced

g

Colitis induced by anti-CTLA-4 antibodies requires depletion of Tregs by Fcg receptors

CTLA-4 treated WT mice and isotype-treated FcRg-deficient mice (Fig. 4, A and B). While we observed an increase in CD4+ T cells expressing IFNg, IL-17A, and both IFNg and IL-17A in WT mice treated with anti-CTLA-4 antibodies, the induction of cytokine-producing CD4+ T cells was not observed in FcRg-deficient mice similarly receiving the anti-CTLA-4 antibodies nor mice receiving isotype antibodies (Fig. 4, C and D). Moreover, under anti-CTLA-4 antibody treatment conditions, the percentages of intestinal Foxp3+ Tregs in WT mice were reduced compared with FcRg-deficient mice and isotypetreated mice with a notable reduction in the proportion of RORgt+ pTreg (Fig. 4, E and F). These data suggest that FcRg-dependent depletion of Tregs is required for gut inflammation in response to treatment with antiCTLA-4 antibodies. We further assessed the function of the antiCTLA-4 antibody Fc domain during ICB-driven intestinal inflammation using a humanized mouse model of CTLA-4 blockade. We challenged a CTLA-4 transgenic mouse strain in which the extracellular and transmembrane domains of the mouse Ctla4 sequence were replaced by the human ortholog (huCTLA-4 KI) with anti-human CTLA-4 antibodies containing a mouse IgG2a Fc domain (ipilimumab mIgG2a), or anti-human CTLA-4 antibodies with three amino acid substitutions (L234A, L235A, P329G) in the Fc domain (ipilimumab LALAPG) rendering the Fc domain inert to interact with FcgRs (36). We found that huCTLA-4 KI mice colonized with WildR microbiota and treated with ipilimumab mIgG2a developed intestinal inflammation as indicated by fecal LCN-2 levels and assessment of intestinal tissue sections (fig. S16, A and B). However, there was minimal or no indication of disease in huCTLA-4 KI mice colonized with WildR microbiota treated with ipilimumab LALAPG, and in huCTLA-4 KI mice re-colonized with SPF microbiota treated with ipilimumab mIgG2a (fig. S16, A and B). Results from this humanized CTLA-4 mouse model further highlight the requirement of a functional Fc domain for microbiota-dependent intestinal inflammation induced by ipilimumab.

p

(Gzma, Gzmb, Prf1) and chemokines (Ccl3, Ccl4, Ccl5) involved in myeloid cell responses (Fig. 3, A and B, and fig. S15A). Anti-CTLA-4 treatment resulted in a decrease in the relative abundance of follicular helper T-cells, naïve CD4+ T cells, Tregs, Tcf7+ CD4+ T cells, TH22 cells, TH17 cells, and TH2 cells (Fig. 3, A and B). To better assess the impact of antiCTLA-4 treatment on distinct Treg subsets, we separated cluster 5 and performed further clustering and supCPM rendering. We observed three distinct Treg clusters: pTreg (cluster 5-1), tTreg (cluster 5-2), and an intermediate Treg subcluster defined by elevated Tbet expression (cluster 5-3) (Fig. 3C and table S3). Consistent with previous findings, anti-CTLA-4 antibody treatment resulted in the selective depletion of pTregs (42% reduction compared with isotype-treated mice) whereas tTregs and Tbet+ Tregs remained intact (Fig. 3D). Although most cells from TH1 clusters 1 and 3 of the isotype-treated mice expressed unique TCRs, CTLA-4 blockade led to a substantial increase in the frequency of clonally expanded TH1 cells whereas a major portion of clonally expanded Tregs were eliminated (Fig. 3E). Further assessment of clonotype abundance in TH1 clusters 1 and 3 revealed that only five clonotypes with a frequency of 14 or higher were detected, all of which were selectively expanded by antiCTLA-4 antibody treatment (fig. S15B). Among the five enriched clonotypes, a highly dominant clonotype (33% of C3 cells from anti-CTLA-4– treated mice) was observed in the cytotoxic PD-1+ TH1 cluster and was 14 times more abundant than the next most frequent clonotype within the same cluster (fig. S15B). Collectively, these observations indicate that colitic TH1 cells are normally restricted under homeostatic conditions but can be selectively induced by anti-CTLA-4 antibodies in the presence of microbes of the WildR microbiota.

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

Discussion

AC KNOWLED GME NTS

,

8 of 9

y

The authors thank K. Martin and A. Aboud for animal husbandry, P. Kuffa for laboratory management, T. Duval, L. Keeton, and L. Kennedy of the University of Michigan (UM) Germ-Free Animal Core, T. Tamsen and M. Coon of the UM Advanced Genomics Core, W. Trim and A.M. Deslauriers of the UM Flow Cytometry Core, S. McClintock of the UM Pathology Flow Cytometry Core, C. Strayhorn of the UM Dentistry Histology Core, K. Warren of the UM Tissue and Molecular Pathology Histology Core, and the UM Host Microbiome Initiative. Funding: This work was supported by NIH grants R01 DK121504 and R01 DK095782 and a grant from Takeda/Millennium Pharmaceuticals (to G.N.). B.C.L. was supported by a Canadian Institutes of Health Research Fellowship. Additional funding provided by a Crohn’s and Colitis Foundation Senior Research Award (to R.C.), NIH grants R01 DE026728 and R01 DE030691 (to Y.L.L.), and NSF grant IOS-2107215 (to Y.X.). Research reported in this publication was supported by the National Cancer Institutes of Health under Award Number P30 CA046592 by the use of the following Cancer Center Shared Resources: Single Cell Spatial Analysis, and Tissue and Molecular Pathology. Author contributions: B.C.L., I.K., J.Y., L.V., R.C., and M.M. performed experiments, and analyzed data. Y.S. and M.H.S. provided advice, discussion, and critical materials. N.I. performed 16s rRNA analysis, and Y.X. and Y.L.L. performed single-cell immune profiling analyses. W.Z. designed experiments and provided advice. B.C.L. and G.N. conceived the project and designed experiments. G.N. supervised the study. B.C.L and G.N. wrote the manuscript with input from all authors. Competing interests: Y.S. and M.H.S. are employees of Takeda Pharmaceuticals International Co., Cambridge, MA, U.S.A. The other authors declare that they have no competing interests. Data and materials availability: C57BL/6

y g

5 January 2024

1. A. D. Waldman, J. M. Fritz, M. J. Lenardo, Nat. Rev. Immunol. 20, 651–668 (2020). 2. W. Zou, J. D. Wolchok, L. Chen, Sci. Transl. Med. 8, 328rv4 (2016). 3. F. Martins et al., Nat. Rev. Clin. Oncol. 16, 563–580 (2019). 4. E. Soularue et al., Gut 67, 2056–2067 (2018). 5. Y. Wang et al., Inflamm. Bowel Dis. 24, 1695–1705 (2018). 6. A. M. Luoma et al., Cell 182, 655–671.e22 (2020). 7. A. V. Heul, T. Stappenbeck, J. Allergy Clin. Immunol. 141, AB119 (2018). 8. Y. Zhou et al., J. Exp. Med. 220, e20221333 (2023). 9. E. Perez-Ruiz et al., Nature 569, 428–432 (2019). 10. F. Wang, Q. Yin, L. Chen, M. M. Davis, Proc. Natl. Acad. Sci. U.S.A. 115, 157–161 (2018). 11. K. Adam, A. Iuga, A. S. Tocheva, A. Mor, PLOS ONE 16, e0246168 (2021). 12. M. C. Andrews et al., Nat. Med. 27, 1432–1441 (2021). 13. K. Dubin et al., Nat. Commun. 7, 10391 (2016). 14. R. C. Simpson et al., Nat. Med. 28, 2344–2352 (2022). 15. S. P. Rosshart et al., Cell 171, 1015–1028.e13 (2017). 16. L. K. Beura et al., Nature 532, 512–516 (2016). 17. D. Masopust, C. P. Sivula, S. C. Jameson, J. Immunol. 199, 383–388 (2017). 18. S. P. Rosshart et al., Science 365, 6452 (2019).

y

Lo et al., Science 383, 62–70 (2024)

RE FERENCES AND NOTES

19. B. Chassaing et al., PLOS ONE 7, e44328 (2012). 20. E. V. Russler-Germain, S. Rengarajan, C. S. Hsieh, Mucosal Immunol. 10, 1375–1386 (2017). 21. T. Tanoue, K. Atarashi, K. Honda, Nat. Rev. Immunol. 16, 295–309 (2016). 22. K. Atarashi et al., Science 331, 337–341 (2011). 23. G. J. Britton et al., Immunity 50, 212–224.e4 (2019). 24. M. B. Geuking et al., Immunity 34, 794–806 (2011). 25. S. K. Lathrop et al., Nature 478, 250–254 (2011). 26. C. Ohnmacht et al., Science 349, 989–993 (2015). 27. J. L. Round, S. K. Mazmanian, Proc. Natl. Acad. Sci. U.S.A. 107, 12204–12209 (2010). 28. E. Sefik et al., Science 349, 993–997 (2015). 29. M. Xu et al., Nature 554, 373–377 (2018). 30. S. C. Sasson et al., Gastroenterology 161, 1229–1244.e9 (2021). 31. Z. Zhai, Y. L. Lei, R. Wang, Y. Xie, Bioinformatics 38, 2496–2503 (2022). 32. K. R. James et al., Nat. Immunol. 21, 343–353 (2020). 33. A. M. Thornton et al., J. Immunol. 184, 3433–3441 (2010). 34. F. Arce Vargas et al., Cancer Cell 33, 649–663.e4 (2018). 35. D. Bauché et al., J. Immunother. Cancer 8, e001584 (2020). 36. Y. Sato et al., Cancer Immunol. Immunother. 71, 2421–2431 (2022). 37. A. Sharma et al., Clin. Cancer Res. 25, 1233–1238 (2019). 38. Y. Bulliard et al., J. Exp. Med. 210, 1685–1693 (2013). 39. M. J. Selby et al., Cancer Immunol. Res. 1, 32–42 (2013). 40. T. R. Simpson et al., J. Exp. Med. 210, 1695–1710 (2013). 41. T. Takai, M. Li, D. Sylvestre, R. Clynes, J. V. Ravetch, Cell 76, 519–529 (1994). 42. K. Kim et al., Proc. Natl. Acad. Sci. U.S.A. 111, 11774–11779 (2014). 43. W. Wang et al., Nature 569, 270–274 (2019). 44. Y. Wang et al., Nat. Med. 24, 1804–1808 (2018). 45. H. Xu et al., J. Exp. Med. 218, e20210777 (2021). 46. B. Akagbosu et al., Nature 610, 752–760 (2022). 47. R. Kedmi et al., Nature 610, 737–743 (2022). 48. M. Lyu et al., Nature 610, 744–751 (2022). 49. W. Ise et al., Nat. Immunol. 11, 129–135 (2010). 50. K. S. Peggs, S. A. Quezada, C. A. Chambers, A. J. Korman, J. P. Allison, J. Exp. Med. 206, 1717–1725 (2009). 51. L. F. Mager et al., Science 369, 1481–1489 (2020). 52. T. Tanoue et al., Nature 565, 600–605 (2019). 53. M. Vétizou et al., Science 350, 1079–1084 (2015).

g

The establishment of a robust animal model of ICB-driven colitis would be beneficial for gaining mechanistic insights into irAEs observed in the clinic and for improving current antitumor ICB therapies with reduced toxicities. In this study, we demonstrate the critical role of the microbiota in colitis induced by ICB and reveal key contributions of CD4+ TH1 cells and IFNg in the pathogenesis of disease which is consistent with previously reported features of the human disease (4, 6) (fig. S20). Several strategies have been proposed for suppressing colitis without altering the beneficial antitumor effects of ICB including neutralization of TNF or IL-6 (8, 9). Moreover, the modification of the gut microbiota by fecal microbiota transplantation can mitigate intestinal inflammation induced by ICB (44). Here, we report that CTLA-4 inhibitors that lack an Fc domain do not induce colitis in mice which are otherwise susceptible to colitis when receiving conventional anti-CTLA-4 antibodies. Importantly, anti-CTLA-4 VHHs can effectively stimulate antitumor responses in mice without causing gut irAEs, thus circumventing the necessity for the neutralization of various proinflammatory cytokines or other strategies to modify the microbiota to remedy intestinal inflammation. These results also indicate that many immunological redundancies exist in the gut to prevent inappropriate T cell activation by the gut microbiota and that although the microbiota-induced Tregs are present, CTLA-4 blockade by itself is not sufficient to induce colitis. In several genetic models, direct perturbation of gut pTregs renders mice more susceptible to intestinal immunopathologies (26, 28, 29, 45). Furthermore, mice lacking a population of tolerogenic RORgt+ antigen-presenting cells,

which are required for the induction of bacteriareactive pTregs, appear to develop intestinal inflammation associated with exaggerated proinflammatory T cell responses (46–48). However, CTLA-4 can also restrain T cell activation independently of Tregs in autoimmunity and antitumor responses (49, 50). Therefore, further studies are needed to understand the potential cumulative effects of T cell CTLA-4 blockade and anti-CTLA-4 antibody-mediated depletion of Tregs during colitis. We find that a CTLA-4 inhibitor lacking an Fc domain had similar antitumor immunitypromoting effects as conventional anti-CTLA-4 antibodies in mice harboring WildR microbiota. However, some—but not all—reports indicate that anti-CTLA-4 antibodies containing an Fc domain with high binding affinity to Fc receptors and functional Fc receptor activity is required for optimal tumor clearance through the selective depletion of intratumoral Tregs (34–40). This discrepancy may be explained, at least in part, by the dissimilarities in gut microbiota composition described in these animal models, as select gut bacterial strains or microbial communities have a profound impact on extra-intestinal immunity and responsiveness to CTLA-4 blockade for anticancer therapy (51–53). Given that the gut microbiota of free-living mice is highly immunogenic, the threshold of immune tolerance disruption required for activating optimal antitumor responses may be lower in WildR mice. Moreover, as recent data indicate that experimental mouse models that use the microbiota of free-living animals more accurately recapitulate human responses to immune-modulating therapies (18), our work provides evidence for the potential utility of an Fc-null anti-CTLA-4 antibody, either alone or in combination with PD-1 or PD-L1 blockade, in effectively stimulating anticancer immune responses without inducing intestinal inflammation.

p

by reduction in tumor sizes during the treatment course compared with isotype-treated controls (Fig. 5D and fig. S17, C and D). Moreover, only mice receiving both anti-CTLA-4 and anti-PD-1 antibodies developed colitis whereas isotype control mice or mice receiving H11-HLE and anti-PD-1 antibodies did not display evidence of intestinal inflammation (Fig. 5, E and F). We further evaluated the effects of H11HLE nanobodies and conventional anti-CTLA-4 antibodies, in combination with PD-L1 blockade, in a syngeneic B16F0 melanoma implantation model (43). Treatment with H11-HLE and antiPD-L1 antibodies was similarly effective as antiCTLA-4 antibodies and anti-PD-L1 antibodies in limiting subcutaneous B16F0 tumor growth but did not induce overt intestinal inflammation (fig. S19, A to E). These results indicate the potential benefits of an Fc-lacking CTLA-4 inhibitor in combination with PD-1 or PD-L1 blockade without incurring the complications of colitis.

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

mice harboring WildR microbiota were obtained from Taconic Biosciences via a material transfer agreement. Fecal microbiota 16s sequencing data and single CD4+ T cell RNA sequencing data are available at the NCBI SRA under BioProject number PRJNA944829. License information: Copyright © 2024 the authors, some rights reserved; exclusive licensee American Association for the Advancement of Science. No claim to original

US government works. https://www.sciencemag.org/about/ science-licenses-journal-article-reuse SUPPLEMENTARY MATERIALS

science.org/doi/10.1126/science.adh8342 Materials and Methods

Figs. S1 to S20 Tables S1 to S3 References (54–60) Submitted 16 March 2023; resubmitted 8 September 2023 Accepted 17 November 2023 10.1126/science.adh8342

p g y y g y ,

Lo et al., Science 383, 62–70 (2024)

5 January 2024

9 of 9

RES EARCH

POLYMERS

Thermodynamically stable plumber's nightmare structures in block copolymers Hojun Lee1, Sangwoo Kwon2, Jaemin Min1, Seon-Mi Jin3, Jun Ho Hwang3, Eunji Lee3, Won Bo Lee2, Moon Jeong Park1* Block copolymer self-assembly affords diverse nanostructures, spanning from spheres and cylinders to networks, offering meticulous control over properties and functionalities at the nanoscale. However, creating thermodynamically stable network structures with high packing frustration remains a challenge. In this study, we report a methodology to access diverse network structures such as gyroid, diamond, and primitive phases from diblock copolymers using end group and linker chemistry. The stability of the medial packing of polymer chain ends (plumber's nightmare structure) over skeletal aggregation (gyroid) is attributed to the interplay between the strength of the end-end interactions and the initial shape of the curvature. Our study establishes an approach to develop tailored network structures from block copolymers, providing an important platform for using block copolymers in nanotechnology applications.

1 of 7

,

5 January 2024

y

Lee et al., Science 383, 70–76 (2024)

Several experimental and theoretical studies  on diblock copolymers have identified the Ia3d structure as the most stable network phase,

y g

*Corresponding author. Email: [email protected]

Emergence of plumber’s nightmare for diblock copolymers

y

Department of Chemistry, Pohang University of Science and Technology (POSTECH), Pohang 37673, Republic of Korea. School of Chemical and Biological Engineering, Institute of Chemical Processes, Seoul National University, Seoul 08826, Republic of Korea. 3School of Materials Science and Engineering, Gwangju Institute of Science and Technology (GIST), Gwangju 61005, Republic of Korea.

2

g

1

Beyond the Ia 3d structure, block copolymers have also been theoretically predicted to  andIm3m  space form cubic lattices withPn3m groups that have Schwarz diamond (D) and primitive (P) surfaces, respectively (28). However, these structures have a higher packing  structure because frustration than the Ia3d excessive chain stretching is required to access the constant-mean-curvature structure, thereby substantially limiting their feasibility for experimental observations (29). In particular, diblock copolymers have difficulty filling the space to form minimal P surfaces with six intersecting tubes compared with gyroid (three tubes) and diamond (four tubes) structures (30). To explore ways for diversifying the type of accessible network structures from block copolymers, extensive experimental studies (31, 32) have been conducted. However, all observed double primitive structures were not in thermodynamic equilibrium. One current challenge is the creation of thermodynamically stable network structures with high packing frustration. A recent theoretical work by Reddy et al. (33) may provide important insights in this regard. According to them, manipulating the stability of the network structures by controlling the packing of polymer chain ends could be a prospective method to realize structures with high pacing frustration (34). However, there is no known method yet to control the arrangement and localization of the polymer chain ends.  Fddd, In this study, the stabilization of Ia3d,  and Im3m structures was realized using diend-functionalized block copolymers with different linkages. A polystyrene-b-poly(ethylene oxide) (PS-b-PEO, SEO) diblock copolymer with a nominally symmetric molecular weight of 7.5-b-7.4 kg mol–1 and a PEO volume fraction (fPEO) of 0.48 was synthesized. As briefly shown in Fig. 1A, the hydroxyl terminal of SEO was first converted to a diol or primary amine group to form an intermediate, followed by a Michael

p

S

tarting with the theoretical prediction of the phase behavior of diblock copolymers by Leibler (1), experimental efforts have established phase diagrams of block copolymers based on the block volume fraction (f), degree of polymerization (N), and Flory-Huggins interaction parameter (c) between two blocks (2, 3). Recent advances in polymer synthesis have further allowed the exploration of the effects of the molecular architecture (4, 5), interfacial molecular composition (6, 7), and conformational asymmetry (8–11) on the self-assembly behavior of block copolymers. Among the various self-assembled structures formed by block copolymers, network structures characterized by triply periodic minimal surfaces are particularly intriguing (12) owing to their improved mechanical (13, 14) and transport properties (15–17) across inherently bicontinuous nanodomains. Through the controlled domain sizes of the network structures with integrated functional moieties, block copolymers can serve as advanced nanomaterials in soft electronics (18, 19) and energy storage systems (20, 21). In this context, the structure ) property relationship of the gyroid ( Ia3d structure (minimal G surface) has been of particular interest (14, 15). However, the gyroid structure is stable only over a narrow window of the phase diagram of block copolymers, and its feasibility relies on tedious composition screening (22). The manipulation of the polymer chain conformation at the interfacial region (23–26) and compositional nonuniformity at the interface (27) have been shown to play a  structures. crucial role in stabilizing the Ia3d

reaction with diethyl vinylphosphonate (dPE). After the ethyl group was removed by hydrolysis, two types of SEOs with diphosphonic acid (dPA) end groups were prepared, in which the termini were connected to PEO by different linker molecules. Hereafter, the dPA-ended SEOs are referred to as SEO-N-dPA (amine linker) and SEO-O-dPA (ether linker). The detailed synthetic procedures and data are provided in figs. S1 and S2. The end group conversion rates were higher than 95% and the molecular weight distribution of SEO was unaffected by the terminal modification reactions (fig. S3). Figure 1B shows representative small-angle x-ray scattering (SAXS) profiles and transmission electron microscopy (TEM) images of pristine SEO, SEO-amine, SEO-N-dPA, SEO-diol, and SEO-O-dPA. Prior to the measurements, the freeze-dried SEOs were annealed at 130°C for one week and cooled down to 25°C. Pristine SEO exhibited a lamellar (lam) structure with a domain spacing (d10) of 15.9 nm. SEO-amine and SEO-diol also exhibited analogous lam structures with a d10 of 16.4 nm. Notably, the inclusion of dPA end groups in the SEO intermediates led to the evolution of peculiar network morphologies. For SEO-N-dPA, we saw a mixture of orthorhombic networks (Fddd, hkl  (hkl index in black) index in blue) and Ia3d structures, with Fddd being dominant. The unit cell parameters of Fddd were (a:b:c) = (1:1.97:3.41), with a = 20.0 nm. Bright-field TEM micrographs were recorded in each area  were observed, as shown where Fddd andIa3d in the bottom inset images of Fig. 1B. Notably, upon changing only one linker molecule from amine to ether, SEO-O-dPA exhib structure with a d110 of 31.4 nm ited an Im3m (hkl index is marked in the scattering profile of Fig. 1B). The top insets of Fig. 1B show the tomography images and a captured TEM im age for the direct visualization of the Im3m structure in SEO-O-dPA. These clearly reveal the primitive cubic morphology in the [100] projection with the interconnected PS domains (bright phases). Unprecedentedly, diverse network structures were observed from a single SEO by functionalizing the end group, which accounts for less than 1% of the total number of repeating units. Electron energy loss spectroscopy mapping in TEM was performed to elucidate the type of ele ments constituting each domain of the Im3m structure. The bright-field TEM micrograph and oxygen mapping image of SEO-O-dPA in the  structure are shown [111] projection of the Im3m in fig. S4, revealing that the PEO chains form minor phases (shown dark by RuO4 staining).

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

A

SEO-N-dPA

Michael reaction

SEO

hydrolysis

(211)

B

SEO-O-dPA

SEO-O-dPA

4

100 nm

2

SEO-diol (315)

(242) (062)

g

1

10

50 nm

SEO-O-dPA (131) (400) (202) (422)

(113)

(220)

10

p

Scattering Intensity (a.u.)

Im3m

(310) (222) (400) (330)(411) (420) (332) (422) (431) (521) (440) (611)(532) (541) (631) (710)(550)

3

10

(110) (200) (211), (111) (220)

10

Fig. 1. Stabilization of diverse network structures using di-end– functionalized block copolymers. (A) Synthetic routes for the preparation of polystyrene-b-poly(ethylene oxide) (SEO) having diphosphonic acid (dPA) end groups. (B) SAXS profiles of end-functionalized SEOs measured at 25°C. hkl indices are indicated in parenthesis (blue, Fddd; black, Ia3d). q, scattering vector. a.u., arbitrary units. (Bottom insets) Bright-field TEM images of SEO-N-dPA. (Top insets) TEM tomography images and a captured TEM image are shown for the visualization of the Im3m structure in SEO-O-dPA. Dotted SAXS profiles show the morphologies of SEO intermediates, SEO-diol, and SEO-amine.

SEO-N-dPA SEO-N-dPA y

0

10

SEO-amine SEO

-1

10

0.2

0.4

0.6

0.8

1.0

q (nm-1)

Fddd

(40). The P surface can be bent to form the G and D surfaces, complying with the aD:aG:aP = 1:1.58:1.28 ratio (41). This was not the case with   our experiments, in which the Im3m-to-Ia 3d  transitions occurred without changes -to-Im3m in the domain size. Thus, a transition mech and Ia3d  structures anism between the Im3m through the merging and splitting of nodes was proposed. As schematically depicted in the upper inset of Fig. 2A, when viewed along the  and [111] direction, the (211) planes of Im3m  were parallel to the center of the nodes. Ia3d  two enantiomeric nodes Upon cooling Ia3d, with three intersecting channels merged through PEO crystallization, resulting in a  These new node with six channels of Im3m. are illustrated on the right side of each scattering profile.  structure of SEO-O-dPA was viThe Im3m sualized by electron density reconstruction using MATLAB and SUPERFLIP software with structure factors extracted from the SAXS profile in Fig. 2A (42). Subsequently, we performed Le Bail refinement using JANA2006 (43). On 2 of 7

,

5 January 2024

SEO-O-dPA. As shown by the SAXS profiles in Fig. 2A, heating the sample from 25° to 50°C  caused the destabilization of Im3m, as evidenced by the disappearance of the (110) peak.   phase transition, This led to an Im3m-to-Ia 3d and the Ia 3d structure remained intact up to  150°C (fig. S5). When the sample with the Ia3d structure was cooled from 110° to 25°C, an immediate reemergence of the (110) peak of  structure was observed along with the Im3m  Upon monitoring the the coexistence of Ia3d. changes in morphology of the cooled sample at room temperature, the slow reenactment  from Ia3d  through the coinciding of Im3m d211 of two cubic structures was observed. This  transition is indicates that the Im 3m-to-Ia3d  structure thermally reversible, and the Im3m is in equilibrium at room temperature. The SAXS results obtained for unannealed SEO-O-dPA during in situ temperature changes are shown in fig. S6. The results for the annealed sample during sample rotation are presented in fig. S7. Minimal D, G, and P surfaces are related to each other by the Bonnet transformation

y

Lee et al., Science 383, 70–76 (2024)

Ia3d

y g

which exists in the narrow phase window between the hex and lam structures (22). The  structure, called plumber’s nightmare, Im3m was identified in block copolymer–templated mesoporous silica by Finnefrock et al. (35) two decades ago; however, evidence of primitive cubic structures in equilibrium does not exist for any neat block copolymers. In fact, computational studies based on Gaussian statistics have reported that excessive stretching of the polymer chains in double diamonds and double primitives is not preferred over gyroids owing to the entropic penalty (36, 37). For block copolymer–homopolymer blends, the possibility of the existence of a thermodynamically stable plumber’s nightmare structure was theoretically predicted (28, 38, 39). However, it is only applicable over a finite region of the phase diagrams when the requirements for the composition or conformational asymmetry of the block chains and the architecture or length of the homopolymer are met. This prompted us to investigate the ther structure of mal stability window of the Im3m

50 nm

50 nm

1.2

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

p g y y g

Lee et al., Science 383, 70–76 (2024)

5 January 2024

sity isosurface and charge-density contour maps  structure of SEO-O-dPA along the of the Im3m [100] and [111] directions, confirming the characteristic structural patterns of the plumber’s nightmare. The lower panels of Fig. 2C display bright-field TEM images and the related inverse Fast Fourier transformed (FFT) images from the [111], [211], and [531] projections. Additional sets of TEM tomography images are shown in fig. S10 and movie S1. A close correlation between the hexapod stability and crystallization of dPA-ended PEO

chains was envisaged for SEO-O-dPA. Essen is favored over Ia3d  at low temtially, Im3m peratures when the dPA end groups are well arranged at the center of the PEO domains of SEO-O-dPA. As the temperature increases, the PEO crystals melt, and the increased PEO chain mobility disorganizes the end-end arrangement. This results in the splitting of the bulky hexapod into smaller tripods in the direction of reducing packing frustration. This was investigated in more detail by simultaneous SAXS and wide-angle x-ray scattering 3 of 7

,

the basis of the goodness-of-fit parameter and the weighted profile R-factor, we affirmed the reliability of our structure determination. The representative Le Bail refinement result in comparison with the experimental SAXS profile is presented in fig. S8. Rigorous structural validation through the simSAXSLee software package built on the MATLAB platform (fig. S9) demonstrates the close correlation between the measured scattering intensity of the hkl reflection (Ihkl) and calculated Ihkl values (44). Figure 2C displays the resultant electron den-

dPA-ended PEO arrangements in hexapod (Im3m) and tripod (Ia3d). (C) Electron density isosurface and charge-density contour maps of the Im3m structure of SEO-O-dPA in the [100] and [111] directions and the bright field TEM and related inverse FFT images from the [111], [211], and [531] projections.

y

 structures in SEO-O-dPA. (A) SAXS Fig. 2. A deeper look into the Im3m profiles of SEO-O-dPA upon heating and cooling. The epitaxial relationship between the Ia 3d and Im 3m structures and changes in the nodes owing to the structural transition are schematically illustrated in the insets. (B) Proposed

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

A

mPEO mPEO-O-dPA mPEO-N-dPA

-N-dPA: N-dPA: Zwitterionic

Absorbance

-O-dPA: H-bond

1150

1100

1050

2700 2600 2500 2400

1000

Wavenumber (cm-1)

B

lam

mPEO

twisted lam

mPEO-O-dPA

D

helical

@ 25 oC

mPEO-N-dPA

p

Fig. 3. Factors affecting the stability of the network structures of end-functionalized SEOs. (A) FT-IR spectra, end-end interactions, and (B) 2D scattering patterns of mPEO, mPEO-O-dPA, and mPEO-N-dPA. (C) SAXS profiles of di-end– functionalized SEOs (14.5-b-9.5 kg mol–1) measured at 25°C. Schemes depicting the medial versus skeletal chain packing of PEO are shown in the top right. (D) Phase diagram of SEOs with different types of di–end groups and linker molecules. The strength of end-end interaction and the degree of packing frustration were included as variables. (E) Projected phase diagram of dPA-ended SEOs in a low-cN regime. Conventional phase diagram is indicated with gray dotted lines for comparison.

C

g y

Scattering Intensity (a.u.)

vs.

SEO-O-dPE

E

L

L

SEO-O-dPA

0.2

0.4

0.6

0.8

1.0

1.2

N 0.2

-1

q (nm )

analyses with isothermal crystallization of SEO-O-dPA (fig. S11). For SEO-N-dPA with the amine linkage, in the temperature range 25° to 170°C, coexisting  were always observed phases of Fddd and Ia3d  over Fddd in(fig. S12). The fraction of Ia3d creased with increasing temperature, and the Ia 3d -to-Fddd transition occurred reversibly with cooling, which was relatively rapid comLee et al., Science 383, 70–76 (2024)

5 January 2024

  transition. The pared with the Ia3d-to-Im 3m results of molecular dynamics (MD) simulations based on the MARTINI approach are shown in fig. S13 (45). Contrary to the flat interface observed for pristine SEO, percolated structures appeared for SEOs with dPA end groups. Further, despite the small number fraction (180 ppm) (37) and N-heterocyclic carbenes (NHCs) (typically >185 ppm) (38). This divergence underscores a distinct electronic structure separating 4 from the realm of PHCs and NHCs featuring a s2p0 electronic state. It is worth noting that, without the presence of Rh coordination, attempts to synthesize a diphosphinocarbene through photolysis of 2 were unsuccessful, instead leading to the generation of a 1,4,5,6-tetrahydro-1,4,2-diazaphosphinine derivative, 8 (fig. S8) (39). The UV-visible spectrum of 4[BPh4] in tetrahydrofuran (THF) reveal two broad peaks at 436 and 385 nm (fig. S54). Based on time-dependent density functional theory calculations, these peaks arise from electronic transitions of highest occupied molecular orbital (HOMO)→lowest

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

p g y y g y , Fig. 2. Synthesis, characterization, and reactivity of metallocyclic carbenes. (A) Schematic representation of the synthetic route to the metallocyclic carbenes 4[BPh4] and 4[NTf2]. Mes, mesityl; Ad, 1-adamantyl; Tf, trifluoromethanesulfonyl. (B) Solid-state structures of 3 and 4[BPh4]. Hydrogen atoms and the noninteracting tetraphenyl borate anion are omitted for clarity. (C) Reactivity of 4[BPh4] or 4[NTf2] toward 4-dimethylaminopyridine, AgNTf2 and AdNC.

is polarized toward the P–N s*-antibonding orbitals of the 1,3,2-diazaphospholidine framework. Conversely, the LUMO, at −3.69 eV, largely resembles a s-type sp2-hybridized orbital centered at C(1). This electron distribution is also illustrated by the electron localization function (43) plot of 4 showing an intensified localized electron density at the C(1) p orientation (fig. S57). This is in contrast to the patterns calculated for NHCs (fig. S58). This observation Hu et al., Science 383, 81–85 (2024)

5 January 2024

confirms that 4 exhibits a s0p2 electronic state, which is markedly different from the states observed in previously isolated singlet carbenes. Moreover, our calculations suggest that an increase in the P–C–P bond angle does not affect the s0p2 electronic configuration of 4 (figs. S59 and 60). Based on gauge-independent atomic orbitalDFT NMR analyses, the distinct electronic configuration of 4 enhances three magnetic shielding components, leading to a prominent

upfield chemical shift for the carbene carbon (fig. S62). Intrinsic bond orbital (IBO) (44) calculations offer deeper insights into the bonding scenario of 4 (Fig. 3C and fig. S63). At the C(1) atom, a s bond is formed with both P(1) and P(2) (fig. S63, A and B). Simultaneously, an essentially nonbonding lone pair orbital resides at C(1), oriented perpendicularly to the CP2Rh plane (Fig. 3C, IBO1). This lone pair has minor 3 of 5

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

Together, these electron push-pull effects contribute substantially to the stability of the s0p2 electronic state of the carbene 4. To further understand the impact of d orbital size on the stabilization of the s0p2 electronic configuration of carbenes, we computationally examined the Co and Fe analogs of 4, denoted as 4-Co and 4-Fe (figs. S66 and 67). Although the bonding framework of 4-Co aligns with that of 4 suggesting that d orbital size might play a limited role in stabilizing the s0p2 configuration, 4-Fe presents as a formal Fe(II) complex. In this case, the carbon appears formally dianionic as a result of intramolecular charge transfer, highlighting the critical influence of the energy level of d electrons. Their considerable reducing power can lead to distinctly different bonding scenarios. Accordingly, the Wiberg bond indices for C(1)–P(1), Rh(1)–P(1), and C(1)–Rh(1) in 4 are 1.36, 0.41, and 0.26, respectively. The natural population analysis shows positive charges on the P atoms [1.61 atomic units (a.u.)] whereas

both Rh(1) (−0.20 a.u.) and C(1) (−1.13 a.u.) exhibit negative charges. In light of these findings, four plausible resonance forms of 4 are depicted in Fig. 3B, resembling those observed for NHCs (19). Nevertheless, it is imperative to underscore that in the context of 4, ylide structures emerge as a consequence of s delocalization/donation as opposed to p donation, characteristic of NHCs. Reactivity of 4 as a p-donor and s-acceptor

Contrary to PHCs (37) and NHCs (19), which function as s-donors and p-acceptors, our computational insights point toward carbene 4 operating as a p-donor and s-acceptor. In support of this notion the interaction of 4[BPh4] with the potent s-donor Lewis base, 4-dimethylaminopyridine (DMAP), results in the occupancy of the vacant s orbital of 4 by DMAP, culminating in the formation of 5[BPh4] in 89% yield (Fig. 2C), as evidenced by the x-ray structure of 5[BPh4] in fig. S50. When the small Lewis acid AgNTf2 is employed, the

p

delocalization toward P(1) and P(2), which aligns with the HOMO of 4. Although the two lone pairs at P(1) and P(2) predominantly coordinate to Rh(1), they also delocalize to the unoccupied s orbital of C(1) (IBO2 and IBO3). Furthermore, one of the filled d orbitals at Rh (1) provides slight stabilization for the in-plane unoccupied s orbital of C(1) (IBO4). Theoretically, this in-plane s delocalization/donation resembles the p-donation found in hitherto known singlet carbenes featuring the s2p0 electronic state (fig. S64). However, in this instance, the formally vacant orbital at C(1) manifests as an in-plane sp2-hybridized s orbital, diverging from the conventional out-of-plane pp-orbital. This inplane triple donation is instrumental in pushing the lone pair of electrons into the C(1) p orbital (Fig. 3D). Conversely, the P–N s*-antibonding orbitals within the 1,3,2-diazaphospholidine scaffold play a crucial role in drawing p electron density away from C(1), as evidenced by the analysis of second-order perturbation theory using the natural bond orbital method (fig. S65).

g y y g y , Fig. 3. Theoretical analysis of 4. (A) Energy diagram for the frontier Kohn-Sham orbitals. HOMO, highest occupied molecular orbital; LUMO, lowest unoccupied molecular orbital. (B) Illustration of the four distinct resonance structures. (C) Visual representation of selected intrinsic bond orbitals. (D) Schematic representing the essence of stabilization: push/push/push s-electron delocalization/donation and pull/pull p-electron negative hyperconjugation. Hu et al., Science 383, 81–85 (2024)

5 January 2024

4 of 5

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

37. J. D. Masuda et al., Chem. Asian J. 2, 178–187 (2007). 38. D. Tapu, D. A. Dixon, C. Roe, Chem. Rev. 109, 3385–3407 (2009). 39. S. Marrot, T. Kato, F. P. Cossío, H. Gornitzka, A. Baceiredo, Angew. Chem. Int. Ed. 45, 7447–7450 (2006). 40. E. P. Beaumier et al., J. Am. Chem. Soc. 142, 8006–8018 (2020). 41. Y. Ishida, B. Donnadieu, G. Bertrand, Proc. Natl. Acad. Sci. U.S.A. 103, 13585–13588 (2006). 42. P. Pyykkö, M. Atsumi, Chemistry 15, 12770–12779 (2009). 43. A. Savin, R. Nesper, S. Wengert, T. F. Fässler, Angew. Chem. Int. Ed. 36, 1808–1832 (1997). 44. G. Knizia, J. E. M. N. Klein, Angew. Chem. Int. Ed. 54, 5518–5522 (2015). 45. M.-M. Gan et al., Chem. Rev. 118, 9587–9641 (2018). 46. D. R. Tolentino et al., J. Am. Chem. Soc. 141, 9823–9826 (2019). 69. L. Falivene, L. Cavallo, Coord. Chem. Rev. 344, 101–114 (2017). AC KNOWLED GME NTS

Funding: We gratefully acknowledge financial support from the National Natural Science Foundation of China (22350004 and, 22271132 to L.L.L. and 22301122 to J.L.), Shenzhen Science and Technology Innovation Program (JCYJ20220530114806015), Guangdong Innovation and Entrepreneurial Research Team Program (2021ZT09C278), and Guangdong Provincial Key Laboratory of Catalysis (2020B121201002). We also acknowledge the assistance of SUSTech Core Research Facilities. The theoretical work was supported by the Center for Computational Science and Engineering at SUSTech. Author contributions: L.L.L. devised and supervised the project and wrote the manuscript. C.H. synthesized and isolated all compounds 1-8, determined the X-ray crystal structures and the NMR analysis. L.L.L. and C.H. performed the computational work. X.-F.W. was involved in the preparation of 1 and provided support for synthesizing Li(THF)1.5CN2TMS. J.L. helped in the synthesis of 5[BPh4] and 7[BPh4] and provided valuable discussions. X.-Y.C. provided help in the determination of X-ray crystal structures. Competing interests: The authors declare no competing financial interests. Data and materials availability: General information, experimental procedures, 1 H NMR/13C NMR/31P NMR/19F NMR/11B NMR spectra, high resolution mass spectrometry data, infrared spectrometry data and computational details are provided in the Supplementary Materials. Full crystallographic data for structures 2, 3, 4[BPh4], 5[BPh4], 6[NTf2], 7[BPh4] and 8 are available free of charge from the Cambridge Crystallographic Data Center under reference numbers 2267592, 2267593, 2267596, 2267599, 2267595, 2267590 and 2302945, respectively. License information: Copyright © 2024 the authors, some rights reserved; exclusive licensee American Association for the Advancement of Science. No claim to original US government works. https://www.sciencemag. org/about/science-licenses-journal-article-reuse

y g

SUPPLEMENTARY MATERIALS

y

science.org/doi/10.1126/science.adk6533 Materials and Methods Supplementary Text Figs. S1 to S73 Tables S1 to S11 References (47–68)

5 of 5

,

Submitted 3 September 2023; accepted 20 November 2023 10.1126/science.adk6533

y

5 January 2024

1. J.-B. Dumas, E. M. Péligot, in Annales de chimie et de physique (1835); pp. 5-74. 2. W. von E. Doering, A. K. Hoffmann, J. Am. Chem. Soc. 76, 6162–6165 (1954). 3. M. Gomberg, J. Am. Chem. Soc. 22, 757–771 (1900). 4. M. N. Hopkinson, C. Richter, M. Schedler, F. Glorius, Nature 510, 485–496 (2014). 5. J. P. Moerdyk, D. Schilter, C. W. Bielawski, Acc. Chem. Res. 49, 1458–1468 (2016). 6. M. Melaimi, R. Jazzar, M. Soleilhavoup, G. Bertrand, Angew. Chem. Int. Ed. 56, 10046–10068 (2017). 7. G. Guisado-Barrios, M. Soleilhavoup, G. Bertrand, Acc. Chem. Res. 51, 3236–3244 (2018). 8. A. Igau, H. Grutzmacher, A. Baceiredo, G. Bertrand, J. Am. Chem. Soc. 110, 6463–6466 (1988). 9. A. J. ArduengoIII, R. L. Harlow, M. Kline, J. Am. Chem. Soc. 113, 361–363 (1991). 10. D. Enders, O. Niemeier, A. Henseler, Chem. Rev. 107, 5606–5655 (2007). 11. E. Peris, Chem. Rev. 118, 9988–10031 (2018). 12. Y. Xia, D. Qiu, J. Wang, Chem. Rev. 117, 13810–13889 (2017). 13. A. V. Zhukhovitskiy, M. J. MacLeod, J. A. Johnson, Chem. Rev. 115, 11503–11532 (2015). 14. C. A. Smith et al., Chem. Rev. 119, 4986–5056 (2019). 15. L. Oehninger, R. Rubbiani, I. Ott, Dalton Trans. 42, 3269–3284 (2013). 16. W. Liu, R. Gust, Coord. Chem. Rev. 329, 191–213 (2016). 17. D. Bourissou, O. Guerret, F. P. Gabbaï, G. Bertrand, Chem. Rev. 100, 39–92 (2000). 18. K. Hirai, T. Itoh, H. Tomioka, Chem. Rev. 109, 3275–3332 (2009). 19. H. V. Huynh, Chem. Rev. 118, 9457–9492 (2018). 20. J. Vignolle, X. Cattoën, D. Bourissou, Chem. Rev. 109, 3333–3384 (2009). 21. A. Kalemos, T. H. Dunning Jr., A. Mavridis, J. F. Harrison, Can. J. Chem. 82, 684–693 (2004). 22. B. Chen, A. Y. Rogachev, D. A. Hrovat, R. Hoffmann, W. T. Borden, J. Am. Chem. Soc. 135, 13954–13964 (2013). 23. J. P. Wagner, J. Am. Chem. Soc. 144, 5937–5944 (2022). 24. G. Maier, J. Endres, Chemistry 5, 1590–1597 (1999). 25. L. Pauling, J. Chem. Soc. Chem. Commun. 688–689 (1980). 26. P. H. M. Budzelaar, P. R. Schleyer, K. Krogh-Jespersen, Angew. Chem. Int. Ed. 23, 825–826 (1984). 27. R. Wehrmann, H. Klusik, A. Berndt, Angew. Chem. Int. Ed. 23, 826–827 (1984). 28. H. Meyer, G. Baum, W. Massa, S. Berger, A. Berndt, Angew. Chem. Int. Ed. 26, 546–548 (1987). 29. H. Meyer, G. Baum, W. Massa, A. Berndt, Angew. Chem. Int. Ed. 26, 798–799 (1987). 30. Y. Shibutani, S. Kusumoto, K. Nozaki, J. Am. Chem. Soc. 145, 16186–16192 (2023). 31. J. Ruiz, V. Riera, M. Vivanco, M. Lanfranchi, A. Tiripicchio, Organometallics 17, 3835–3837 (1998). 32. J. Ruiz et al., Angew. Chem. Int. Ed. 42, 4767–4771 (2003). 33. J. Ruiz, M. E. G. Mosquera, G. García, F. Marquínez, V. Riera, Angew. Chem. Int. Ed. 44, 102–105 (2004). 34. J. Vignolle et al., J. Am. Chem. Soc. 129, 978–985 (2007). 35. M. J. Menu et al., J. Organomet. Chem. 372, 201–206 (1989). 36. T. Kato, H. Gornitzka, A. Baceiredo, A. Savin, G. Bertrand, J. Am. Chem. Soc. 122, 998–999 (2000).

g

Hu et al., Science 383, 81–85 (2024)

RE FERENCES AND NOTES

p

p-lone pair of 4[NTf2] coordinates to the Ag center, giving rise to 6[NTf2] with a 63% yield. X-ray diffraction analysis of single crystals of 6[NTf2] (fig. S51) unequivocally shows Ag(1) orienting toward the C(1) p lone pair, marked by a C(1)–Ag(1) bond length of 2.684(15) Å. Moreover, discernible Rh–Ag and arene–Ag interactions manifest, as denoted by the Rh(1)–Ag(1), C(2)–Ag(1), and C(3)–Ag(1) separations, which stand at 2.849(6) Å, 2.634(8) Å, and 2.587(8) Å, respectively. The p-coordination between C(1) and Ag(1), supported by DFT calculations (figs. S68 and 69), stands in contrast to the conventional s-coordination seen in NHC silver complexes (45). In a further notable development, 4[BPh4] undergoes a reaction with AdNC, a reagent known for [1+1] cross-coupling reactions with ambiphilic or electrophilic singlet carbenes (20), yielding 7[BPh4] at a 73% yield, as corroborated by the x-ray structure delineated in fig. S52. Although similar results, such as the emergence of a ketenimine, were recorded for cyclic(alkyl)(amino)carbenes (CAACs) (46) possessing the s2p0 electronic state and 4[BPh4], our computational exploration reveals distinct mechanistic scenarios for a CAAC and 4 when reacting with isocyanide (figs. S70 to 73). This distinction is driven by the specific electronic states. Over three decades have passed since the pioneering work by Bertrand (8) and Arduengo (9) on the isolation of stable singlet carbenes. These molecules have since become indispensable tools in chemistry and have even found applications in medicine and materials science. The molecular design principles and synthetic strategies we present here offer a clear roadmap for the creation and stabilization of singlet carbenes with a reversed s0p2 electronic state and the capacity for s-accepting and p-donating behavior. Given their distinct electronic properties, we expect that these singlet carbenes can be harnessed more broadly for the activation of small molecules and for fine-tuning the physical and chemical properties of transition-metal complexes to suit specific applications.

RES EARCH

PEROVKSITES

Supramolecular assembly of blue and green halide perovskites with near-unity photoluminescence Cheng Zhu1,2,3†, Jianbo Jin4†, Zhen Wang1, Zhenpeng Xu1, Maria C. Folgueras1,2,3, Yuxin Jiang2,4, Can B. Uzundal2,4, Han K. D. Le2,4, Feng Wang2,3,5, Xiaoyu (Rayne) Zheng1,6, Peidong Yang1,2,3,4* The metal-halide ionic octahedron is the optoelectronic unit for halide perovskites, and a crown ether–assisted supramolecular assembly approach can pack various ionic octahedra into tunable symmetries. In this work, we demonstrate near-unity photoluminescence quantum yield (PLQY) blue and green emission with the supramolecular assembly of hafnium (Hf) and zirconium (Zr) halide octahedral clusters. (18C6@K)2HfBr6 powders showed blue emission with a near-unity PLQY (96.2%), and green emission was also achieved with (18C6@K)2ZrCl4Br2 powders at a PLQY of 82.7%. These highly emissive powders feature facile low-temperature solution-based synthesis conditions and maintain high PLQY in solution-processable semiconductor inks under ambient conditions, and they were used in thin-film displays and emissive three-dimensional–printed architectures that exhibited high spatial resolution.

1 of 8

,

5 January 2024

y

Zhu et al., Science 383, 86–93 (2024)

We explored a supramolecular synthetic route in which 18C6 greatly increased the solubility

y g

*Corresponding author. Email: [email protected] †These authors contributed equally to this work.

Crown ether–assisted supramolecular assembly

y

Department of Materials Science and Engineering, University of California Berkeley, Berkeley, CA 94720, USA. 2Materials Sciences Division, Lawrence Berkeley National Laboratory, Berkeley, CA 94720, USA. 3Kavli Energy NanoScience Institute, Berkeley, CA 94720, USA. 4Department of Chemistry, University of California Berkeley, Berkeley, CA 94720, USA. 5Department of Physics, University of California Berkeley, Berkeley, CA 94720, USA.USA. 6Energy Technologies Area, Lawrence Berkeley National Laboratory, Berkeley, CA 94720, USA.

g

1

blocks in the crystal structure (19–21), and, given the ionic nature and high chemical tunability of halide perovskite structures (22), different compositions and packing geometries of [MX6]n− could be explored for light-emission applications. The vacancy-ordered double perovskite (A2MX6 phase) has been proposed to incorporate tetravalent metal cation octahedra, such as [TeX6]2− (23–25), [SnX6]2− (26, 27), and [PtX6]2− (28, 29). Although the [PbX6]4− octahedra are corner-shared in all three dimensions in the prototypical CsPbX3 structure (30), the [MX6]2− octahedra in the A2MX6 phase are isolated because a vacancy occupies every other M site in the crystal structure (23). A few [MX6]n− emitters as well as some non-octahedral emitters (31) with high PLQY (~95%) have been identified with yellow emission, such as [SnX6]4− (32). However, emission of high PLQY with shorter wavelengths is still very rare. The isolated nature of the octahedra affects their optoelectronic properties in that the strong coupling of the exciton with lattice vibrations greatly lowers the energy level of the exciton and forces it into transient self-trapped exciton (STE) states with a range of self-trapped energy levels (33, 34). As a result, the A2MX6 systems generally have broadband emissions with a large Stokes shift. Although the A2MX6 phase has been widely studied in various compositions, octahedra with Hf4+ or Zr4+ centers, especially [HfBr6]2− and [ZrBr6]2− octahedra, have rarely been the subject of research (35, 36), even though they have interesting optoelectronic properties. Cs2HfBr6 crystals have a blue emission with the PL peak at 435 nm (35), and colloidal Cs2ZrBr6 nanocrystals have been demonstrated to have a green emission with a PLQY of 45% (36). There are several reasons why they are less explored. Theoretical (37) and experimental studies (38) have shown that the Hf4+ and Zr4+ metal centers are extremely air- and moisture-

p

B

lue and green emission with high photoluminescence quantum yield (PLQY) is at the forefront of solid-state lighting and color display research. Although Si and Zn codoped GaN can exhibit a PLQY of 90% (1), these covalent semiconductors require high purity to prevent rapid nonradiative recombination at crystal structure defects (2–5) and rely on solid-state synthesis at elevated temperatures near 1000°C (6). As an alternative to covalent semiconductors, ionic halide perovskites have received attention given their high optical absorption coefficient (7), tunable bandgap (8–10), high defect tolerance (11, 12), and efficient photo- and electroluminescence (13, 14). For example, the blue and green emissive colloidal CsPbClxBr3-x quantum dots have exhibited PLQY values of ~80% (15, 16). In addition, low-dimensional halide perovskites like the n = 1 Ruddlesden-Popper phase (C6H5CH2NH3)2PbBr4 show blue emission with a PLQY of 79% (17). Despite the notable optoelectronic properties of lead-based halide perovskites, the toxicity of lead and the complex colloidal synthesis complicate largescale applications. Moreover, suitable ligands are still needed to prevent aggregation of these low-dimensional nanostructures during use (18). Recent studies have revealed that the optoelectronic properties of halide perovskites stem from the [MX6]n− (where M is a metal cation and X is a halide anion) fundamental building

sensitive in the A2MX6 phase. Their synthesis requires the vertical Bridgman-Stockbarger method at ~1000°C in sealed quartz ampoules (36, 39, 40). Finally, it is difficult to prepare high-purity samples that do not contain a secondary impurity, such as CsBr (36). Thus, a new methodology is needed for the synthesis of more stabilized and purer solid phases containing the [HfBr6]2− or [ZrBr6]2− octahedra. Recently, we proposed a general crown ether– assisted supramolecular assembly approach for tetravalent metal octahedra (41). Two crown ether@alkali metal complexes can sandwich a tetravalent metal octahedron into a (crown ether@A)2MX6 dumbbell structural unit. The composition of the dumbbell structural unit is highly tunable, with crown ether = 18-crown-6 (18C6) or 21-crown-7 (21C7); A = Cs+, Rb+, or K+; M = Te4+, Sn4+, Se4+, Ir4+, Pt4+, Zr4+, Hf4+, or Ce4+; and X = Cl−, Br−, or I−. In this work, we extended this general supramolecular assembly approach to [HfBr6]2− octahedra to achieve a structure with formula (18C6@K)2HfBr6 that features blue emission with near-unity (96.2%) PLQY. We also optimized the synthetic route by replacing the challenging high-temperature solid-state synthesis with a low-temperature organic solution– based synthesis. Moreover, an efficient green emission was also achieved by tuning the composition of the (crown ether@A)2MX6 dumbbell structural unit. (18C6@K)2ZrCl4Br2 demonstrated green emission with 82.7% PLQY. By studying the photophysics of the supramolecular assembled samples, we could attribute the emission to STE states and observed a very strong electron-phonon coupling constant (represented by the Huang-Rhys parameter) of >90 for (18C6@K)2HfBr6. The supramolecular assembled samples had longer PL lifetimes (in the microsecond timescale) compared with those of other halide perovskite systems that reflected a low rate of nonradiative recombination. The structural integrity and impressive optical properties of the supramolecular assembled solid powders were further maintained by generating a powder suspension in nonpolar organic solvents, such as dichloromethane (DCM), to create an ink system. Polystyrene (PS) polymer was dissolved into the ink to further increase the solution processability. We used these inks to fabricate thin films through fast solvent evaporation. In combination with a digitally controlled excitation source, the (18C6@K)2HfBr6/PS composite thin film could be used as a display with bright color contrast and fast response time. A solution-processable ink also allowed three-dimensional (3D) printing of the powders into various blue-, green-, and dual-color– emitting structures.

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

p g y y g

Zhu et al., Science 383, 86–93 (2024)

5 January 2024

ature to 200°C by combining mechanical forces with heat to facilitate solid-state diffusion. Details of the synthesis for (18C6@K)2HfBr6 and K2HfBr6 can be found in the supplementary materials. The crystal structure of (18C6@K)2HfBr6 was determined from single-crystal x-ray diffraction (SCXRD). (18C6@K)2HfBr6 crystallized in  space group with lattice parameters of the R3 a = 14.1332 Å and c = 21.0189 Å (Fig. 1A and table S1). The (18C6@K)2HfBr6 dumbbell structural unit belongs to the S6 point group, where two K+ cations and the Hf4+ cation sit on the S6 axis. The sixfold symmetry axis of the 18C6 and the S6 axis of the Oh-symmetric [HfBr6]2− octahedron were aligned (Fig. 1B). The K2HfBr6

crystals were face-centered cubic (fcc) (Fig. 1C) (42, 43), in which the [HfBr6]2− ionic octahedra were charge balanced by the surrounding K+ cations (Fig. 1D). The purity of the (18C6@K) 2 HfBr 6 and K2HfBr6 powders was investigated with powder x-ray diffraction (PXRD) (Fig. 1E). The PXRD pattern of the (18C6@K)2HfBr6 powders matched with the calculated pattern generated from the single-crystal structure with no visible diffraction peaks from impurities. The quality of the PXRD pattern for the K2HfBr6 powders was much lower because of their extreme air sensitivity. The measurement had to be collected in 5 min with an inert atmosphere 2 of 8

,

of the KBr and HfBr4 precursors in polar organic solvents for low-temperature solutionbased synthesis. A clear precursor solution was obtained with acetonitrile (ACN) at 80°C with the concentration of 4 mM for 18C6 and KBr and 2 mM for HfBr4. Our previous study of the supramolecular assembly of tetravalent metal octahedra (41) indicated that a (18C6@K)2HfBr6 dumbbell structural unit was formed in ACN. We grew (18C6@K)2HfBr6 powders and single crystals using the antisolvent crystallization method (39). K2HfBr6 powders were also synthesized by using a modified solid-state synthesis method. We increased the purity and decreased the synthesis temper-

diffraction patterns. K2HfBr6 showed quite poor PXRD quality because of its poor stability during measurement. a.u., arbitrary unit. (F and G) Band structure and corresponding total pDOS of (18C6@K)2HfBr6 (F) and K2HfBr6 (G). When [HfBr6]2− octahedra were assembled in the supramolecular approach, the dispersion of the bands decreased, and 18C6 contributed to the VB.

y

Fig. 1. Two assemblies of the [HfBr6]2− ionic octahedron. (A and B) The rhombohedral unit cell (A) and the dumbbell-shaped structural unit (B) of (18C6@K)2HfBr6. (C and D) The Fm-3m unit cell [from the Open Quantum Materials Database (OQMD) (36, 37)] (C) and the isolated [HfBr6]2− ionic octahedron building block (D) of K2HfBr6. (E) The PXRD patterns for synthesized (18C6@K)2HfBr6 and K2HfBr6 powders and the calculated

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

(E) The CIE 1931 chromaticity diagram for the emission of (18C6@K)2HfBr6 powders and (18C6@K)2ZrCl4Br2 powders. “B” stands for the blue emission of (18C6@K)2HfBr6, and “G” stands for the green emission of (18C6@K)2ZrCl4Br2. The coordinates for the emission colors of (18C6@K)2HfBr6 and (18C6@K)2ZrCl4Br2 were (0.17438, 0.16922) and (0.30597, 0.41533), respectively.

p

Fig. 2. Blue emission with near-unity PLQY (96.2%) from (18C6@K)2HfBr6 powders and green emission with a PLQY of 82.7% from (18C6@K)2ZrCl4Br2 powders. (A) (18C6@K)2HfBr6 powders under white lamp and 254-nm UV excitation. (B) PL and PLE spectra of (18C6@K)2HfBr6 powders. (C) (18C6@K)2ZrCl4Br2 powders under white lamp and 302-nm UV excitation. (D) PL and PLE spectra of (18C6@K)2ZrCl4Br2 powders.

g 3 of 8

,

Compared with K2HfBr6, (18C6@K)2HfBr6 had greatly enhanced emission intensity. Figure 2A shows the extremely bright blue emission of (18C6@K)2HfBr6 powders under 254-nm ultraviolet (UV) excitation. The photoluminescence (PL) spectrum of (18C6@K)2HfBr6 powders was measured at 275-nm excitation (Fig. 2B). The powders had a blue emission centered at 445 nm (2.79 eV), and the full width at half maximum (FWHM) was 0.73 eV. Photoluminescence excitation (PLE) spectra revealed a large Stokes shift (1.35 eV). The emission intensity of the (18C6@K)2HfBr6 powders was quantified with PLQY measurement, and a near-unity value of 96.2 ± 1.2% was obtained for the (18C6@K)2HfBr6 powders over six measurements from two batches of samples (fig. S5). The specific value for each mea-

y

5 January 2024

Optical characterization of the blue and green emitters

surement is shown in table S2. By contrast, the PLQY of K2HfBr6 powders was 12.8% (fig. S6). K2HfBr6 also had an even larger Stokes shift (1.42 eV) and broader emission, with a peak emission wavelength of 457 nm and a FWHM of 0.90 eV (fig. S7). The color purity of the emission from the [HfBr6]2− octahedra was also enhanced by the supramolecular approach. Figure S8 shows the emission color of (18C6@K)2HfBr6 and K2HfBr6 powders on the CIE 1931 chromaticity diagram. (18C6@K)2HfBr6 had a much purer blue emission color compared with K2HfBr6. The [ZrBr6]2− units enabled high-PLQY green emissions.Upon290-nmexcitation,(18C6@K)2ZrBr6 had a PL peak at 547 nm, and the FWHM of the PL was 0.69 eV (fig. S9). For the same excitation wavelength, the PL peak of K2ZrBr6 was at 560 nm, and the FWHM of the PL was 0.70 eV (fig. S10). The PLQY of (18C6@K)2ZrBr6 was 49.8% (fig. S11), which was slightly greater than the PLQY of K2ZrBr6 (46.3%) (fig. S12). Although the peak position of the PL spectrum was in the green region, simply analyzing the peak emission wavelength was insufficient given the broadness of the STE-based emission because this crystal actually produced a yellow-green emission color (fig. S13). Given the great chemical tunability of the dumbbell structural unit, an alloying approach at the halide site was proposed to achieve a purer green emission with near-unity PLQY. For CsPbX3 (where X = Cl−, Br−, or I−) nanocrystals, the emission color can be easily controlled by tuning the halide composition (44); introducing Cl− in the halide site may generate a shorter wavelength emission color. By carefully tuning the

y g

Zhu et al., Science 383, 86–93 (2024)

The conduction band (CB) of (18C6@K)2HfBr6 was composed of Hf 5d and Br 4p orbitals, as was the CB of K2HfBr6. However, the valence band (VB) compositions were quite different in these two materials. The VB of K2HfBr6 was mainly composed of the Br 4p orbital, but 18C6 contributed to the VB of (18C6@K)2HfBr6. Thus, the 18C6 molecules were electronically coupled to the [HfBr6]2− octahedra, which indicates that the entire (18C6@K) 2HfBr6 dumbbell building block became a new electronic unit. DFT calculations of (18C6@ K)2ZrBr6 showed that the contribution from 18C6 to the VB and the band structures were more discrete than those in K2ZrBr6 (fig. S4).

y

sample holder to prevent the degradation of the powders and measurement of the degradation product. Although the quality of the K2HfBr6 PXRD pattern was not ideal, the most dominant peaks of the fcc K2HfBr6 phase were still identifiable. Moreover, no HfBr4 or KBr diffraction peaks were present (Fig. 1E), which showed that all of the precursor materials transformed into the K2HfBr6 phase. A Raman spectrum of K2HfBr6 further confirmed the presence of the [HfBr6]2− octahedra in the crystal structure (fig. S1). The crown ether–assisted supramolecular approach was generalized to produce other emissive centers. For example, (18C6@K)2ZrBr6 single crystals and powders were successfully synthesized by the same method, and the same crystal structure as (18C6@K)2HfBr6 was obtained (fig. S2 and table S1). Figure S3 shows that phase-pure (18C6@K)2ZrBr6 powders could be obtained with our established solution-based synthesis. K2ZrBr6 powders were also synthesized with the same solid-state method as K2HfBr6. The (18C6@K)2HfBr6 and (18C6@K)2ZrBr6 dumbbell building blocks were also the electronic units of the new crystal. To elucidate the effect of 18C6 on the electronic structures of the assembled [HfBr6]2− octahedra, density functional theory (DFT) calculations were performed on (18C6@K)2HfBr6 (Fig. 1F) and K2HfBr6 (Fig. 1G) to determine their electronic band structures and partial electronic density of states (pDOS). The electronic bands of (18C6@K)2HfBr6 were less dispersive compared with K2HfBr6 because the [HfBr6]2− octahedra were more separated in (18C6@K)2HfBr6.

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

Fig. 3. Photophysical analysis of (18C6@K)2HfBr6 and (18C6@K)2ZrCl4Br2. (A) PL spectra of (18C6@K)2HfBr6 powders at 4, 50, 100, 150, 210, and 293 K. (B) FWHM of the PL spectra of (18C6@K)2HfBr6 powders at different temperatures, with the orange and teal solid lines denoting the least-square fit to Eq. 1 at low (4 to 190 K) and high (190 to 293 K) temperature ranges, respectively. (C) PLE spectroscopy of (18C6@K)2HfBr6 powders. (D) Normalized PL decay curves of (18C6@K)2ZrCl4Br2 and (18C6@K)2ZrBr6 single crystals.

p g y 4 of 8

,

5 January 2024

cluding their large Stokes shift and broadband nature, are primarily attributed to the electronphonon coupling effect. To unravel the STE emission mechanism, we performed lowtemperature PL measurements to examine the electron-phonon coupling in (18C6@K)2HfBr6. With increasing temperatures, the PL peak gradually broadened, and the peak was slightly red-shifted, indicating greater phonon participation at higher temperatures (Fig. 3A). A small shoulder peak at ~550 nm was present that was especially distinct at lower temperatures, which we attribute to the Zr impurity in the HfBr4 precursor (35). Inductively coupled atomic emission spectroscopy revealed an ~0.5 atomic % (at %) ZrBr4 impurity in the asobtained HfBr4 precursor, and that there was an ~0.6 at % Zr4+ impurity in the synthesized (18C6@K)2HfBr6 single crystals (table S5). To deconvolve the emission from (18C6@K)2ZrBr6 impurities, a two-peak Gaussian fitting was applied to the PL spectrum at each temperature. Figure S22 shows an example at 4 K. The FWHMs of the (18C6@K)2HfBr6 peaks were obtained from the Gaussian fittings and are summarized in table S6. The temperature dependence of the FWHM of the emission peak was modeled using the theory of Toyozawa (45), which applies a configuration coordinate model to explain the broadening of the emission originating from electron-phonon

y

Zhu et al., Science 383, 86–93 (2024)

system. The (101) and (110) diffraction peaks of (18C6@K)2ZrCl4Br2 were slightly shifted to larger 2q values compared with the corresponding PXRD peaks of (18C6@K)2ZrBr6, which suggests smaller lattice constants (fig. S21). Figure 2C shows the bright green emission of (18C6@K)2ZrCl4Br2 powders under 302-nm UV lamp excitation. The PL spectrum of (18C6@K)2ZrCl4Br2 powders was measured at 295-nm excitation (Fig. 2D). The green emission had a similar Stokes shift (1.36 eV versus 1.35 eV) and FWHM (0.80 eV versus 0.73 eV) compared to the blue emission of the (18C6@K)2HfBr6 powders, which suggests similar emission properties of the Hf and Zr metal centers in the supramolecular assembly materials system. The PLQY of the emission from (18C6@K)2ZrCl4Br2 powders was 82.7 ± 0.9%, which was determined through the measurement of four samples from two batches (table S4). Therefore, we achieved highly emissive powders with blue and green emission colors based on the supramolecular assembly approach. The blue and green colors of the emissions from (18C6@K)2HfBr6 and (18C6@K)2ZrCl4Br2, respectively, are summarized in the CIE 1931 diagram (Fig. 2E). Next, we conducted a comprehensive photophysics analysis to confirm and gain deeper insights into the STE emission mechanism that underlies these blue and green emissions. The distinctive features of STE emissions, in-

y g

KCl/KBr and ZrCl4/ZrBr4 precursor ratio in the synthesis, the Cl−/Br− ratio in the obtained (18C6@K)2ZrX6 dumbbell structural unit can be precisely controlled. As expected, a larger Cl − /Br − ratio created a more blue-shifted PL (figs. S14 and S15). For example, (18C6@K) 2ZrCl3Br3 and (18C6@K)2ZrCl4Br2 had green emission with PL peaks at 534 and 530 nm, respectively. Increasing the Cl− content to a composition of (18C6@K)2ZrCl4.5Br1.5 changed the PL color to a cyan (bluish green) color. The established halide site alloying approach not only generated a purer green emission color but also boosted the PLQY of the emission to near-unity. For the Cl−/Br− ratio from 1:1 to 2:1 to 3:1, the PLQYs were 69.1, 82.7, and 87.0%, respectively (figs. S16 and S17). Because the 2:1 Cl−/Br− ratio composition had both pure-green emission color and high PLQY, we selected (18C6@K)2ZrCl4Br2 for detailed studies of green emission. (18C6@K)2ZrCl4Br2 single crystals were synthesized by controlling the Cl−/Br− precursor ratio to be 2:1. The formula of (18C6@K)2ZrCl4Br2 was determined by SCXRD (Cl−:Br− = 4.3:1.7) (fig. S18 and table S3) and energy-dispersive x-ray spectroscopy (EDX) elemental mapping (Cl−:Br− = 4.1:1.9) (fig. S19). The Cl and Br atoms were perfectly miscible in the crystal structure. PXRD of the (18C6@K)2ZrCl4Br2 powders (fig. S20) also indicated that this composition was a phase-pure

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

coupling. The FWHM depends on the Boltzmann constant kB, the effective phonon energy Eph, the temperature T, and the Huang-Rhys electronphonon coupling parameter S (46) pffiffiffi FWHM ¼ 2:36 SEph

A

rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Eph coth 2kB T

ð1Þ

The relation between FWHM and temperature is shown in Fig. 3B. Analyzing the data according to Eq. 1 yielded a coupling factor S1 = 92.2 ± 3.6 and an effective phonon energy Eph1 = 21.4 ± 0.5 meV. This phonon mode corresponded to the asymmetric stretching mode (Eg) of the [HfBr6]2− octahedra, which was observed at 20.4 meV (164.5 cm−1) in the Raman spectrum (fig. S23). However, this pho-

non mode was only responsible for STE formation up to 190 K. For temperatures >190 K, a higher energy phonon mode dominated STE formation. Shifting the zero temperature of Eq. 1 by 190 K, a second fit could be obtained with a coupling factor S2 = 108.8 ± 12.4 and an effective phonon energy Eph2 = 25.8 ± 1.6 meV. This phonon mode corresponded to the sym-

Solution Processibility with Polymers Facile casting Single crystals

Thin FIlm +Polystyrene +DCM DMD chip

Powders

Mild stirring and Sonicating Programmable Display

p

Bulk Materials

Processible Inks

PLQY: 90.77%

C

PLQY: 75.01%

D

PLQY: 80.29%

E

PLQY: 68.99%

g

B

y y g

F

G y ,

Fig. 4. Solution processability and display application of highly emissive blue and green semiconductor inks. (A) Schematic illustrating the thin-film fabrication method and the display application. The inks were formed by mixing (18C6@K)2HfBr6 or (18C6@K)2ZrCl4Br2 powders and PS in DCM. Thin films were obtained by drop casting, and they demonstrated programmable display capability. DMD, digital micromirror device. (B) (18C6@K)2HfBr6/PS-DCM ink under white light and 254-nm UV lamp excitation. (C) (18C6@K)2ZrCl4Br2/PS-DCM ink under white light and Zhu et al., Science 383, 86–93 (2024)

5 January 2024

302-nm lamp excitation. (D) (18C6@K)2HfBr6/PS composite thin film under white light and 254-nm UV excitation. (E) (18C6@K)2ZrCl4Br2/PS composite thin film under white light and 302-nm UV excitation. The scale bars for (B) to (E) are 1 cm. The PLQYs of all samples are shown in the photos. (F) Image of the “Cal” logo blue emission on the (18C6@K)2HfBr6/PS composite thin film. (G) Snapshots of a video showing the alphabet, A to Z, with 0.1 s per letter on the (18C6@K)2HfBr6/PS composite thin film. The scale bars for (F) and (G) are 3 mm and 4 mm, respectively. 5 of 8

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

y ,

6 of 8

y g

The high PLQY of the blue and green emission colors in the powders were preserved in thin films, which would enable various optoelectronic device applications (54–56). Because the powders were stable in nonpolar organic solvents, they could be evenly dispersed into solution to form inks. We used DCM because its low boiling point (39.6°C) leads to high volatility for drying films, and we added PS to create inks suitable for drop casting or spin coating by increasing the viscosity (Fig. 4A) (57, 58). The image of (18C6@K)2HfBr6/PS ink under a white lamp (Fig. 4B) shows that a uniform white suspension was achieved that exhibited a bright blue emission under 254-nm excitation. The emission was solely from the (18C6@K)2HfBr6 powders in the ink because the shape of the PL spectrum was the same as the PL shape of the powders (fig. S26). The solution PLQY was 90.8% (Fig. 4B and fig. S27), which was only 5.5% less than the powder PLQY. This reduction was expected because DCM and PS do not absorb strongly in the blue color wavelength region (fig. S28), and the suspended powders in the ink could cause losses through scattering. The (18C6@K)2ZrCl4Br2/PS ink also preserved the green emission of the (18C6@K)2ZrCl4Br2 powders with a solution PLQY of 75.0% (Fig. 4C and fig. S29).

y

5 January 2024

Blue-green dual-color display and 3D printing

The inks could be drop casted under ambient conditions, and, after rapid solvent evaporation, a uniform thin film forms (Fig. 4, A, D, and E). PXRD patterns of the (18C6@K)2HfBr6/PS composite (fig. S30) and the (18C6@K)2ZrCl4Br2/PS composite (fig. S31) showed that the structural integrity of the powders was preserved in the PS matrix. Scanning electron microscopy (SEM) imaging of the (18C6@K)2HfBr6 powders and the (18C6@K)2HfBr6/PS composite thin-film surface (fig. S32) indicated that the submicrometersized powders were uniformly dispersed. Crosssectional SEM imaging of the thin film (fig. S33) proved the presence and uniformity of the powders across the thin film. Under UV irradiation, (18C6@K)2HfBr6/PS and (18C6@K)2ZrCl4Br2/PS composite thin films showed bright blue and green emissions, respectively (Fig. 4, D and E). The shapes of the PL spectra of the thin films were the same as those for the powders (figs. S25 and S34), and the PLQYs were 80.3% (fig. S35) and 69.0% (fig. S36) for blue- and greenemitting composites, respectively. The stability of the air-sensitive Hf and Zr octahedral clusters was further enhanced in the PS polymer composite. Both Cs2HfBr6 and Cs2ZrBr6 double-perovskite structures are predicted to be thermodynamically unstable in the presence of water and oxygen (37), and we found that K2HfBr6 and K2ZrBr6 powders turn from a white to a brownish color after a few minutes of air exposure and became nonemissive. By contrast, the (18C6@K)2HfBr6/PS and (18C6@K)2ZrCl4Br2/PS composites maintained their blue and green emission colors, respectively, after 1 month of storage in the air (fig. S37). The air-stable PS polymers along with the hydrophobic crown ethers could greatly protect the air-sensitive Hf and Zr metal emission centers. We explored display applications of the powderPS composite thin films. A digital mirror device with a pixel resolution of 2560 by 1440 sequentially patterned 250-nm UV light through projection optics onto the (18C6@K)2HfBr6/PS composite thin film with a spot size of 6.9 by 3.9 mm at a frame rate of 60 Hz (schematic of the process is illustrated in Fig. 4A). An emissive blue “Cal” logo was illuminated on the thin film with dimensions 3.8 mm in height and 4.7 mm in width (Fig. 4F). The logo exhibited high luminosity characterized by sharply defined edges. To further demonstrate dynamically changing display luminescence, we illuminated the alphabet sequence (from A to Z) onto the thin film with a fast flipping rate (0.1 s per letter). A video of 2.6 s was recorded (movie S1). Although the duration of each letter was very short, the blue emission with the shape of the letters was sharp and bright, as illustrated in the snapshot photos (Fig. 4G). The size of the letters was only 3.1 mm in width and 3.9 mm in length, but every feature of the letters was clearly visible with similar emission intensity

g

Zhu et al., Science 383, 86–93 (2024)

emitting diode (OLED) community has used a xenon lamp to simulate solar irradiation, dissolving Ir complexes in deuterated toluene for reference measurements of green and blue emission (51, 52). To ensure a fair comparison, we applied identical irradiation energy density (62 mW/cm2) and temperature (35°C), and deuterated toluene was used to disperse the (18C6@K)2HfBr6 and (18C6@K)2ZrCl4Br2 powders. Figure S25 shows the PL intensity decay of the (18C6@K)2HfBr6 and (18C6@K)2ZrCl4Br2 samples under continuous irradiation. Both decay trends could be accurately described by the integrated rate law for the first-order reaction [ln(It/I0) = −kt]. The rate constants of photodegradation were estimated to be 5.1 × 10−3 h−1 and 3.0 × 10−3 h−1 for (18C6@K)2HfBr6 and (18C6@K)2ZrCl4Br2, respectively. Notably, even under stringent irradiation conditions, the PL intensities of (18C6@K)2HfBr6 and (18C6@K)2ZrCl4Br2 decreased to 80% after 43 and 73 hours, respectively. These findings underscore the superior photostability of the supramolecular assembled samples compared with most Ir complexes, rivaling the bestreported green-emitting fac-[Ir(ppy)3] reference (k = 2.6 × 10−3 h−1) (51, 52). Previous studies on the photodegradation of the Ir complexes, such as Ir(ppy)3 and Ir(piq)3, have identified singlet oxygen attack and interaction of the excited-state molecule with its local environment as primary degradation pathways (53).

p

metric stretching mode (A1g) of the [HfBr6]2− octahedra at 25.1 meV (202.5 cm−1). The large Huang-Rhys factor S in both scenarios indicated a very strong electron-phonon coupling in this material. For example, S for CsPbX3 (where X = Br− or I−) is 80%, making it favorable for patterning, display,

52. H.-H. Kuo et al., J. Mater. Chem. C 6, 10486–10496 (2018). 53. S. Schmidbauer, A. Hohenleutner, B. König, Beilstein J. Org. Chem. 9, 2088–2096 (2013). 54. W. Hui et al., Science 371, 1359–1364 (2021). 55. F. Mathies et al., J. Mater. Chem. A 4, 19207–19213 (2016). 56. S. X. Li et al., ACS Appl. Mater. Interfaces 13, 31919–31927 (2021). 57. M. Kaseem, K. Hamad, Y. G. Ko, Eur. Polym. J. 79, 36–62 (2016). 58. Y. Tu et al., J. Mater. Chem. 20, 1594–1599 (2010). 59. H. Cui et al., Science 376, 1287–1293 (2022). 60. R. Hensleigh et al., Nat. Electron. 3, 216–224 (2020).

g

Discussion

and printing applications. The powders with blue and green emissions were also highly compatible with the 3D printing technology. The supramolecular assembly approach for halide perovskite building block catalyzes further investigation into the synthesis and characterization of supramolecular assembled functional materials, laying the foundation for substantial progress in the field.

p

architectural models of the Eiffel Tower (Fig. 5B), after excitation at 254 nm, showed their respective blue and green colors (Fig. 5C). The dimensions of the two Eiffel Towers were within a few centimeters, with high-resolution spatial features (Fig. 5B). The submicrometer scale of these powders and a printing layer thickness of 40 mm enabled even distribution throughout each layer and ensured a homogeneous emission color profile across the entire architectural construct. A single 3D-printed structure could also manifest emissions in both blue and green by alternative resins during the printing procedure. An Eiffel Tower design characterized by blue emissions at its upper and lower segments with green emissions in its central region is shown in Fig. 5D, and a second-order hierarchical lattice structure (octet truss) was realized with one half radiating in blue and the other in green (Fig. 5E). Notably, a close-up view of the boundary between these blue- and greenemitting regions within the octet truss structure revealed the high precision in color transition without any color crossover on either side. Twisted (Fig. 5F and fig. S41) and conformal (Fig. 5G and fig. S42) octet truss architectures with dual emissions were also achieved with bright emissions and high structural accuracy. Other complex topologies, such as cuboctahedron, tetrakaidecahedron, octet truss, and Menger sponge with the blue emitter embedded (Fig. 5H and figs. S43 and S44), were also obtained to exhibit the variety of structures that could be printed with the lightemitting ink. These demonstrations served as a proof of concept for integrating emissive ionic powders with 3D printing technology. The potential applications of 3D-printed lightemitting structures are extensive and constantly evolving, ranging from intricate interior ambient-lighting solutions to seamless integration into wearable devices.

RES EARCH

CATALYSIS

Dealuminated Beta zeolite reverses Ostwald ripening for durable copper nanoparticle catalysts Lujie Liu1†, Jiaye Lu2†, Yahui Yang3, Wolfgang Ruettinger3, Xinhua Gao4, Ming Wang5, Hao Lou6, Zhandong Wang6, Yifeng Liu7, Xin Tao8, Lina Li8, Yong Wang9, Hangjie Li1, Hang Zhou1, Chengtao Wang1, Qingsong Luo1, Huixin Wu1, Kaidi Zhang1, Jiabi Ma5*, Xiaoming Cao2,10*, Liang Wang1*, Feng-Shou Xiao1,11* Copper nanoparticle–based catalysts have been extensively applied in industry, but the nanoparticles tend to sinter into larger ones in the chemical atmospheres, which is detrimental to catalyst performance. In this work, we used dealuminated Beta zeolite to support copper nanoparticles (Cu/Beta-deAl) and showed that these particles become smaller in methanol vapor at 200°C, decreasing from ~5.6 to ~2.4 nanometers in diameter, which is opposite to the general sintering phenomenon. A reverse ripening process was discovered, whereby migratable copper sites activated by methanol were trapped by silanol nests and the copper species in the nests acted as new nucleation sites for the formation of small nanoparticles. This feature reversed the general sintering channel, resulting in robust catalysts for dimethyl oxalate hydrogenation performed with supported copper nanoparticles for use in industry.

,

1 of 8

y

5 January 2024

y g

Liu et al., Science 383, 94–101 (2024)

y

*Corresponding author. Email: [email protected] (J.M.); [email protected] (X.C.); [email protected] (L.W.); [email protected] (F.-S.X.) †These authors contributed equally to this work.

We performed a proof-of-concept experiment by comparing the stability of Beta-deAl (Cu/ Beta-deAl; figs. S1 and S2) and amorphous silica–supported Cu NPs (Cu/SiO2, Cu loading content at 3.0 wt %) in methanol vapor. The Cu NP sizes and distributions were initially characterized by scanning transmission electron microscopy (STEM). The fresh Cu/SiO2 (as prepared by calcination in air and prereduction in hydrogen) showed that the NPs had an average size of 3.0 ± 0.8 nm (fig. S3, A and D). After the methanol treatment at 200°C for 12 hours (methanol feeding rate of 0.03 mlliquid min−1 in hydrogen with a flow rate of 15 ml min−1), the Cu NPs on Cu/SiO2 sintered into ones with an average size of 4.0 ± 1.4 nm (fig. S3, B and E), and some NPs were larger than 7.0 nm (fig. S3, C and E). Methanol-triggered Cu sintering is known to occur under similar conditions and leads to the deactivation of Cu catalysts in industrial processes (7, 28–30). However, the Cu NPs on the Cu/Beta-deAl (Cu content at 3.0 wt %) exhibited an opposite phenomenon after the equivalent methanol treatment. The initial average Cu NP size of 5.6 ± 1.6 nm (fresh catalyst; Fig. 1A and fig. S4, A and B) was reduced to 2.4 ± 0.7 nm, and we could not detect NPs larger than 6.5 nm (Fig. 1B and fig. S4, C and D). The Cu content (3.0 wt %) was unchanged after the methanol treatment. These results suggest a redispersion of Cu NPs on the Cu/Beta-deAl in methanol vapor. The redispersion of Cu NPs on Cu/Beta-deAl was further confirmed by in situ x-ray diffraction (XRD) characterization. Compared with the bare Beta-deAl zeolite, the fresh Cu/BetadeAl exhibited peaks at 43.3° and 36.5° that we assigned to metallic Cu and Cu2O phases, respectively (31, 32) (Fig. 1C and fig. S5). Methanol treatment reduced these diffraction peaks until they disappeared after 12 hours, which confirmed the elimination of large Cu or Cu2O particles. These results were consistent with those of the TEM characterizations and supported the reversed Cu sintering on the Cu/Beta-deAl in methanol vapor (Fig. 1D). By contrast, after the equivalent methanol vapor treatment, the diffractions of Cu-related NPs were enhanced on the Cu/SiO2 because of the formation of larger Cu NPs (fig. S6). These results suggest that the Beta-deAl zeolite was crucial for the redispersion of Cu. The Cu NPs on other zeolites,

g

1 Key Lab of Biomass Chemical Engineering of Ministry of Education, College of Chemical and Biological Engineering, Zhejiang University, Hangzhou 310027, China. 2Joint International Research Laboratory of Precision Chemistry and Molecular Engineering, Feringa Nobel Prize Scientist Joint Research Center, School of Chemistry and Molecular Engineering, East China University of Science and Technology, Shanghai 200237, China. 3BASF Advanced Chemicals Co., Ltd., Shanghai 200137, China. 4State Key Laboratory of High-efficiency Utilization of Coal and Green Chemical Engineering, School of Chemistry and Chemical Engineering, Ningxia University, Yinchuan 750021, China. 5 Key Laboratory of Cluster Science of Ministry of Education, School of Chemistry and Chemical Engineering, Beijing Institute of Technology, Beijing 102488, China. 6National Synchrotron Radiation Laboratory, University of Science and Technology of China, Hefei, Anhui 230029, China. 7 Department of Chemistry, Zhejiang University, Hangzhou 310027, China. 8Shanghai Institute of Applied Physics, Chinese Academy of Sciences, Shanghai 201204, China. 9 Center of Electron Microscopy and State Key Laboratory of Silicon Materials, School of Materials Science and Engineering, Zhejiang University, Hangzhou 310027, China. 10 School of Chemistry and Chemical Engineering, Shanghai Jiao Tong University, Shanghai 200240, China. 11Beijing Advanced Innovation Center for Soft Matter Science and Engineering, Beijing University of Chemical Technology, Beijing 100029, China.

Redispersion of the sintered metal NPs into smaller ones on the support surface could efficiently reverse deactivation. Sintering often occurs by the loss of mobile metal atoms from small NPs and diffusion and capture by larger NPs, a process known as Ostwald ripening (18–20). Reversing this process would require activation of the metal sites on large NPs and the capture of these metal species by the support. However, this process requires hybrid supports that have distinguishable interactions with the metal species that cause them to move from a weak-interaction region to a stronginteraction region (9, 21). For example, Cu species migrated from silica to the ceria on the SiO2-CeO2 mixed supports to form smaller clusters (21), but this movement changed the intrinsic catalytic properties of the Cu NPs. The reverse ripening of metal NPs on a homogeneous support should return the sintered catalysts to ones with identical performances to the original catalysts, but achieving this is thermodynamically challenging. Treatment with halogens that coordinate with the metal sites can initiate their emission from bulky NPs and transmission on the support. Such postregeneration of the deactivated catalysts (22–24) has been used in industrial processes, such as oxychlorination for redispersing Pt NPs on alumina support (23), which can generate environmentally unfriendly wastes and corrode the reactor. Ideally, in addition to the postregeneration of deactivated catalysts, the redispersion that occurs under the reaction processes would realize a catalyst with superior durability against sintering (25–27). We report that dealuminated Beta zeolite (Beta-deAl, Si/Al atomic ratio at ~1250; fig. S1) is an efficient support for the reverse ripening of Cu NPs in methanol vapor at 200°C. This feature maintained small Cu NPs (2 to 3 nm) and even redispersed large Cu particles (15 to

Redispersion of Cu NPs

p

M

etal nanoparticles (NPs) on the surface of heterogeneous catalysts encounter reactant and product molecules at high temperatures (1–7), and these conditions often cause the NPs to sinter into larger ones. The resulting loss of surface area and reduced number of active sites can lead to irreversible deactivation (1–3, 8–12), and it costs billions of dollars per year to shut down the reaction process and replace the catalysts after deactivation (13). For metals with low melting temperatures and, hence, higher surface diffusion, such as copper, sintering can lead to deactivation for various hydrogenation (14, 15) and reforming reactions (16, 17).

250 nm) on siliceous Beta-deAl support. We focused on dimethyl oxalate (DMO) hydrogenation, a structure-dependent reaction that has been industrially used for the conversion of coal to ethylene glycol (EG) (28, 29). As a result, the Cu/Beta-deAl catalyst maintains high conversion and selectivity in a continuous reaction for long periods, even during ambientpressure DMO hydrogenation.

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

A

B

5 nm

fresh methanol-treated

2

50 nm

C Cu2O

4

6 8 10 12 14 Diameter (nm)

50 nm

D

Cu

6

fresh

6h 8h Si

10 h

Si

12 h

Beta-deAl 30

40

50

60

HO

OH

HO Si

silanol nest

70

2 (°)

We evaluated Cu/SiO2 and Cu/Beta-deAl in the hydrogenation of DMO in methanol vapor at 200°C with a liquid hourly space velocity (LHSV) of 0.8 hour−1 (mlDMO mlcat−1 hour−1) based on a feeding rate of DMO per milliliter of catalyst. The data that characterize the performances are shown in Fig. 2A, where the initial DMO conversions were maintained at ~40% for evaluating the catalyst durability. Liu et al., Science 383, 94–101 (2024)

5 January 2024

2 of 8

,

Catalytic evaluation

The Cu/SiO2 catalyst showed continuously decreased DMO conversion with reaction time; the conversion of 35.2% after 24 hours dropped to 19.8% after 80 hours and to 9.1% after 110 hours. TEM characterization of the spent Cu/SiO2 catalyst after testing for 80 hours showed an average Cu NP size of 4.4 ± 1.3 nm (fig. S13), which was larger than that of the fresh catalyst (3.0 ± 0.8 nm). In previous studies of DMO hydrogenation, abundant methanol in the reaction system, from the solvent and product, also caused Cu sintering (28–31). In the DMO hydrogenation under equivalent conditions for Cu/Beta-deAl, we observed improved DMO conversion over time; conversion increased from 33.4 to 45.0% in the initial 12 hours and then remained at 45.1 to 48.1% in the test for another 188 hours (Fig. 2A). TEM

y

such as commercial Beta (Si/Al ratio at 13), ZSM-5 (aluminosilicate MFI zeolite with Si/Al ratio at 13), S-1 (siliceous MFI zeolite), and silanol-modified MFI zeolites (ZSM-5-OH and S-1-OH), also readily sintered into larger NPs after the methanol treatment at 200°C (figs. S7 to S12).

y g

Fig. 1. Redispersion of Cu on the Beta-deAl support by methanol treatment. (A and B) High-angle annular dark-field STEM images of fresh Cu/Beta-deAl (A) and Cu/Beta-deAl (B) after the methanol treatment at 200°C for 12 hours. The insets in (B) show Cu NP size distributions of the fresh and methanoltreated catalysts (bottom left) and a high-resolution TEM image of the methanol-treated catalyst (top right). (C) In situ XRD patterns characterizing the change of Cu NPs on the Cu/Beta-deAl during the methanol treatment at 200°C. (D) Illustration showing the reverse Ostwald ripening of Cu NPs on the BetadeAl support. The silanol nest, which is formed after removing the tetracoordinated Al sites from the Beta zeolite framework, usually involves four silanols in a local region with hydrogen bonds between them.

y

6

Si OH

g

Intensity

4h

p

methanol-treated for 2h

characterizations of the spent Cu/Beta-deAl after reaction for 10 hours showed an average Cu NP size of 2.3 ± 0.4 nm (fig. S14) versus 5.6 ± 1.6 nm for the fresh sample. This result confirmed a self-optimization process in the initial reaction. Interestingly, the formed smallsize Cu NPs on the Cu/Beta-deAl were constant during the reaction period, and the average Cu NP size was 2.5 ± 0.7 nm on the spent Cu/BetadeAl after testing for 80 hours (fig. S15). In the reaction with 1,4-dioxane as solvent, the Cu/ Beta-deAl catalyst also exhibited constant performance, with an average Cu NP size of 2.4 ± 0.9 nm after reaction for 120 hours (figs. S16 and S17). We then adjusted the LHSV of DMO for evaluating the catalysts with high or even full conversion of DMO. After pretreatment with methanol vapor at 200°C for 10 hours, the Cu/ Beta-deAl exhibited full DMO conversion under given reaction conditions (LHSV at 0.4 hour−1, 200°C) with EG selectivity >98.5% (Fig. 2B). Increasing the LHSV to 0.8 hour−1 resulted in a lower DMO conversion at 45.4% with methyl glycolate (MG) from semihydrogenation as a dominant product. This incomplete conversion provided a way to evaluate the activity change of the Cu catalysts. In a continuous test for 120 hours, we periodically switched the LHSV between 0.4 and 0.8 hour−1 every 30 hours, and the Cu/Beta-deAl exhibited constant performance in each period. For example, after the test in which the LHSV was at 0.8 hour−1 and was then switched back to 0.4 hour−1, the Cu/Beta-deAl catalyst gave a DMO conversion and EG selectivity similar to those in the initial reaction. By contrast, the Cu/SiO2 showed continuously decreased DMO conversion with reaction time, suggesting a deactivation process (Fig. 2B). The durability of Cu/Beta-deAl was further evaluated under a constant condition (LHSV at 0.4 hour−1, 200°C) for 200 hours. After an activation period of 12 hours, the DMO was fully converted, with EG selectivity up to 98.7% (Fig. 2C). In this case, the selectivity to ethanol and other heavier products was 67 kPa) would benefit the Cu redispersion (fig. S29). A general view on the methanol-triggered Cu sintering includes methanol decomposition to form CO and COinitiated Cu migration (30, 34, 42). However, CO was unable to change the average diameter

of Cu NPs by exposure of Cu/Beta-deAl to a CO atmosphere at 200°C (figs. S30 and S31). The other molecules, such as H2, N2, and water, were also unable to decrease the Cu NP sizes (figs. S30 and S31). Formaldehyde, which was produced from the methanol dehydrogenation (fig. S32), greatly promoted the redispersion of Cu particles, resulting in an average Cu NP size of 2.3 ± 0.9 nm for the formaldehyde-treated Cu/Beta-deAl sample (figs. S30 and S33). 4 of 8

,

Fig. 3. Methanol-triggered structure change of the Cu particles. (A) Normalized Cu K-edge XANES spectra of fresh and methanol-treated Cu/SiO2 and Cu/BetadeAl catalysts. The methanol treatments were performed at 300°C for 12 hours. (B) Fourier-transformed (FT) magnitudes of the experimental Cu K-edge EXAFS spectra. The plot lines with open circles show fitting results of samples. R, bond distance. (C to F) XRD patterns (C) and STEM images [(D) to (F)] characterizing the mixture of Cu powder and Beta-deAl zeolite during the methanol treatment at 200°C for different periods. The insets in (E) and (F) show particle size distribution of small Cu NPs. (G) Illustration showing the formation of small-size Cu NPs on the Beta-deAl zeolite from the physically mixed commercial Cu powder. d, diameter.

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

,

5 of 8

y

5 January 2024

y g

Liu et al., Science 383, 94–101 (2024)

Previous studies have shown that the Cu on relatively inert supports such as silica sinters through particle migration and coalescence under thermal conditions but tends to proceed by Ostwald ripening under chemical triggering conditions (8, 20, 43), such as the hydrogenation in methanol vapor. The ripening route involves the formation and diffusion of metal atoms, complex intermediates, or both (20, 44, 45). A support with strong interactions might trap the metal atoms or complex intermediates. We used synchrotron vacuum ultraviolet photoionization mass spectrometry (SVUV-PIMS; fig. S41) to detect Cu species in methanol vapor. In these tests, methanol was introduced to the physical mixture of Cu powder with Beta-deAl, and the possible intermediates were captured through a vacuum channel and detected by mass spectrometry. The signals of methoxyl [mass/charge ratio (m/z) of 31] and formaldehyde (m/z of 30) species from methanol conversion (m/z of 32) appeared (Fig. 4A), as well as other possible species of (CHO)Cu1* (m/z of ~92 and 94), (CH2O)Cu1* (m/z of ~93 and 95), and (CH3O)Cu1* (m/z of ~94 and 96) (Fig. 4B). These results suggest the existence of Cu1-related intermediates during the migration. In addition, the signal intensities for (CHO)Cu1* and (CH3O)Cu1* gradually increased with temperature during the methanol treatment (fig. S42), which is related to the easier formation and faster escape of these intermediates at higher temperatures. These insights are in good agreement with general knowledge about the Ostwald ripening processes with chemicaltriggered metal sintering (8, 20, 44–47), but here we provide direct experimental evidence and also reverse this process with the Beta-deAl support. We also performed density functional theory (DFT) calculations, in which we first considered the desorption of a Cu atom from the Cu NP matrix as the primary step that initiates ripening. The direct detachment of a Cu atom from the matrix is difficult because of the strong Cu–Cu bonding, but the methanol-derived species could coordinate with the surface Cu atom and induce its migration. We considered the formation process of methanol-derived species (Fig. 4C and figs. S43 and S44) and showed

y

The redispersion of Cu on the Beta-deAl zeolite was further confirmed by methanol treatment of a physical mixture containing Beta-deAl and commercial Cu powders (15 to 250 nm; figs. S34 and S35). The XRD patterns recorded the change of Cu crystals in methanol vapor for different periods. The diffraction peaks assigned to Cu crystals decreased in intensity during the treatment period, which suggested a continuous migration of Cu species from the bulky crystals (Fig. 3C). A STEM image of the fresh sample showed only bulky Cu particles with undetectable small Cu NPs (Fig. 3D). The small-size Cu NPs were observed on the Beta-deAl support when methanol was introduced to the mixture at 200°C (an initial stage of 5 hours; fig. S36). After the methanol treatment at 200°C for 70 hours, more Cu NPs appeared with an average particle size of 2.1 ± 0.4 nm (Fig. 3E and fig. S37; bulky ones were not included in the calculation), whereas the bulky Cu particles were still observed but their sizes were smaller than those of the parents. After the treatment in methanol vapor for 190 hours, the Cu peaks were undetectable in

Mechanism study

that the production of *CH3O, *CHO, and *CO was exothermic, whereas the production of other species was endothermic (figs. S43 and S44 and table S5). Lower energy inputs for the Cu detachment and migration than for the ligand desorption for (CHO)Cu1* and (CH3O) Cu1* suggest that the migration is more favorable in the reverse ripening process (fig. S43). For example, the *CHO exhibited an energy input of 0.42 eV for Cu detachment and migration from the Cu matrix to the silanol nest (four silanols in a local region; fig. S45), whereas direct desorption of the *CHO without Cu detachment required 1.03 eV (fig. S43). These results indicated that the complexes would preferentially detach the Cu atoms from the surface of the Cu matrix to form the intermediates that were ready for the migration, which is consistent with the results of SVUVPIMS. In this case, lower energy inputs for (CHO)Cu1* migration from the Cu matrix to the silanol nest than for (CH3O)Cu1* suggest that (CHO)Cu1* is a preferential intermediate in the reverse ripening process. The *CHO could be further dehydrogenated into *CO, but the *CO was unable to assist with the detachment of Cu atoms from the Cu matrix; thus, CO would be directly desorbed at the reaction temperature (fig. S43), which is in good agreement with the experimental results (fig. S30). In addition, the free-energy barrier for the dehydrogenation of *CHO to *CO was slightly lower than that of (CHO)Cu1* moving from the matrix to the Cu/Beta-deAl surface (0.29 eV). These results indicated that *CHO decomposition to CO competes with the detachment of (CHO)Cu1* (fig. S44). We further studied the methanol-triggered Cu detachment by exploring the reactions between the gas-phase Cux+ clusters and methanol with time-of-flight mass spectrometry (48, 49). In our tests, the 63Cu2+ cations were generated by laser ablation of the Cu matrix surface, and then methanol was introduced for reaction. As shown in Fig. 4D, the peak intensity of 63Cu2+ decreased after reaction with methanol and led to the formation of abundant Cu1 species [63Cu1(CH3OH)+]. Increasing the partial pressure of methanol further reduced the Cu2+ and enhanced the Cu1+ signal magnitudes. Although the reactions of 63Cu2+ with CH3OH could not fully represent the types of ligandCu intermediate because of the obvious scale difference between the atomic 63Cu2+ cluster and Cu NP surface, these results strongly supported that methanol triggers Cu–Cu bond cleavage. In sharp contrast, the 63Cu2+ cations were inert toward CO and H2 under similar reaction conditions, which confirms the undetectable ability of these gases to trigger the Cu detachment (Fig. 4D). These results were also in good agreement with the results obtained when Cu/Beta-deAl catalyst was treated with CO and H2 (figs. S30 and S31).

g

Redispersion of bulky Cu particles

the XRD pattern, which confirmed the removal of bulky Cu crystals. A STEM image of the sample showed an average Cu NP size of 2.2 ± 0.5 nm on the Beta-deAl support, with undetectable bulky Cu particles (Fig. 3F and figs. S37 and S38). The equivalent test on the mixture of commercial Cu powders with amorphous silica resulted in unchanged XRD patterns, and small Cu NPs were absent (figs. S39 and S40). Based on these results, we present a schematic illustration showing the formation of small-size Cu NPs on the Beta-deAl zeolite from the bulky Cu crystals (Fig. 3G).

p

We characterized the Cu/SiO2 and Cu/BetadeAl catalysts before and after the methanol treatment at 300°C (the temperature required to amplify sintering or redispersing NPs on the catalysts) by x-ray absorption fine structure spectroscopy. The x-ray absorption nearedge structure (XANES) spectrum of fresh Cu/SiO2 fell between those of Cu foil and Cu2O, which suggested a multivalent Cu0 and Cu+ composition (Fig. 3A) that was consistent with the general feature of the Cu/SiO2 with Cud+-OSiOx interaction (7, 28–31, 34, 43). The methanoltreated Cu/SiO2 was more metallic because the spectrum was near that of the Cu foil that was caused by Cu NPs sintering that destroyed the interaction with silica. Extended x-ray absorption fine structure (EXAFS) analysis (Fig. 3B) showed an increased coordination number (CN) of the Cu-Cu shell (3.5 to 8.2) and the absence of Cu-O or Cu-O-Cu signals after the methanol treatment (table S4). By contrast, the fresh Cu/Beta-deAl exhibited a XANES spectrum near that of Cu foil. After the methanol treatment, the Cu species were partially oxidized, giving a XANES spectrum near that of Cu2O. The EXAFS fitting results showed the CN of the Cu-Cu shell at 9.0 on the fresh Cu/Beta-deAl, which dropped to 2.3 with the CN of the Cu-O shell at 1.5 and the CN of the Cu-O-Cu shell at 1.7 on the methanoltreated sample (Fig. 3B and table S4). Considering that the methanol treatment with hydrogen carrier provided a reductive environment, the formation of more-oxidized Cu species could be attributed to the formation of a Cud+-O-SiOx interaction during the methanol-triggered redispersion of Cu.

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

A

C

B 20000

Intensity

Free energy (eV)

CH3OH

Intensity

2

CH3O* *CHO

100

CH3O

(CH3O)Cu1* (CHO)Cu1*

0.90 eV

1

CH3OH*

0

1.08 eV 1.09 eV 0.81 eV 0.96 eV

CH3OH(g) CH3O* *CHO

HCHO

0.88 eV 0.70 eV

Formation

Migration

-1

30

31

92

32

m /z

93

94

95

96

D

63

E

Cu2+

Gas -phase migration

(CHO)Cu1*(g)

He

0.68 eV 63

Cu1(CH3OH)+

63

+

Cu2H

Intensity

CH3OH, 0.65 Pa

H

H

H

O

O

O

-0.26 eV

OH

silanol nest

Si

Si

Si O O

CH3OH, 1.15 Pa

(CHO)Cu1*(g)

CO

60

70

80

90

100

110

120

130

140

Beta-deAl

Surface migration 0.40 eV 0.22 eV

H

H

H

O

O

O

O O

-0.20 eV

OH

silanol nest

Si

Si

Si

O O

O

Beta-deAl

y

m/z

F

O O

O

g

H2

2

0.95

TS1

Si

0

H

CO(g)+H 2(g)

-1.64 -2.10

Cu1-Cu 3

TS5 -3.72

Cu@Beta-deAl+CO(g)+H2(g)

-3.77

(CHO)Cu1*+Cu3

HCu4+CO(g)

(CHO)Cu1*+HCu4

-3.86 CO(g)

Cu5+CO(g)+H2(g)

y g

(CHO)Cu1*+H(Zeolite)

Cu4-Cu 5 -6

Cu 4-Cu 5

Cu 1-Cu 3

TS4

-1.16

-2

C

(CHO)Cu 1*

-0.57

-0.99

-4

O

TS3 TS2

Free energy (eV)

Reaction Coordinate

m /z

p

Fig. 4. Experimental and theoretical insights of the reverse ripening process. (A and B) Data showing the signals at different m/z regions in the SVUV-PIMS tests that characterize the mixture of Cu powder and BetadeAl during methanol treatment. The background signals were excluded. The possible species corresponding to the peaks are given, where the signal of m/z 31 should be from the fragmentation of methanol. (C) Free energies of (CH3O)Cu1* and (CHO) Cu1* intermediates, and the formation and migration of the Cu1 intermediates from the Cu particles to surface hydroxy groups with the methanol treatment. (D) Time-of-flight mass spectra for reactions of the mass-selected 63Cu2+ cations with methanol, CO, and hydrogen. The reaction time was about 1.1 to 1.2 ms, and the effective reactant gas pressures for CO and H2 were around 1.0 Pa. (E) Illustration showing the gas-phase and surface migration of Cu species on the catalyst surface. (F) Energy profile showing the nucleation process from Cu1 to Cu5 at the silanol nests. TS, transition state; Cux, Cu cluster with x number of Cu atoms.

-5.04 y ,

-8

-10

TS1

TS2

TS3

TS4

TS5

Reaction Coordinate

With respect to the migration of Cu1 intermediates on the zeolite surface, both surface and gasphase migration might occur during the reverse Ostwald ripening process (figs. S46 to S49) (20, 25, 27, 34, 44–46). With the (CHO)Cu1 intermediate as a model, the gas-phase migration would cost 0.68 eV for desorption from the solid surface and −0.26 eV for capture by the silanol nests (Fig. 4E). In the case of surface migration Liu et al., Science 383, 94–101 (2024)

5 January 2024

(Fig. 4E), the first step is the migration of the (CHO)Cu1 intermediate from the Cu cluster to the adjacent terminal silanol on the zeolite surface (0.40 eV). Next, the (CHO)Cu1* migrates on the terminal silanol (0.22 eV) (20) and finally arrives at the silanol nest (−0.20 eV). Slightly lower energy input indicates that the surface migration would be the preferential pathway for (CHO)Cu1* diffusion during the dynamic change of the

Cu species that occurs with the methanol treatment. Figure 4F shows an energy diagram that characterizes the stepwise nucleation process by which the silanol nest captures the (CHO)Cu1* intermediate, which, with a free energy of −1.16 eV, is obviously exothermic. With respect to the interaction between (CHO)Cu1* and the silanols in the nest, a stable Cud+-(OSi)2 linkage 6 of 8

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

18. 19. 20. 21. 22. 23. 24. 25. 26. 27. 28. 29. 30. 31. 32. 33. 34. 35. 36. 37. 38. 39. 40. 41. 42. 43. 44. 45. 46. 47. 48. 49.

AC KNOWLED GME NTS

We thank G. Zhu and F. Chen for assistance with TEM characterization and the Center for High Performance Computing at Shanghai Jiao Tong University for providing the computing resources of the Siyuan-1 cluster. We appreciate use of the Hard X-ray Spectroscopy Beamline (BL11B), the User Experiment Assist System, and the Chinese Academy of Sciences–Shanghai Science Research Center of the Shanghai Synchrotron Radiation Facility as well as TILON Group Technology Limited for data collection. We appreciate being able to obtain the SVUV-PIMS measurements through use of the Atomic and Molecular Physics Beamline (BL09U) at the National Synchrotron Radiation Laboratory in Hefei, China. Funding: This work was supported by the National Key Research and Development Program of China (2022YFA1503502 and 2019YFA0405602); the National Natural Science Foundation of China (22288101, 22241801, 22202176, 22022302, and 92045303); the China Postdoctoral Science Foundation (2021M702802); the Shanghai Municipal Science and Technology

7 of 8

,

1. B. C. Gates, Chem. Rev. 95, 511–522 (1995). 2. P. Munnik, P. E. de Jongh, K. P. de Jong, Chem. Rev. 115, 6687–6718 (2015). 3. C. Mondelli, G. Gözaydın, N. Yan, J. Pérez-Ramírez, Chem. Soc. Rev. 49, 3764–3782 (2020).

15. 16. 17.

y

RE FERENCES AND NOTES

14.

y g

We have demonstrated the reverse ripening phenomenon of Cu NPs on the dealuminated Beta zeolite support, in which the Cu NPs tend to become smaller under the methanol vapor treatment. This feature enables the realization of an ideal catalyst with superior durability for the hydrogenation of DMO that outperforms the classic Cu-silica–based catalysts. We believe that zeolites with abundant silanol nests act as powerful supports to optimize the dynamic change of Cu NPs, and the reverse ripening concept on this support should open a productive route to more-stable catalysts for industrial processes. By considering the importance of Cu-silica–based catalysts in multiple industrial reaction processes, our work provides a step forward in overcoming the disadvantage of instability and may improve the vitality of scalable applications.

12. 13.

A. T. Bell, Science 299, 1688–1691 (2003). X. Li et al., Nature 611, 284–288 (2022). E. V. Makshina et al., Chem. Soc. Rev. 43, 7917–7953 (2014). M. B. Gawande et al., Chem. Rev. 116, 3722–3811 (2016). S. Hu, W.-X. Li, Science 374, 1360–1365 (2021). J. Jones et al., Science 353, 150–154 (2016). J. C. Matsubu et al., Nat. Chem. 9, 120–127 (2017). T. W. van Deelen, C. Hernández Mejía, K. P. de Jong, Nat. Catal. 2, 955–970 (2019). Z. Li et al., Nat. Catal. 1, 349–355 (2018). M. D. Argyle, C. H. Bartholomew, Catalysts 5, 145–269 (2015). X. Jiang, X. Nie, X. Guo, C. Song, J. G. Chen, Chem. Rev. 120, 7984–8034 (2020). W. Zhou et al., Chem. Soc. Rev. 48, 3193–3228 (2019). D. Li et al., Nat. Catal. 5, 99–108 (2022). D. R. Palo, R. A. Dagle, J. D. Holladay, Chem. Rev. 107, 3992–4021 (2007). P. Wynblatt, N. A. Gjostein, Acta Metall. 24, 1165–1174 (1976). S. C. Parker, C. T. Campbell, Phys. Rev. B 75, 035430 (2007). R. van den Berg et al., ACS Catal. 5, 4439–4448 (2015). D. Yao et al., Nat. Commun. 14, 1123 (2023). S. Feng et al., Nat. Commun. 10, 5281 (2019). S. Chen et al., Chem. Soc. Rev. 50, 3315–3354 (2021). A. Bishara, K. M. Murad, A. Stanislaus, M. Ismial, S. S. Hussian, Appl. Catal. 7, 351–359 (1983). Y. Nishihata et al., Nature 418, 164–167 (2002). N. Felvey et al., J. Am. Chem. Soc. 144, 13874–13887 (2022). H. Li et al., ACS Catal. 13, 1197–1206 (2023). J. Zheng et al., Science 376, 288–292 (2022). J. Gong et al., J. Am. Chem. Soc. 134, 13922–13925 (2012). J. Lin, X. Zhao, Y. Cui, H. Zhang, D. Liao, Chem. Commun. 48, 1177–1179 (2012). C. Xu et al., Nat. Commun. 9, 3367 (2018). J. Sun et al., Sci. Adv. 4, eaau3275 (2018). H. Yue et al., AIChE J. 58, 2798–2809 (2012). J. Zheng et al., J. Phys. Chem. C Nanomater. Interfaces 119, 13758–13766 (2015). C.-S. Chen, W.-H. Cheng, S.-S. Lin, Chem. Commun. 2001, 1770–1771 (2001). Z. Gao, B. Ma, S. Chen, J. Tian, C. Zhao, Nat. Commun. 13, 3343 (2022). C. Wen, Y. Cui, W.-L. Dai, S. Xie, K. Fan, Chem. Commun. 49, 5195–5197 (2013). S. Qing et al., Int. J. Hydrogen Energy 44, 16667–16674 (2019). C. D. Chang, A. J. Silvestri, J. Catal. 47, 249–259 (1977). Z. Liu, J. Liang, Curr. Opin. Solid State Mater. Sci. 4, 80–84 (1999). S. Lin et al., J. Am. Chem. Soc. 143, 12038–12052 (2021). B. Eren et al., Science 351, 475–478 (2016). R.-P. Ye et al., ACS Catal. 10, 4465–4490 (2020). K. Ma et al., Chem. Sci. 10, 2578–2584 (2019). R. Ouyang, J.-X. Liu, W.-X. Li, J. Am. Chem. Soc. 135, 1760–1771 (2013). Q. Wan et al., Nanoscale 10, 17893–17901 (2018). T. W. Hansen, A. T. Delariva, S. R. Challa, A. K. Datye, Acc. Chem. Res. 46, 1720–1730 (2013). Y. X. Zhao, Z. Y. Li, Z. Yuan, X. N. Li, S. G. He, Angew. Chem. Int. Ed. 53, 9482–9486 (2014). Z. Yuan et al., J. Phys. Chem. C Nanomater. Interfaces 118, 14967–14976 (2014).

y

5 January 2024

Discussion

4. 5. 6. 7. 8. 9. 10. 11.

g

Liu et al., Science 383, 94–101 (2024)

silanol nests, through (CHO)Cu1* intermediates to bulky Cu particles with the formation of CO and H2, which is known as the classical Ostwald ripening route. This process is endothermic, with an energy of 0.76 eV. These findings indicate that the nucleation of Cu atoms at locations with small Cu clusters at silanol nests is preferred over ripening on larger Cu particles. The redispersion of Cu NPs on the Cu/BetadeAl thus appeared to proceed through the following steps: (i) the activation of an alcohol molecule on the surface of the Cu matrix to promote the cleavage of the Cu–Cu bond, (ii) the formation of the Cu1 intermediate, (iii) the migration of the Cu1 intermediate to silanol nests, and (iv) the anchoring of the Cu1 and then more Cu atoms to form clusters. The silanol nest capture and methanol decomposition would cooperatively regulate the (CHO)Cu1-mediated reverse ripening route. Notably, except for methanol, the other alcohols such as ethanol, n-propanol, n-butanol, and n-pentanol failed to trigger the redispersion of Cu under the equivalent tests (fig. S61). We calculated the free energy in each step for the Cu NP redispersion process for different alcohols. As shown in table S8, the biggest difference in the energy diagram for the alcohol-triggered Cu redispersion occurred at the step during which the (RCO)Cu1* intermediate was captured by the silanol nests. The results show that the migration of (RCO)Cu1* intermediates with butanol to the silanol nests is much more difficult than with methanol. The (RCO)Cu1* with a longer carbon chain would have a larger steric hindrance (fig. S62), thus prohibiting the redispersion of Cu species on Beta-deAl.

p

with adjacent silanol groups formed, which was supported by a crystal orbital Hamilton population (COHP) analysis (figs. S50 and S51). CO and H2 were preferentially produced from the reaction between *CHO and the hydrogen of the silanols (fig. S52). As a result, the Cu1 intermediates could be captured, and the silanol nest of the Beta-deAl zeolite displayed a crucial role because of its strong interaction with the Cu sites (fig. S53). This result is further supported by Fourier transform infrared spectroscopy, which showed the reduced signals that were associated with silanol nests on the Cu/Beta-deAl during the methanol treatment (fig. S54). However, the Beta zeolites with a different dealumination degree (table S6) and the different MFI zeolites, which both have a relatively low content of silanol nests, failed to realize the Cu redispersion (figs. S11, S12, and S55 to S59). Compared with the silanol nest, the terminal silanol has a much weaker interaction with the (CHO)Cu1* intermediate, as confirmed by the COHP values that represent the strength of the Cu–O bond (fig. S50). This renders the terminal silanols inoperable with respect to capturing the Cu intermediate (fig. S52), where these (CHO)Cu1* intermediates would be trapped by another bulky Cu particle, leading to Ostwald ripening (20, 43). As a result, the general Cu/SiO2 catalyst with only terminal silanols failed to have a redispersing feature but caused the rapid Cu sintering to form larger particles (29, 33, 34, 36, 43). To further confirm this hypothesis, we rationally synthesized a siliceous Beta zeolite (Beta-Si) with abundant terminal silanols rather than silanol nests (fig. S58) and used it as support for Cu NPs. Cu sintered rather than redispersed after the methanol treatment on the Beta-Si (fig. S60). After capturing the first Cu atom, the silanol nest with a Cu atom became a nucleation site, and the second and third (CHO)Cu1* species were captured under the assistance of silanol groups to form Cu3 clusters with a total freeenergy change less than −2.61 eV. The Cu clusters were efficiently stabilized by the Cud+-OSi linkage, which was in good agreement with the experimental results on abundant positively charged Cu sites after the methanol treatment (Fig. 3, A and B). The Cu3 cluster could efficiently capture more Cu1-CHO intermediates, with a total free-energy change at −1.27 eV for capturing the fourth and fifth Cu atoms, which finally led to the formation of small-size Cu NPs (Fig. 4F). Regarding the single step of Cu1 intermediate capture on the Cu4 cluster to form a Cu5 cluster at the silanol nests, the adsorption energy is −2.80 eV, whereas the Cu1 adsorption on the bulky Cu surface is −2.22 eV (table S7). These results suggest that it is more energetically favorable for Cu1 to be trapped by the small Cu clusters at the silanol nest than at the bulky Cu surface. We also considered the channel of Cu migration from the Cu5 cluster at the

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

Major Project (2018SHZDZX03); Group Research Asia of BASF Advanced Chemicals Co., Ltd.; and the Qizhen Project of Zhejiang University. Author contributions: L.Liu performed the catalyst preparation, characterization, and catalytic tests. J.L. and X.C. performed the theoretical simulations and wrote the corresponding sections of the paper. Y.Y. and W.R. provided helpful discussions on catalytic performances, catalyst structures, and comparisons with the commercial catalysts. Y.L., X.G., X.T., and L.Li participated in the XRD and XAFS characterizations. H.Li, H.Z., C.W., Q.L., H.W., K.Z., and Y.W. participated in the catalyst synthesis and

characterization. M.W., J.M., H.Lou, and Z.W. performed the mass spectra characterization and analyzed the data. L.W. and F.-S.X. designed the study, analyzed the data, and wrote the paper. Competing interests: The authors declare no competing interests. Data and materials availability: All data are available in the manuscript or the supplementary materials. License information: Copyright © 2024 the authors, some rights reserved; exclusive licensee American Association for the Advancement of Science. No claim to original US government works. https://www.science.org/ about/science-licenses-journal-article-reuse

SUPPLEMENTARY MATERIALS

science.org/doi/10.1126/science.adj1962 Materials and Methods Figs. S1 to S62 Tables S1 to S8 References (50–65) Submitted 12 June 2023; accepted 5 December 2023 Published online 21 December 2023 10.1126/science.adj1962

p g y y g y ,

Liu et al., Science 383, 94–101 (2024)

5 January 2024

8 of 8

RES EARCH

SIGNALING COMPLEXES

Molecular insights into atypical modes of b-arrestin interaction with seven transmembrane receptors Jagannath Maharana1†, Fumiya K. Sano2†, Parishmita Sarma1†, Manish K. Yadav1, Longhan Duan3, Tomasz M. Stepniewski4, Madhu Chaturvedi1, Ashutosh Ranjan1, Vinay Singh1, Sayantan Saha1, Gargi Mahajan1, Mohamed Chami5, Wataru Shihoya2, Jana Selent4, Ka Young Chung3, Ramanuj Banerjee1*, Osamu Nureki2*, Arun K. Shukla1*

1 of 8

,

5 January 2024

y

Maharana et al., Science 383, 101–108 (2024)

To visualize atypical modes of barr recruitment, we focused our efforts on M2R, which has a short C terminus, with most of the potential phosphorylation sites localized in ICL3,

y g

*Corresponding author. Email: [email protected] (R.B.); [email protected] (O.N.); [email protected] (A.K.S.) †These authors contributed equally to this work.

Structures reveal atypical barr binding modes to GPCRs

y

1 Department of Biological Sciences, Indian Institute of Technology Kanpur, Kanpur, India. 2Graduate School of Science, The University of Tokyo, Tokyo, Japan. 3School of Pharmacy, Sungkyunkwan University, Suwon, Republic of Korea. 4Research Program on Biomedical Informatics, Hospital del Mar Research Institute and Pompeu Fabra University, Barcelona, Spain. 5BioEM Lab, Biozentrum, University of Basel, Basel, Switzerland.

a P-X-P-P–type phosphorylation motif in the C terminus of a broad set of GPCRs, where P is a phosphorylation site, as a critical determinant of barr interaction and activation (17, 18). There are several GPCRs, for example the human muscarinic receptor subtype 2 (M2R), that contain a short C terminus with few potential phosphorylation sites, but they harbor phosphorylation sites primarily in their third intracellular loop (ICL3) (5, 17, 21–23). Sitedirected mutagenesis and biochemical studies have shown that phosphorylation sites in the intracellular loops of some of these receptors contribute to barr binding (24, 25). Whether these receptors engage the same binding interface with barrs and impart similar activation features as GPCRs with phosphorylation sites on their C terminus remains unexplored in terms of direct structural visualization. Several seven transmembrane receptors (7TMRs), such as the human decoy D6 receptor (D6R), are sometimes classified as nonsignaling or nonfunctional GPCRs, as they lack functional G protein coupling. However, these proteins robustly interact with and signal through barrs (26–29). The molecular mechanisms used by these receptors, known as atypical chemokine receptors (ACKRs) or arrestin-coupled receptors (ACRs), to bind and activate barrs are also mostly elusive with respect to the binding interface and activation-dependent conformational changes vis-à-vis prototypical GPCRs (30–33).

g

b

-arrestins (barrs) are multifunctional proteins that interact with and regulate a large repertoire of G protein– coupled receptors (GPCRs) at multiple levels (1–4). The interaction of GPCRs and barrs is typically conceived to be driven primarily by agonist-induced receptor phosphorylation and receptor activation, although emerging studies have started to suggest additional contributing factors such as membrane interaction, catalytic activation, and the role of specific phospholipids (2–10). Structures of GPCR-barr1 complexes have provided the first glimpse of high-resolution information about this interaction (11–16). However, considering the divergent sequences and phosphorylation patterns of GPCRs, the molecular mechanisms driving the broadly conserved nature of GPCR-barr interaction and activation have been more elusive. Recent studies have shed light on phosphorylationmediated components of GPCR-barr binding through broadly conserved phosphorylation motifs identified in a large number of GPCRs (17–20). For example, structural and biophysical studies have proposed the framework of phosphorylation codes and modulatory sites in the GPCR C terminus as a possible mechanism governing phosphorylation-mediated barr interaction (19, 20). Two independent structural studies identified that the presence of

p

b-arrestins (barrs) are multifunctional proteins involved in signaling and regulation of seven transmembrane receptors (7TMRs), and their interaction is driven primarily by agonist-induced receptor activation and phosphorylation. Here, we present seven cryo–electron microscopy structures of barrs either in the basal state, activated by the muscarinic receptor subtype 2 (M2R) through its third intracellular loop, or activated by the barr-biased decoy D6 receptor (D6R). Combined with biochemical, cellular, and biophysical experiments, these structural snapshots allow the visualization of atypical engagement of barrs with 7TMRs and also reveal a structural transition in the carboxyl terminus of barr2 from a b strand to an a helix upon activation by D6R. Our study provides previously unanticipated molecular insights into the structural and functional diversity encoded in 7TMR-barr complexes with direct implications for exploring novel therapeutic avenues.

and D6R, which is an intrinsically barr-biased receptor with no detectable G protein activation despite robust barr binding and signaling. We also determined the structure of barr2 in the basal conformation and in complex with a phosphopeptide derived from a prototypical GPCR, the complement C3a receptor (C3aR) (Fig. 1). For M2R, we used a full-length, wildtype receptor phosphorylated in cellulo through coexpression of a membrane-tethered G protein– coupled receptor kinase 2 construct (GRK2CAAX) and agonist-induced phosphorylation followed by incubation with purified barr1 and Fab30 to reconstitute the complex (fig. S1, A and B). Subsequently, we attempted to determine the structure of this complex using cryo–electron microscopy (cryo-EM). Although the receptor component was not resolved at high resolution, presumably because of inherent flexibility, we successfully determined the structure of receptor-bound barr1 at 3.1-Å resolution with focused refinement (Fig. 1C and fig. S2). Attempting to reduce the flexibility of the receptor component in this complex, we crosslinked the preformed M2R-barr1-Fab30 complex using on-column glutaraldehyde cross-linking (34) followed by cryo-EM data collection. The receptor exhibited flexible positioning relative to barr1, such that we could determine the structure of only the receptor-bound barr1 at 3.2-Å resolution (Fig. 1C and figs. S1, C and D, and S3). These structural snapshots nevertheless allowed us to identify the phosphorylated region of the ICL3 in M2R that forms the key interaction interface with barr1 and thereby allowed us to synthesize and validate the corresponding phosphopeptide (M2Rpp) (fig. S4, A and B) and determine the structure of the M2Rpp-barr2-Fab30 complex at 2.9-Å resolution (Fig. 1C and figs. S1, E and F, and S5). For D6R, we have reported previously that the critical determinants of barr recruitment are located primarily in its C terminus (27). We therefore generated a set of phosphopeptides corresponding to the phosphorylated D6R and tested their ability to activate barrs in vitro using Fab30 reactivity or limited proteolysis as readouts (fig. S4, C to F). On the basis of these assays, we identified D6Rpp2 (referred to hereafter as D6Rpp) as activating barrs most efficiently, and we used it to reconstitute D6Rpp-barr1/2-Fab30 complexes (fig. S1, G and H) and determined their structures at 3.4- and 3.2-Å resolution, respectively (Fig. 1D and figs. S6 and S7). We also determined the structures of wild-type barr2 in its basal conformation stabilized by Fab6 (Fig. 1A and figs. S1I and S8) and barr1 in complex with a C terminus phosphopeptide of the complement C3a receptor (C3aR), a prototypical GPCR (Fig. 1B and figs. S1J and S9), as references for basal and typical active conformations. EM densities of the phosphorylated receptor domains and the key loops in barrs in these

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

A

GPCR phosphorylation

B

C

Prototypical GPCRs Complement 3a receptor (C3aR)

D

GPCRs with long ICLs Muscarinic receptor M2 (M2R)

-arrestin biased 7TMRs Atypical chemokine receptor 2 (ACKR2 or D6R)

Agonist

Agonist

M2R

Phosphorylation by GRKs

?

ICL3 GRK

-arrestin recruitment

-arrestin

i

i

?

arr1

i

G-protein

iv

M2R

C3aR

arr2

Agonist D6R

Agonist M2R

C3aR

GPCR

vii

M2R

arr1

ii

iv

Fab30

arr1

ii

arr1

v

arr1

viii

arr2 Fab30

Fab30

arr2

Fab30

arr1

arr1

D6R

D6R

arr1 Fab30

ii

arr2

i

M2R

Fab30 Fab6

-arrestin

arr2

Fab30

ii

v Fab30

arr1

arr2

D6Rpp

Fab6

3.9

3.5

iii

M2RICL3 Fab30

M2R-ICL3

C3aRpp

Fab30

ix

3.4

3.2

Fab30

iii

Fab30

vi

arr1

arr2 M2Rpp

D6Rpp

arr2 D6Rpp

2 of 8

,

In contrast to prototypical GPCRs, some chemokine receptors, such as CXCR7, D6R, and a complement C5 receptor (C5aR2), lack G protein coupling but maintain robust barr recruitment and downstream signaling (27, 35–38). These receptors, referred to as ACKRs or ACRs, are essentially intrinsically barr-biased and represent an excellent model system to probe structural and functional diversity of barrs. We thus attempted to reconstitute D6R-barr complexes using coexpression of the receptor, GRK2 or GRK6, and barr1/2, followed by in cellulo assembly of the complex through

y

barr signaling complexes with atypical chemokine receptors

y g

mutagenesis. Subsequently, we measured agonistinduced barr1 recruitment to these mutants vis-à-vis the wild-type receptor using NanoBiT and coimmunoprecipitation assay. Mutation of T-V-S-T, but not T-N-T-T, nearly ablates barr binding (Fig. 2, L and M, and fig. S12). These observations establish the key contribution of the T-V-S-T motif in M2R-ICL3 in driving barr recruitment and underscore the shared mechanism of barr activation by M2R and other prototypical GPCRs despite distinct receptor domains engaging barrs.

y

C3aRpp-barr1 (iii). (C) A 3D reconstruction (left) showing a “hanging” mode of complex organization in M2R. High-resolution structures of M2R-ICL3–bound barr1/2 are shown below. 2D class average (i), overall 3D map (ii), and structure of M2R-barr1 (iii); M2R-barr1 of cross-linked complex (iv, v, and vi); and M2Rpp-barr2 (vii, viii, and ix). (D) 2D class average (i), overall dimeric 3D map (ii), and structure (iii) of D6Rpp-barr1, and D6Rpp-barr2 (iv, v, and vi). The estimated resolutions for all the structures are shown next to each map.

(Fig. 2, I and J). The barr1 and barr2 in these structures exhibit an interdomain rotation of ~18° and 23°, respectively (Fig. 2, F, H, and J); disruption of the three-element and polar-core network (figs. S11, A to D, and S24); and reorientation of the critical loops compared with the basal conformation (fig. S11E). The phosphate groups in the M2R-ICL3 stretch resolved in these structures are organized in a P-X-P-P pattern, where P is a phosphorylation site, and are engaged in ionic interactions with conserved Lys and Arg residues in barrs organized in a K-R-K–type pattern involving R7/R8, K10/ K11, K11/K12, R25/R26, K107/K108, and K294/ K295 (Fig. 2K). A comprehensive list of residueresidue contacts between the phosphopeptides and barrs is provided in data S2. The sequence analysis of M2R reveals that there are two plausible P-X-P-P–type motifs in the ICL3, one represented by T307-V-S309-T310, which is observed in the structures presented here, and the other represented by T340-NT342-T343 (Fig. 2L). To validate the contribution of the T-V-S-T stretch in M2R-ICL3 in barr engagement and activation, we generated two different mutants of the receptor with the phosphorylation sites in each of these P-X-P-P motifs changed to Ala residues by site-directed

5 January 2024

2.9

g

Maharana et al., Science 383, 101–108 (2024)

M2Rpp

arr1 M2R-ICL3

Fig. 1. A structural approach to understand the atypical modes of barr interaction with 7TMRs. (A) Cryo-EM structure of full-length barr2 sheds light on its basal-state conformation. 2D class average (i), overall 3D map of barr2 bound to Fab6 (ii), and structure of barr2 alone (iii). (B) b-arrestins adopt two distinct modes of interaction with phosphorylated typical GPCRs. The phosphorylation pattern of complement receptor C3aR was used to delineate the “hanging” mode of barr interaction. 2D class average (i), overall dimeric 3D map (ii), and structure of

above-mentioned structures are presented in fig. S10. M2R-barr1-Fab30 resembles a hanging conformation observed previously for prototypical GPCRs (11, 34) with a space between the receptor and barr components, presumably owing to their interaction mediated primarily through the long ICL3 (~150 residues) in the M2R (Fig. 2, A to E). This space is observed in M2R complexes with both isoforms of barrs and after receptor phosphorylation by either GRK2 or GRK6 (Fig. 2, A to D), suggesting that hanging conformations represent a major M2R-barr population irrespective of barr or GRK isoforms. Glutaraldehyde cross-linking appears to stabilize a more closely engaged complex, as reflected in negative-staining twodimensional (2D) class averages (Fig. 2G), but did not improve resolution of the receptor component in cryo-EM. The structure of M2Rbound barr1 revealed a phosphorylated stretch of ICL3 in the receptor that harbors the residues from E305 to G313 with four phosphorylation sites (Thr307, Ser309, Thr310, and Ser311) and docks on the N-domain of barr1 (Fig. 2, F and H). M2Rpp derived from the ICL3 sequence visualized in M2R-bound barr1 structure binds to an analogous interface on barr2

3.2

vi

arr1

arr1 Basal arr2

Fab30

iii

D6Rpp

arr1

M2RICL3

3.1

p

iii

Fab30

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

p g y y g y ,

Fig. 2. Structural insights into ICL3-driven barr interaction with M2R. (A to D) Negative-staining EM class averages of M2R, endogenously phosphorylated by GRK2/6 in complex with barr1 or barr2. (E) Cryo-EM 2D classes, 3D reconstruction of “hanging” M2R-barr1-Fab30 complex. (F) Structure of barr1 bound to phosphorylated M2R-ICL3. The EM density of ICL3 and surrounding residues within 4 Å are shown in the inset. barr1 attains an active conformation with a Maharana et al., Science 383, 101–108 (2024)

5 January 2024

C-domain rotation of 18.4° with respect to the N-domain. (G) Representative negative-staining EM 2D classes depicting the effect of cross-linking. Yellow arrows show potential transition of the complex subunits. (H) Structure of the crosslinked M2R-barr1 complex. The EM density of ICL3 and surrounding residues within 4 Å are shown in the inset. C-domain rotation value with respect to N-domain is 18.6°. (I) Sequence of phosphopeptide derived from the ICL3 of M2R. (J) Structure 3 of 8

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

of M2Rpp-barr2 in ribbon representation. M2Rpp is shown in yellow and barr2 in blue. Density map of phosphopeptide and surrounding residues within 4 Å are displayed to the right. barr2 attains an active conformation with 23.4° rotation of C-domain upon activation with M2Rpp. (K) The phosphorylated residues from ICL3 making critical contacts with Lys and Arg residues of barr1 (upper) and barr2 (lower) are highlighted in blue. (L) Cartoon representation illustrating the presence of possible phosphorylation clusters in the ICL3 of M2R. Mutations of the two phosphor-motifs: TVST and TNTT were generated to assess the barr recruitment measured by bystander NanoBiT assay (receptor+SmBiT-barr1+LgBiT-CAAX). Substitution of phosphosites of TVST to AVAA leads to abrupt reduction in barr recruitment, whereas TNTT to ANAA substitution maintains barr recruitment, suggesting a critical role played by TVST on barr recruitment to M2R (mean ± SEM; n = 3 inde-

,

4 of 8

y

5 January 2024

A cryo-EM structure of a chimeric M2R with engineered vasopressin receptor 2 C terminus (M2-V2R) with barr1 has been determined previously (11); however, the ICL3 of M2R was not resolved in the structure. It thus has not been known how M2R or other similar GPCRs with a short C terminus but relatively longer ICL3 engage barrs (42). Our structures of M2R-barr1 and M2Rpp-barr2 underscore that the key interaction interface and the activation mechanism remains conserved despite distinct domains on the receptor being used to engage barrs. Our work reveals how two barr isoforms are able to interact with and regulate a broad set of receptors with a structurally conserved interface and activation mechanism. Comparing barr structures determined thus far, including the C3aRpp-barr1 structure presented here, a higher interdomain rotation is observed in barr2 than in barr1 (fig. S17). We speculate that this observation may provide a mechanism for how class B GPCRs, which have a relatively stable barr interaction, exhibit apparently higher affinity for barr2 over barr1 (43). A comparison of our M2R-bound barr1 structure with a previously reported M2R-V2R-barr1 complex reveals the hanging conformation of barr1 with respect to the receptor (fig. S18). This observation further underlines the occurrence of the hanging conformation as a major population in the context of native M2R-barr interaction and offers a structural framework to design guided experiments to probe functional outcomes in future studies. However, the active conformations of barr1 were similar in terms of the interacting residues on N-domain, key loops, and C-domain rotation values (fig. S19). We observe that M2R-barr complexes can also form engagements that resemble fully engaged conformations of GPCR-barr complexes, as is apparent in cryo-EM and negative-staining data and more pronounced after cross-linking. Thus, it is conceivable, although not yet established, that barr-mediated M2R internalization and desensitization follows the same paradigm as other GPCRs, with the hanging conformation

y g

Previous structures of activated barrs, either in complex with phosphopeptides or full-length receptors, have used either truncated barrs or the C terminus is not resolved structurally. Even in the crystal structure of barr2 in its basal conformation, only part of the C terminus is structurally resolved (39, 40). We determined the cryo-EM structure of wildtype, full-length barr2 and resolved a longer stretch of the C terminus (Fig. 4, F and G, and fig. S16). A stretch of the barr2 C terminus adopts a b strand in its basal conformation and docks to the N-domain to maintain the barr in an inactive conformation. A previous structure of the b-appendage domain of adaptin (AP2) in complex with a peptide corresponding to the C terminus of barr1 also exhibits an a-helical conformation of the peptide that is positioned onto a groove in the platform subdomain of the b-appendage (Fig. 4E) (41). The propensity of the C terminus in barr1 and barr2

Discussion

y

Resolution of an extended barr C-terminal tail

to adopt an a-helical conformation should be explored further.

g

Maharana et al., Science 383, 101–108 (2024)

the key loop regions compared with the basal state, and disruption of the three-element and polar core network (Fig. 3, G and H, and figs. S11 and S24). A comprehensive list of residueresidue contacts between the phosphopeptides and barrs are given in data S2. We observe that the distal C terminus of barr2 (Tyr391 to Lys408) in the D6Rpp-bound conformation adopts an a-helical structure, which is positioned in the central crest of barr2 (Fig. 4, A and B) through extensive interactions (fig. S15). This a helix in barr2 forms a key dimerization interface for the two protomers in this structure and is arranged in an antiparallel coiled-coil fashion with contacts across the two protomers (Fig. 4C and data S3). We analyzed this a helix using molecular dynamics (MD) simulations and observed that it exhibits robust stability over a 2-ms simulation time frame (Fig. 4D). We also observed that this stretch of barr2 C terminus has a propensity to adopt an a-helical conformation even in isolated form, meaning, without the barr2 core being present. We did not observe this a-helical structure in D6Rpp-bound barr1, although the corresponding segment is not resolved in the structure.

p

agonist stimulation and stabilization using Fab30. Although we observed clear complex formation and a typical architecture by negative staining that is reminiscent of the hanging conformation (Fig. 3, A and B), attempts to scale up the complex for cryo-EM analysis were not successful. Therefore, we focused our efforts to determine the structures of barrs in complex with a phosphorylated peptide corresponding to the C terminus of D6R (D6Rpp) (Fig. 3D). We first confirmed that D6R-barr interaction depends on receptor phosphorylation by truncating the C terminus of D6R harboring the phosphorylation sites, which resulted in near-complete ablation of agonistinduced barr1 recruitment (Fig. 3C). Subsequently, we characterized D6Rpp using in vitro proteolysis and Fab30 reactivity assays (fig. S4, C to F) and further validated barr activation by this peptide using hydrogen/deuterium exchange mass spectrometry (HDX-MS). We observed that D6Rpp binding resulted in robust activation of barrs, as reflected by conformational changes in multiple b strands and loop regions in the N-domain (Fig. 3, E and F, and fig. S13). We also observed notable differences between the HDX-MS pattern of barr1 versus barr2, such as reduced solvent exposure of b strands XIV and XV in the C-domain of barr2, which suggests isoformspecific differences between activation of barr1 versus barr2. We determined the structures of barr1 and barr2 in complex with D6Rpp, stabilized by Fab30, at resolution of 3.4 and 3.2 Å, respectively (Figs. 1D and 3, G and H). We observed a similar interaction interface of D6Rpp on N-domains of barr1 and barr2, although seven phosphates were resolved in the barr2 structure compared to three in barr1 (Fig. 3, G and H). However, we observed that three phosphate groups, on Ser348, Ser350, and Ser351, were organized in a P-X-P-P pattern and were engaged in interactions with selected Lys and Arg residues in the N-domain of barrs (Fig. 3I). As in M2R, there are two putative P-X-P-P motifs in D6Rpp, but our structural snapshots only capture barrs associated with one of them (fig. S14). We also observed interdomain movement in D6Rpp-bound barrs, reorientation of

pendent experiments; normalized with respect to highest ligand concentration signal for M2RWT as 100%). (M) Role of TVST in barr recruitment is further corroborated by coimmunoprecipitation assay. On carbachol stimulation, M2RAVAA showed drastic reduction in barr1 recruitment. A representative blot and densitometry-based quantification are presented (mean ± SEM; n = 4 independent experiments; normalized with M2R 30-min stimulation condition signal as 100%; two-way analysis of variance, Tukey’s multiple comparisons test). The exact P values are as follows: M2RWT 0 min versus 15 min, P = 0.0006; M2RWT 0 min versus 30 min, P =