2 FEBRUARY 2024, VOL 383 ISSUE 6682 Science

Citation preview

EDI TO R I A L

Earning respect and trust

R

espect for, and trust in, science may be at an all-time low. In the United States, a 2023 Pew Research poll showed that only 57% of the population believed science has had a positive impact on society, and a Gallup poll showed that confidence in higher education was down to 36%. If the Gallup poll were done now, support would likely be even lower, given recent events with university presidents, from questions about their research integrity to their explanations for policies on student speech. I’m frequently asked what can be done about all of this, especially in the realm of science. Many scientists think the challenge has largely to do with science communication, which is certainly important. But first, the scientific community must begin to conduct itself in the same manner that it is asking of the public, and that means treating everyone in the scientific community with respect. As Editor-in-Chief of the Science family of journals, I work with a highly skilled staff in scientific publishing. The professional editors are scholars in their fields and intensely dedicated to the goal of a robust scientific record. In addition, the visuals experts produce striking and educational imagery; the social media crew runs widely subscribed accounts; gifted news reporters provide outstanding global coverage; and a sharp communications team spreads the message far and wide. Plenty of scientists recognize the importance of all of this. But far too often, I see my colleagues at the journals treated disrespectfully by authors, reviewers, and readers who consider running a research lab as somehow more meaningful than anything else in science. This is simply ludicrous. If anything, the challenges that science is experiencing now are not due to a lack of success in the laboratory. They are due to a lack of emphasis on other aspects of science—great teaching; communicating; policy-making; and performing the hard intellectual labor of choosing, from the mass of research, those discoveries that deserve publication in a top journal— and then working with authors to make the findings publishable. The notion that lab work is the only purposeful endeavor in science is obtuse and is an example of precisely what leads to the view that scientists are

intellectual elites who do not value the contributions or abilities of anyone except themselves and the small group they deign to recognize as their peers. Every time this academic hauteur is revealed to the public, confidence is lost for a simple reason—scientists like these are not inspiring the people’s trust. When I was a university administrator, I was frequently visited by graduate students who were in distress after they had informed their adviser that they did not intend to pursue an academic research career. Suddenly, their adviser became less interested in them. I was dismayed by faculty who had apparently forgotten that they worked at a school, where helping students achieve success in the life that they choose is the goal. Academic researchers should be excited for students who want to contribute to scientific publishing, education, policy, and other endeavors where science needs much more help than it does in producing more grants and papers. Nowhere is this elitism more apparent than in the behavior I sometimes see from academics toward the staff at Science’s journals. Many seem to think that having highly cited work and membership in exclusive academies gives them license to be dismissive of others. This is pure arrogance and ignorance. Professional editors are scientists who are highly capable and trained to handle papers. Too often, an editor’s decisions are attacked as thoughtless output from an underling rather than insightful determinations from a true colleague. I have worked with both excellent academic departments and with outstanding professional editors, and talking science in both environments is equally stimulating and challenging. Furthermore, the people who handle the visual, communication, and technical details do important things that no researcher can. As it turns out, these individuals are not just among the most capable people in scientific publishing, they’re among the most capable people in the enterprise of science. The way to restore trust in science and higher education is by earning it. Let’s start by recognizing everyone within the scientific community as peers in the scientific quest. –H. Holden Thorp

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

PHOTO: CAMERON DAVIDSON

“…the scientific community must…conduct itself in the same manner that it is asking of the public…”

10.1126/science.ado3040

SCIENCE science.org

2 FEBRUARY 2024 • VOL 383 ISSUE 6682

463

NEWS

“

Birth is an unpredictable process.

”

Migration researcher Lea Müller-Funk, in Times Higher Education, about criticism by her and other pregnant scientists that funders have shown inflexibility by scheduling their interview for a grant application close to their due date or recent delivery.

IN BRIEF Edited by Jeffrey Brainard

PLANETARY SCIENCE

A

fter 72 flights on Mars—67 more than originally planned—NASA’s Ingenuity helicopter flew its final sortie last month. The tiny helicopter (above), a technology demonstration deployed from the Perseverance rover, lasted for 3 years, covering 17 kilometers during more than 2 hours of flight, until at least one of its carbon-fiber blades fractured during a crash landing. Ingenuity served as a scout for

Wellcome aims to boost diversity | The Wellcome Trust last week announced a £20 million fund for Black, Bangladeshi, and Pakistani researchers based in the United Kingdom as a step toward diversifying the scientific workforce. The London-based trust, which is one of the largest philanthropic funders of science in the world and focuses on health studies, notes that researchers from those groups are underrepresented in U.K. academia: For example, Black people make up 4.4% of the working age population in England and Wales, but just 2.9% of the U.K.’s research

FUNDING

464

2 FEBRUARY 2024 • VOL 383 ISSUE 6682

Perseverance as it climbed toward the rim of Jezero crater, helping it avoid hazardous terrain and collect rock samples. The mission could pave the way for future martian helicopters, starting with the expensive, troubled campaign to collect Perseverance’s samples, rocket them off the planet, and return them to Earth. Should the rover fail, NASA is considering relying on a larger helicopter to ferry the samples to the return rocket.

community. Under the program, eligible scientists can apply for research grants of up to £200,000 for work lasting up to 2 years. Wellcome aims to initially distribute about £4.5 million annually for 4 years, in one of the largest U.K. initiatives of its kind. Last week applications also closed for a pilot funding scheme by the Royal Society that will give five Black scientists up to £690,000 over 4 years to cover salary and research expenses.

WHO OKs low-cost COVID-19 shot P U B L I C H E A LT H | The World Health Organization (WHO) last week gave

emergency use approval to a low-cost COVID-19 vaccine, years after observers questioned whether regulators in India rushed to authorize its use there without sufficient evidence. Corbevax, a vaccine based on part of the coronavirus’ spike protein, was developed at Texas Children’s Hospital and Baylor College of Medicine and is notable because its developers didn’t seek patent protection on it. In 2020 they freely licensed it to Biological E Limited, a global vaccine supplier based in India. In December 2021, the Drugs Controller General of India gave emergency use authorization for adults and later greenlighted it for science.org SCIENCE

PHOTO: NASA/JPL-CALTECH/ARIZONA STATE UNIVERSITY/MALIN SPACE SCIENCE SYSTEMS

Breakdown finally grounds Mars helicopter

adolescents and as a booster. Biological E was able to sell the two-dose vaccine to the government at the extraordinarily low price of 145 rupees ($1.90) per dose, and people in India have received more than 100 million doses. It is the 14th COVID-19 vaccine to receive an emergency use listing from WHO, whose endorsements often influence whether a low-income country considers a vaccine.

Museums shut Native displays | Several U.S. museums have closed or covered up displays featuring cultural artifacts taken from Native American archaeological sites because of new U.S. rules requiring tribes’ permission to display them. The American Museum of Natural History announced last week it closed its Eastern Woodlands and Great Plains halls, which contained such artifacts, and covered over other displays. Similar moves were taken by Chicago’s Field Museum and Harvard University’s Peabody Museum of Archaeology and Ethnology, The New York Times reported. These actions began after the Biden administration updated the Native American Graves Protection and Repatriation Act last month. It now requires museums to obtain consent from tribal groups before displaying cultural items, such as funerary objects. The changes also set a deadline of 2029 for holders of these, other artifacts, and bodily remains to prepare them for repatriation to tribes—a process that tribes say has already taken too long. The government estimates institutions nationwide hold the remains of some 96,000 Native American people, which museums typically do not display publicly. RESEARCH ETHICS

ILLUSTRATION: SYLVAINE JACQUINOT AND JACK BAKER

Probes to study black holes, Venus S PAC E S C I E N C E | The European Space Agency (ESA) last week approved constructing an orbiting gravitational wave detector and a spacecraft to study Venus. The Laser Interferometer Space Antenna will measure with subatomic precision the distances between three spacecraft flying in formation 2.5 million kilometers apart. Those data could reveal ripples in spacetime created when pairs of supermassive black holes spiral together and merge. The mission is scheduled for launch in 2035. Around then, EnVision is set to arrive at Venus and gather evidence about why that near-twin of Earth ended up so different. Using radar, optical, and gravity-field sensors, it will observe Venus’s interior and its scorching surface and thick atmosphere for clues about how they interact.

SCIENCE science.org

ARCHAEOLOGY

Ice age jewelry shows diversity

I

ce age hunters in Europe created at least nine distinct styles of jewelry, suggesting separate cultures within a group once regarded as monolithic, a study has found. The so-called Gravettian culture, which lasted from about 34,000 to 24,000 years ago, spanned what is now Portugal to Russia. A research team identified differences within this broader culture by comparing thousands of handcrafted beads and other personal adornments (right) from 112 widely spaced burial and habitation sites. They identified 134 types of beads based on their raw materials and other design elements. The team discovered that the adornments clustered geographically—people in northwest Europe often wore tube-shaped shells of Dentalium mollusks, for example. Yet based on ancient DNA from these Gravettian sites, closely related people didn’t always use the same styles of adornments, and unrelated people sometimes did—raising new questions about what the different cultural styles signified and how they spread, the study’s authors say this week in Nature Human Behaviour.

Open-access papers’ wider reach P U B L I S H I N G | Open-access articles attract more citations from scholars in a wider range of locations, institutions, and disciplines than paywalled papers, a study has found. A research team analyzed citations to 19 million scholarly works of both types, published from 2010 to 2019. Open-access papers consistently scored higher on two scales that quantified the variety of disciplines and locations among the citing authors, according to the paper, published last month in Scientometrics. In 2014, for example, the open-access papers drew citations from scholars at about 31 institutions, on average, versus 21 for paywalled papers.

BY THE NUMBERS

0

Cases of cervical cancer among nearly 30,000 Scottish women, ages 24 to 32, vaccinated as girls against human papillomavirus, which causes the cancer. (Journal of the National Cancer Institute)

The diversity of citing authors more reliably indicates a paper’s reach than citation counts, the study’s authors say. They also reported a “citation diversity advantage” for open-access articles deposited in public repositories (“green” open access) over those published as “gold” open access, whose authors pay publishers to make them immediately free to read.

Nonopioid pain drug scores well B I O M E D I C I N E | An experimental compound relieved postsurgical pain in two large trials while avoiding opioids’ risks of addiction and other side effects, its maker announced this week. VX-548, developed by Vertex Pharmaceuticals, blocks a sodium channel on pain-sensing neurons. The trials tested its effects after surgeries that removed bunions and abdominal fat. Participants who got VX-548 reported more pain relief than those receiving a placebo, but less than those given a standard combination of acetaminophen and the opioid hydrocodone. Vertex plans to seek U.S. regulatory approval for VX-548 to treat moderate to severe acute pain. It is also testing VX-548 for use in chronic pain, a complex set of conditions with fewer effective treatments.

2 FEBRUARY 2024 • VOL 383 ISSUE 6682

465

NE WS

IN DEP TH

Sailors on 19th century ships measured ocean temperatures with buckets. Correcting bias in those records would reduce uncertainty over how much Earth has warmed.

CLIMATE CHANGE

Is the world 1.3°C or 1.5°C warmer? Global warming’s uncertainty hinges on old logbooks from merchant ships By Paul Voosen

L

ast month’s announcement that 2023 was the hottest year in history was no surprise. But it came with one: No one knows exactly how much the world has warmed. One group of climate scientists found the planet has warmed 1.34°C over the 1850–1900 average, whereas another found temperatures had risen 1.54°C. In the past, such differences often came down to how groups created a global average and filled in temperatures for remote areas without weather stations, such as polar regions. But the current disagreement is not over present temperatures, but rather the past. The warmth of the ocean in the late 19th century is a key part of the baseline against which the warming of the planet is measured—and figures are at odds. The tension has spurred a flurry of new efforts to identify and correct bias in old tempera466

2 FEBRUARY 2024 • VOL 383 ISSUE 6682

ture logs recorded by sailors. “The impetus to resolve this is getting bigger,” says Robert Rohde, the lead scientist of Berkeley Earth, one of the five main groups that produce a global temperature record. The 0.2°C discrepancy between models does not call into question the humandriven warming of the past century. And mitigating climate change is just as urgent for 1.34°C of warming as for 1.54°C, says Gavin Schmidt, director of NASA’s Goddard Institute for Space Studies, which also produces a global temperature record. But a key symbol is at stake: the arbitrary 1.5°C threshold for “dangerous” climate warming, which policymakers settled on in the 2015 Paris agreement. Climate scientists may never be able to say precisely when the world has passed that milestone, Schmidt says. “There’s enough uncertainty to make it problematic.” No estimate of global temperature is possible without including the oceans, which

cover 70% of the planet’s surface. Today, researchers average data from satellites, weather stations, and buoys to estimate temperatures across the planet’s surface. But ocean temperature records in the 19th century were few and far between. A global record began in the 1850s thanks to a controversial figure, Matthew Fontaine Maury, a superintendent at the U.S. Naval Observatory who avidly supported slavery and would go on to serve the Confederacy (Science, 3 September 2021, p. 1070). Maury wanted the United States to keep up with powers like the United Kingdom’s East India Company, which had discovered optimal shipping routes by measuring powerful ocean currents. So he encouraged merchant sailors to collect weather observations, including measurements of water temperature from buckets heaved to the deck; if captains shared the data with the government, they would receive naval charts in return. science.org SCIENCE

from tree rings and corals. The practice spread to other navies and Other techniques for weeding out bias merchant marines. Over time, wooden are more painstaking. Elizabeth Kent, a buckets gave way to canvas and rubber climatologist at the U.K.’s National Oceanoones, and when steam ships took over, sailgraphy Centre, and colleagues hunt for ors began to measure water temperatures overlooked details in the logbooks that first through engine intake valves and later can help them identify unknown ships with sensors along the hull. Each method and infer their data collection methods. biased the reading: Canvas buckets, for exFor example, a zigzagging vessel—a sign of ample, exposed the water to evaporative tacking in the wind—is likely a sailing ship cooling, whereas intake valves, warmed by that sampled with buckets. There’s a lot of the ship itself, heated the water. work to be done with existing records, she Today, two organizations maintain these says. “It’s a massive mix of stuff we know historical sea surface temperature records: well and some that’s a horror show.” the U.S. National Oceanic and Atmospheric Grouping ships from different countries Administration (NOAA) and the U.K.’s Met and looking for differences in measureOffice. They both catalog the same underments when those fleets cruised the same lying data, but differ in how they approach stretch of ocean at the same time can also a key question. “How to correct the bucket reveal bias, Chan says. This led to the distemperature?” says Boyin Huang, a NOAA covery that, after the 1930s, temperature oceanographer who leads work on the hismeasurements from Japanese ships tended toric baseline and will present its sixth verto be 0.35°C colder than those from other sion at this month’s Ocean Sciences Meeting countries. This wasn’t because of any oddin New Orleans. NOAA does so by crossity in Japanese data collecting. Rather, checking the bucket temperatures with when the U.S. Air Force air temperatures taken at was digitizing these rethe same place and time, cords after World War II, whereas the Met Office putting them on punch relies on a “bucket model” cards, it dropped the decito estimate the water’s mal to save space. “They temperature before it was floored everything to the scooped up. whole degree,” Chan says. Each method has its A staggering numflaws. The Met Office, for Duo Chan, ber of logbooks have yet example, makes assumpUniversity of Southampton to be digitized, says Ed tions about what type Hawkins, a climate scientist at the Uniof bucket was used when it wasn’t docuversity of Reading. The U.K.’s National Armented. Meanwhile, air temperatures have chives has 6 million pages that are so far their own biases, which depend on the untouched, for example. “We could at least weather and time of the observations—and double the quantity of data we have availeven the height of the ship decks, which able,” Hawkins says. grew taller over time. Machine reading and other artificial inUnderscoring doubts about NOAA’s data telligence techniques could accelerate the set, it shows less ocean warming over the work. But science agencies haven’t pressed past 170 years than the Met Office record, suggesting that the land warmed far faster for this, Kent says, and the field lacks the than the ocean during this time, to a degree hands to make progress. “For the importhat climate models show to be implausible. tance of these data sets, the number of Each of the five global temperature groups people working on it is just unimaginably must pick one of the two baselines, even small,” she says. “People think it’s all done, though they are starkly different. “They it’s all fine.” can’t both be right,” Rohde says. Refining the records offers more than Duo Chan, a climate scientist at the Unijust increased certainty about the pace of versity of Southampton, says using temglobal warming today, Chan says. It could peratures from nearby island or coastal help illuminate how ocean warming varweather stations to adjust shipboard obies from basin to basin and shed light on servations can better compensate for the a puzzling observation—the eastern Pacific bucket biases. It has the additional benefit Ocean’s seeming resistance to warming, a of removing two strange trends: cooling matter of importance for those who study that began in the late 1800s, followed by the future of the El Niño climate pattern rapid warming from 1910 to World War II. (see story, p. 472). Better records would Remove these artifacts, he says, and “you also help shore up models’ projections of get a much smoother temperature evoluglobal warming, Chan says. “If we do not tion.” His proposed corrections also line up know the past,” he says, “we cannot give better with temperature records inferred much credit to the predictions we make.” j

“If we do not know the past, we cannot give much credit to the predictions we make.”

SCIENCE science.org

SCIENCE POLICY

NSF bets big on turning research into prosperity Agency funds first round of 10 “innovation engines” designed to spur regional economic growth By Jeffrey Mervis

M

olly Hemstreet, a community organizer and entrepreneur in rural western North Carolina, doesn’t fit the typical hard science profile of someone receiving millions from the National Science Foundation (NSF). But her vision for reviving North Carolina’s struggling textile industry by using more recycled materials, green manufacturing practices, and a well-trained workforce was tailor-made for a new NSF program designed to fuel economic prosperity in neglected communities across the United States. This week, NSF announced that Hemstreet’s project is one of the first 10 it is funding under an ambitious and unprecedented program called Regional Innovation Engines (RIEs). If all goes according to plan, each engine will receive up to $160 million over 10 years—by far the largest sum NSF has ever invested in an individual project, and four to five times what’s awarded to a typical NSF science and technology center. Rather than supporting research and facilities, the RIEs are meant to bring together academia, business, and community leaders to drive economic growth. NSF Director Sethuraman Panchanathan says they are part of the agency’s “commitment to create opportunity everywhere and enable innovation anywhere.” Overall, the 10 engines awarded on 29 January—selected from 188 proposals— involve lead institutions from 15 states. They include initiatives to advance regenerative medicine, enhance semiconductor and aerospace manufacturing, develop clean energy production and storage technologies, and address water supply and climate change challenges. 2 FEBRUARY 2024 • VOL 383 ISSUE 6682

467

NE WS | I N D E P T H

they are located—that have not benefited from the massive investments driving today’s digital and high-tech economy. By that metric, Hemstreet says, parts of North Carolina, which is not an EPSCoR state but has seen its textile industry decline, certainly qualify. In 2008 she started a company that has grown into the country’s largest worker-owned apparel manufacturing facility. And in 2015 she formed the nonprofit, called the Industrial Commons, that was awarded the RIE grant. “I think what appealed to NSF is that we have practical experience bringing people together, developing programs to engage and train young people, and figuring out what the next generation of this industry should look like,” she says. Even states with large research institutions have regions that are underserved, notes Anthony Atala, a pediatric urologist and biomedical engineer who directs the Wake Forest institute. “Sure, Winston-Salem is thriving,” he says. “But if you go 20 minutes from the medical school, you see communities that are like ghost towns, where their industries have died and there are no jobs.” He adds that the engines program is also trying to address a problem that plagues the entire U.S. academic system. “We’re very good at innovating in First lady Jill Biden (right) went to Forsyth Technical Community College to science, but we’re not announce that North Carolina is getting two innovation engines. very good at innovating for manufacturing,” he explains. “So unless let the public know what we’re doing here.” you design and build the right equipment, In creating the engines program, Panchanathan promised Congress that it train the workforce, and develop an effiwould help address a historical geographic cient distribution system, you could end up funding imbalance, in which just six states with a product that can’t be produced and receive 40% of the agency’s budget, now made available on a mass scale.” $10 billion. A 44-year-old program that Each engine will receive $15 million to rev steers small amounts of funding to 25 haveup in the first 2 years, with the promise of not states, called the Established Program to $145 million more over the next 8 years if it Stimulate Competitive Research (EPSCoR), meets certain milestones. NSF says its initial has done little to change that imbalance. investment is being matched by $365 million So far, however, only two EPSCoR from private sector partners. states—Louisiana and North Dakota—are NSF has $200 million in its current budhome to institutions chosen to lead an enget to launch the first cohort of engines. gine. Louisiana State University will lead But it is hoping to get enough money from an effort to transition the state’s massive Congress in future appropriations to susoil and gas industry to cleaner alternatain them and fund subsequent cohorts tives. The North Dakota engine, led by of engines over the next decade. That will North Dakota State University, focuses on depend in part on whether the engines live developing agricultural technologies. But up to a promise Panchanathan made at Panchanathan says the program is dethe Forsyth event: to create “world-leading signed to serve communities—wherever hubs of innovation.” j The political stakes are high. Panchanathan has made the program the flagship of NSF’s new directorate for technology, innovation, and partnerships, created under the CHIPS and Science Act, a 2022 law to bolster the U.S. semiconductor industry. Last week the White House sent first lady Jill Biden to North Carolina to announce that Hemstreet’s textile project would be one of two led by institutions in the state, which voted narrowly for former President Donald Trump in 2020. “Today’s investment of $30 million is going to North Carolina to create great jobs,” the first lady said at a 26 January visit to Forsyth Technical Community College, which will help conduct the training component of a second North Carolina engine led by Wake Forest University’s Institute for Regenerative Medicine. “And we hope you’ll

468

2 FEBRUARY 2024 • VOL 383 ISSUE 6682

HUMAN EVOLUTION

In Europe, an early, cold dawn for modern humans Moderns made mysterious ice age artifacts—implying overlap with Neanderthals By Andrew Curry

M

ore than 45,000 years ago, small bands of hunters chased horses, reindeer, and mammoth over a vast expanse of tundra that stretched across most of northern Europe. They rarely stayed anywhere for long, leaving behind a scattering of stone tools and traces of the odd campfire in the depths of caves. For more than a century, archaeologists debated whether these artifacts were left by some of the last Neanderthals to roam Europe—or the first modern humans to brave the northern reaches of the continent. A trio of papers this week in Nature and Nature Ecology & Evolution may help settle the question. Between 2016 and 2022, archaeologists recovered fragments of hominin bone from a cave in the central German village of Ranis. The bones were at least 45,000 years old, and their DNA has now identified them as the remains of our species. “We now have a Homo sapiens population in northern Europe long before Neanderthals disappeared,” says Marcel Weiss, an archaeologist at the Friedrich Alexander University of ErlangenNuremberg who supervised the excavations. What’s more, the bones were found with a type of stone blade known from other sites across northern Europe, from the British Isles to modern-day Poland. Archaeologists once assumed they were the handiwork of Neanderthals, but the Ranis bones hint that the tools—a style called Lincombian-Ranisian-Jerzmanowician (LRJ)—are modern humans’ calling card. “This suggests that early humans were far more widely spread, much earlier than we thought,” says University of Vienna archaeological scientist Tom Higham, who was not involved with the research. “What seems to be emerging is a complex mosaic science.org SCIENCE

PHOTO: MARCEL WEISS

In the German village of Ranis, archaeologists dug through 8 meters of rock and soil to find evidence of modern human presence in Europe 48,000 years ago.

pattern” in northern Europe, with pioneering bands of modern humans sharing the continent with Neanderthals. The Ranis bones aren’t the only evidence for H. sapiens’s early presence in Europe: In 2022, members of the same team reported finding 45,000-year-old modern human remains at a cave in Bulgaria called Bacho Kiro. A woman’s skull reported last year from Zlatý ku ˚n ˇ, a site in the Czech Republic, had wellpreserved modern human DNA and may be more than 43,000 years old. Another team has claimed still older H. sapiens finds— including a tooth from a cave in southern France that may be 54,000 years old. But signs of early incursions to the south and east did not shatter a long-held model of how modern humans conquered the continent. “A couple of decades ago, we had this scenario where modern humans came in and replaced Neanderthals in a big wave,” says archaeologist Shannon McPherron of the Max Planck Institute for Evolutionary Anthropology, a co-author of the new research. The new evidence from Ranis, added to Bacho Kiro and Zlatý ku ˚n ˇ, suggests that rather than a single wave, small groups of modern humans moved from Africa into Europe piecemeal starting about 48,000 years ago, overlapping with Neanderthals for many millennia. “That implies coexistence and competition and interaction. It’s a much more complex and diverse process,” says Carles Lalueza-Fox, an archaeologist who now directs the Museum of Natural Sciences of Barcelona and was not part of the research team. Genetic evidence has confirmed that the two groups sometimes met and interacted. DNA results from Bacho Kiro, for example, showed that people there had Neanderthal ancestors within six generations, although the Zlatý ku ˚n ˇ woman had no recent Neanderthal ancestry. Analysis of the genetic reSCIENCE science.org

sults from the Ranis individuals is ongoing, but early results hint at the mobility of these small bands, showing close connections to the skull found at Zlatý ku ˚n ˇ, more than 500 kilometers to the south. The new evidence was hard-won. To reach the LRJ layers in Ranis, the team had to break and remove boulders the size of small cars and dig a shaft through 8 meters of debris left by a long-ago cave collapse. But the cave’s cold depths helped preserve organic material for radiocarbon dating and genetic testing. “Ranis is really exceptional—it’s one of the best-preserved sites of this age I’ve ever worked on,” says archaeological scientist Helen Fewlass of the Francis Crick Institute, who led the radiocarbon dating. Boxes of broken bones collected in the 1930s and stored in nearby Halle, Germany, held more human remains from the same cave layer. And the bones and teeth of reindeer, woolly rhinoceros, cave bears, hyenas, and horses, all identified by their distinct protein signatures, provided clues to the environment the ancient hunters faced. Oxygen isotopes from horse teeth in the cave’s LRJ layers, for example, captured a hyperlocal weather report from 48,000 years ago. The average forecast? What researchers call “periArctic,” or 7°C to 15°C colder than modernday Germany. “These guys spread in a very hostile environment, like the north of Scandinavia today,” says Jean-Jacques Hublin, a paleoanthropologist at the College of France who led the Ranis research. That, too, comes as a surprise. H. sapiens originated in the tropical latitudes of Africa. Given the lack of evidence for tailored clothing prior to about 40,000 years ago, researchers long assumed early modern humans took advantage of periodic warm spells to venture north into Europe. Yet the finds at Ranis confirm that the

cold was no barrier. Within a few millennia or less, small bands of modern humans evidently transitioned from tropical or subtropical forests and grasslands in southwest Asia and the Mediterranean to northern Europe’s treeless, icy steppe. “It was really unexpected,” says co-author Sarah Pederzani, an archaeological geochemist at the University of Utah. “It’s an opportunity to think about what the draw factor could be for cold environments—maybe they’re more attractive than we thought.” Not everyone is sold, yet. University of Tübingen archaeologist Nicholas Conard argues it’s too early to attribute all LRJ sites— many of them excavated in the 20th century and poorly documented—to H. sapiens. “The paper uses the problematic idea that one can look at a stone tool and say which kind of hominin made it,” he says. “The problem is that the so-called LRJ is like a wastebasket one can put all sorts of assemblages into.” It’s possible, for example, that Neanderthals made the LRJ tools found at other sites or worked with modern humans to fashion them. “I don’t doubt their claims about Ranis,” says Dirk Leder, an archaeologist at the Lower Saxony State Office for Cultural Heritage. “The question is, was the LRJ exclusive to Homo sapiens, or was it a coproduction?” Though they apparently managed to make a go of it for millennia, ultimately the Ranis people and their contemporaries “weren’t entirely successful,” Hublin says. “They didn’t replace the Neanderthals living farther south, and at least when we try to trace the descendants of people of this time, from Bacho Kiro, it seems we have very little of their genome in later populations.” About 40,000 years ago, a new wave of modern humans arrived and proliferated on a much larger scale. It was those people who soon pushed Neanderthals to the margins, and then to extinction. j 2 FEBRUARY 2024 • VOL 383 ISSUE 6682

469

NE WS | I N D E P T H

PUBLISHING

Citation manipulation found to be rife in math Problem is so severe that an influential top-researchers list has excluded the entire field By Michele Catanzaro

C

liques of mathematicians at institutions in China, Saudi Arabia, and elsewhere have been artificially boosting their colleagues’ citation counts by churning out low-quality papers that repeatedly reference their work, according to an unpublished analysis seen by Science. As a result, their universities—some of which do not appear to have math departments—now produce a greater number of highly cited math papers each year than schools with a strong track record in the field, such as Stanford and Princeton universities. These so-called “citation cartels” appear to be trying to improve their universities’ rankings, according to experts in publication practices. “The stakes are high— movements in the rankings can cost or make universities tens of millions of dollars,” says Cameron Neylon, a professor of research communication at Curtin University. “It is inevitable that people will bend and break the rules to improve their standing.” In response to such practices, the publishing analytics company Clarivate has excluded the entire field of math from the most recent edition of its influential list of authors of highly cited papers, released in November 2023. The startling new analysis is the work of Domingo Docampo, a mathematician at the University of Vigo with a longstanding interest in university ranking systems. Over the past few years, Docampo had noticed that Clarivate’s list of highly cited researchers (HCRs) was gradually being taken over by lesser known mathematicians. “There were people that published in journals that no serious mathematician reads, whose work was cited by articles that no serious mathematicians would read, coming from institutions that nobody knows in mathematics,” he says. So he decided to delve into Clarivate’s data from the past 15 years to explore exactly which universities were publishing highly cited papers and who was citing them. The data showed that between 2008 and 2010, institutions such as the University of California, Los Angeles (UCLA) and Princeton produced the greatest number of highly cited math papers (defined as the top 1% by citation number), with 28 and 27, respectively. But in 2021 to 2023, institutions with

470

2 FEBRUARY 2024 • VOL 383 ISSUE 6682

little mathematical tradition, many based in China, Saudi Arabia, and Egypt, had displaced them. In this period, China Medical University in Taiwan topped the list with 95 highly cited math papers—compared with none a decade earlier. UCLA, meanwhile, had just a single highly cited paper. Docampo found patterns that suggested citation cartels were at work. Most telling, the citations to the top papers often came from researchers at the same institution as the cited paper’s authors. For instance, between 2021 and 2023, two prolific publishers of highly cited papers—China Medical University and King AbdulAziz University, which boasted 66 top papers in that period—each also published hundreds of studies referencing highly cited papers. The studies that

referenced highly cited papers were also regularly published in predatory journals, Docampo found, where rogue citation practices may be more easily accepted. Other scientists agree that the evidence points to widespread citation manipulation. “We have a number of researchers trying to boost their citations artificially in a manner that does not at all reflect their scientific quality,” says Helge Holden, chair of the Abel Prize committee, one of the most prestigious awards in math. “This can only be condemned.” Yueh-Sheng Chen, chief secretary of China Medical University, says his university did not engage in the practice. “We know nothing about the targeted citation and are not involved in such manipulation,” he says. The

involvement of “internationally renowned experts and scholars in fields such as applied mathematics” is part of the institution’s interdisciplinary approach to medicine, he adds. King AbdulAziz University did not reply to Science’s request for comment. Clarivate declined to comment on the issue. However, in online statements about its decision to exclude mathematicians from the most recent HCR list, the company says it was concerned by “strategies to optimize status and rewards through publication and citation manipulation, especially through targeted citation of very recently published papers.” Math is especially vulnerable to manipulation because the field is small, the company writes. “The average rate of publication and citation … is relatively low, so small increases in publication and citation tend to distort the representation and analysis of the overall field.” But citation manipulation is happening in other, larger disciplines, too, says Félix de Moya Anegón, a bibliometrician at the University of Granada—it’s just not as visible. Ilka Agricola, chair of the Committee on Electronic Information and Communication of the International Mathematical Union, worries that by singling out math, Clarivate may have conveyed the impression that the field is infiltrated by “fraudulent scientists.” “We very much regret that no other option was seen than to no longer list mathematics at all,” she says. Clarivate says it is taking “advice from external experts … to discuss our future approach to the analysis of this field.” Docampo is working on a more refined metric, which weights citations according to the quality of the citing journals and institutions. Other researchers say citation manipulation is simply a symptom of a flawed system of evaluation. Citations and similar metrics are not refined enough to monitor individual performance, says Ismael Rafols, a researcher at the Centre for Science and Technology Studies of the University of Leiden, and people are always going to find ways to game the system. Holden agrees: “The bottom line is that citations are not a good measure of scientific quality.” j Michele Catanzaro is a freelance science journalist and lecturer in journalism at the Autonomous University of Barcelona. science.org SCIENCE

The dainty spreading earthmoss is the first plant to have part of its genome synthesized.

BIOLOGY

First step taken toward artificial plant genome Scientists make partially synthetic version of moss chromosome, aiming to harness plant for industry By Mitch Leslie

PHOTO: BRITTA ROTHGÄNGER/ALBERT LUDWIG UNIVERSITY OF FREIBURG

R

esearchers have crafted synthetic genomes for several types of bacteria, and an 18-year-long project to do the same for brewer’s yeast is close to completion. Now, a group in China has tackled a multicellular organism, synthesizing part of the genome of a type of moss. The achievement, reported last week in Nature Plants, could smooth the way for creating artificial genomes for other multicellular organisms—and for turning the moss into a factory for medicines and other products. The Chinese team only reworked part of one chromosome in its chosen species, the spreading earthmoss (Physcomitrium patens). But the work is “a necessary step” toward a fully artificial plant genome, says Ian Ehrenreich, a synthetic genomicist at the University of Southern California. It is also “a wake-up call to people who think that synthetic genomes are only for microbes,” adds synthetic biologist Tom Ellis of Imperial College London. By redesigning an organism’s genome, researchers can probe questions such as what sequences are essential and how its organiSCIENCE science.org

zation affects gene function. They can also bestow new capabilities on an organism, potentially boosting its value for agriculture, industry, medicine, or other uses. Instead of fashioning chromosomes from scratch, synthetic biologists have started with natural ones and made a variety of revisions and simplifications. The yeast project, for instance, has pared away unneeded DNA, added a new chromosome to house genes necessary for protein synthesis, and introduced other changes. Although yeasts bunch up in certain situations, they generally go through life as single cells. So even though they are eukaryotes (organisms with cell nuclei), their genomes can only reveal so much about more complex, true multicellular eukaryotes such as plants and animals. Synthetic biologist Junbiao Dai of the Agricultural Genomes Institute at Shenzhen, developmental biologist Yuling Jiao of Peking University, and colleagues wanted to go further. But instead of working with Arabidopsis thaliana, the popular model plant that researchers have studied for decades, they picked a moss. For one thing, unlike an Arabidopsis cell, a single moss cell can grow into an entire plant. As a result, the team

has to engineer just one cell to transform a whole organism. Furthermore, a key DNAswapping process that allows synthetic segments to integrate into chromosomes occurs more frequently in the moss than in Arabidopsis. For simplicity’s sake, the Chinese project, dubbed SynMoss, started with part of the short arm of chromosome 18, the smallest such limb in the plant’s 26-chromosome genome. The researchers then set about trimming and tidying the DNA. They eliminated transposons, mobile DNA elements found in eukaryotes; added short labels to mark the altered arm; standardized the three-letter genetic codes that halt protein synthesis; and made other tweaks. In total, they shrank that section of the chromosome by 56%. Then, the scientists introduced the now partly synthetic structure into individual moss cells and nurtured their growth. The resulting plants appeared normal. They were the same size and shape as unaltered moss, grew reproductive structures, and produced spores. The modified plants were just as resistant to high salt levels and other stresses as their natural counterparts. However, the team found that some genes in the synthetic region were more active than normal, a change that could be damaging. The results support the contentious view that transposons aren’t essential for multicellular eukaryotes. The finding won’t end the controversy over whether these sequences are beneficial or harmful, says plant synthetic biologist Jennifer Nemhauser of the University of Washington. But for scientists who argue they are advantageous, “the challenge has been set,” she says. Researchers have already harnessed spreading earthmoss to produce some chemicals—a drug synthesized by genetically engineered versions of the plant is in clinical trials, for instance. The new SynMoss project should boost these efforts, says plant biologist Ralf Reski of the Albert Ludwig University of Freiburg, who has been studying the moss since 1985. Although the plant has already been genetically modified, “this paper takes it to the next level” by paving the way for even larger—and more useful—changes, he says. Dai, Jiao, and colleagues will soon tackle the rest of chromosome 18’s short arm and then an entire synthetic moss genome. “We aim to complete this within the next 10 years,” Dai says. Even with the help they hope to get from other labs, that’s an ambitious goal. The plant’s genome is about 40 times bigger than yeast’s. j 2 FEBRUARY 2024 • VOL 383 ISSUE 6682

47 1

NE WS

FEATURES

THE CORAL CHRONICLES On a remote Pacific island, clues to El Niño’s future are preserved in ancient reefs

T

he 500 residents of Tanovusvus do not typically worry about water. The village hugs the coast of Espiritu Santo, the largest island here in Vanuatu, a tropical archipelago in the western South Pacific Ocean. The lush foliage that surrounds it testifies to the several meters of rain it receives annually. The trucksize plastic tank that holds drinking water for the village has rarely run dry—especially during the past 3 years, when an unusually long-lived La Niña climate event brought warmth and wetness to the western Pacific. But one day in October 2023, Abel Kalo, an outreach coordinator from Vanuatu’s Meteorology & Geo-Hazards Department, arrived in Tanovusvus with a warning: The rains would be drying up. El Niño, La Niña’s 472

2 FEBRUARY 2024 • VOL 383 ISSUE 6682

By Paul Voosen, in Vanuatu opposite number, was here. Months earlier, its signature pattern had emerged: The warm ocean waters that had fueled storms over Vanuatu began to slosh east toward South America. Before inspecting the water tank, which was cracked and leaking and less than one-quarter full, Kalo told the villagers, “You have to be careful.” Speaking in Bislama, Vanuatu’s English-based creole, he said, “These next few months you can’t take this water for granted.” The El Niño event, now at its peak, is driving weather extremes not just in Vanuatu, but all over the planet. Drought has struck Australia, as well as the Amazon, where intolerably hot waters have suf-

focated endangered pink dolphins. Rains have drenched Peru, spreading dengue, while warm waters intruding near its coast have disrupted the world’s largest anchovy fishery and forced the nation to cancel a lucrative fishing season. Those same warm waters accelerated Hurricane Otis, which devastated Acapulco and Mexico’s Pacific coast in October 2023. The effects have been truly global: By suppressing the Pacific’s ability to absorb heat from the atmosphere, El Niño helped make 2023 the hottest year in history by a huge margin. “Every time there’s an El Niño event, all kinds of weird funky things happen,” says Pedro DiNezio, a paleoclimate modeler at the University of Colorado Boulder. Researchers can now predict El Niño and La Niña events months in advance, science.org SCIENCE

N E WS

A wave-eroded coral outcrop in Vanuatu is part of a once-submerged reef exhumed by tectonic forces.

Tanovusvus’s terraced coral shoreline was special, Kalo and Partin explained. Buried just beneath the water line were ancient corals, preserved from a time in Earth’s history when frigid temperatures may have caused large changes in the ocean’s behavior—“big signal land,” as Partin calls it. If found intact and drilled out, these corals could capture what El Niño was like during this cold extreme. And that, in turn, could reveal how it will behave in the future, under an opposite climatic extreme. “The way forward is to bring paleo data to bear on this,” says Gabriel Vecchi, a climate scientist at Princeton University. TWENTY THOUSAND YEARS AGO, at the peak of

mitigating some of these harms. But the future of this global pacemaker is poorly understood. Will a warming world cause more frequent or severe events? Will it suppress them? Climate models have disagreed drastically, and the historical record is too limited to show a change. This uncertainty was why Kalo arrived in Tanovusvus not just to deliver a warning, but to pursue an opportunity. He was joined by Judson Partin, a paleoclimatologist at the University of Texas (UT) at Austin who would be staying there for the next 2 months with a small crew of researchers. The duo met with several dozen locals near the village’s central mango tree, answering questions with a bullhorn. When Kalo said they were studying El Niño, many villagers nodded—in Vanuatu, knowing the state of the Pacific, with its impact on farming and fishing, is as important as knowing the season. SCIENCE science.org

the last ice age, Earth was 6°C cooler than today. Glaciers buried the northern continents. But in the tropical Pacific, things wouldn’t have looked so different, with one exception: Every island would have been much, much taller. With the planet’s water locked up in ice sheets, sea levels were 120 meters lower. Corals continued to grow around these towering islands, capturing the swings of El Niño and La Niña events as chemical signals within their skeletons much like tree rings. Later, when the ice sheets melted, the reefs at Vanuatu and elsewhere were submerged, putting their ancient records of El Niño and La Niña out of reach. Mostly. In the 1980s, Frederick Taylor, a UT geologist, discovered that Espiritu Santo was a geological marvel. Like the rest of Vanuatu’s islands, it was formed in spurts of volcanism some 10 million years ago, triggered by the Australian tectonic plate slipping beneath the New Hebrides Plate at an ocean trench. But in just the past several thousand years, a seafloor mountain has plunged into the trench. Like a shoe slipped under a carpet, it has shoved Espiritu Santo upward, with parts of the island rising some 7 millimeters a year—outpacing sea-level rise. As the land rose, it exhumed the submerged corals. Today many of them sit just meters below the surface, safe from the rain that would alter their chemistry but within easy reach. Partin’s arrival at UT in 2008 helped galvanize Taylor to return to these buried corals. The pair bonded over dreaming up the kind of drill that could do the job. Delivering a large, traditional rig to Espiritu Santo by helicopter would be too expensive, and it wouldn’t be any easier by sea: Landing craft would struggle with the rough surf and jagged reef. They needed a new kind of drill, lightweight and modular, that could be carried to the reef and assembled by hand. A portable rig would also be easier to re-

locate as they searched for the best targets. The oldest corals probably lay no more than 15 meters down, but there was really no telling where along the shoreline the team would find corals intact after thousands of years of pulverizing wave action. “People try to bro up, asking, ‘How deep you drilling?’” says Partin, a straight-talking Southerner. “I tell them we’re not trying to drill deep. We’re trying to drill a lot.” In 2019, they brought a prototype drill to Tanovusvus. In their second hole, they hit pay dirt: an intact 1.5-meter-long section of Porites coral, the best species for recording ancient climate. It represented a 160-yearlong record from 16,000 years ago, just as the ice sheets began to melt. “That core alone can justify all the effort,” Taylor says. They found another massive coral in a cliff just above the water, fresh enough not to have been altered by weather. It dated back 12,000 years, to the last gasps of the ice age. Next, the researchers had to tease out El Niño’s signal from each annual layer of the coral. Corals typically build their skeletons from calcium carbonate precipitated out of the surrounding water. But during El Niño events, when it’s colder on Vanuatu, larger strontium atoms in the water can slip into the crystal because there’s less energy to make its molecules vibrate. Besides measuring the strontium signal, the researchers compared the carbonate ions’ levels of heavy oxygen, an isotope with two extra neutrons, with their normal “light” oxygen. Rainfall is enriched in light oxygen because it evaporates more readily. So during Vanuatu’s dry El Niño events, when corals are exposed to less rainwater, heavy oxygen levels in the skeletons rise. Together, these indicators showed that during frigid periods El Niño and La Niña events were only slightly weaker and less frequent than the present, Partin says. That might seem like good news, as it would imply that in a future, warmer world, the climate swings might be only a little stronger and more frequent. But there was a wrinkle: By coincidence, both coral records came from times when a conveyor belt of currents in the Atlantic Ocean had weakened, a condition that is theorized to boost El Niño. Without the conveyor belt slowdown, El Niño might have been much weaker than today during those cold periods, suggesting in turn that it would grow stronger with warming. To be sure, Partin says, the team needs corals older than 16,000 years, when the Atlantic conveyor belt was strong. They had to go back to Tanovusvus. ON AN EARLY MORNING in October 2023, Partin and Robert Domeyko, a UT graduate student, were sitting inside a cinder-block 2 FEBRUARY 2024 • VOL 383 ISSUE 6682

473

NE WS | F E AT U R E S

The Pacific’s past rises from the sea Ancient corals can capture the climatic effects of El Niño and La Niña events that occurred thousands of years ago. But most reefs from the last ice age are submerged by more than 100 meters of water. On the island of Espiritu Santo in Vanuatu, however, tectonic forces have lifted ancient corals back to the surface. Espiritu Santo

Ice melting ends (7000 years ago)

Ice age peaks (20,000 years ago)

Drill rig Present day

Seas rise Tectonic uplift brings corals to surface

Growth during El Niño and La Niña

Coral reef

Sedimentation and layers of new coral

Weather patterns felt worldwide—and written in stone El Niño and La Niña events are sparked by fluctuations in trade winds that intensify through feedback loops. These events cause drought and downpours along the Pacific Rim and elsewhere, and they can change global temperatures by 0.1°C. Corals can preserve signals from these events in their calcium carbonate skeletons, which they precipitate from seawater.

La Niña During La Niña in Vanuatu, warm waters and increased rainfall mean less strontium and less heavy oxygen are incorporated into coral skeletons.

Walker circulation

1

Strong trade winds confine warm water to west Pacific.

2

Winds stir up cold water in the east, increasing air temperature differences and strengthening winds.

3

Warmth and moisture fuel storms in Asia and Oceania.

3

Calcium Carbon Equator Heavy oxygen

1

Vanuatu Warmer

2

Normal oxygen

Coral polyp

Colder

El Niño During El Niño in Vanuatu, cooler waters and dry conditions mean corals incorporate more strontium and more heavy oxygen into their skeletal structure.

3

La Niña

Strontium

El Niño

Equator

1

Heavy oxygen Periods of skeletal growth

2

Warmer

Colder

Porites lutea

Diploastrea heliopora

1

Weak trade winds allow warm water to flow east.

2

Warm waters decrease air temperature difference, weakening winds.

3

Storms shift toward the Americas.

U

A

T

U

CREDITS: (MAP) M. HERSHER/SCIENCE; (DATA) FREDERICK TAYLOR; (GRAPHIC, OPPOSITE PAGE) N. BURGESSS AND M. HERSHER/SCIENCE

N

house vacated for their stay by Petulvo Tasi, went flying as everyone ducked. It’s the Domeyko sat outside the house to finish the brother-in-law of Kalmanmele, the viltrade-off of using aluminum rather than sweating before hosing off in a makeshift lage chief. Life in Tanovusvus, which is off steel, Partin said. “It’s not crazy heavy, but shower. But the joy was short-lived. Tasi the electrical grid, begins at sunrise; roostthen you lose strength.” As long as the crew soon walked by with a warning of his own: ers and dogs start up even earlier. Partin had was careful with the cable, they could con“Have you heard about the cyclone?” A been there for 2 weeks, his stomach adjusttinue, he decided. The work resumed—but storm shaped by El Niño was about to ining to simple meals of root vegetables and this time with hard hats. terrupt the search for El Niño’s past. rice and nightly communal doses of kava, a Finally, late in the day, Partin noticed root drink that is a sedative and, he says, a that the water coming up from the boreFOR MUCH OF HUMAN HISTORY, the equatodigestive aid. “It’s just a magical thing,” he hole was not the usual gray mix of sedirial Pacific was thought to be a dull place. said. “You’re mixing all your biomes, and it ments and pulverized corals. It had turned Sailors long warned of the doldrums, the helps you get in line sooner.” white, a sign of an intact coral. “Maybe it’s belt where northeasterly and southeasterly The day’s drilling crew of locals arrived just me wishing,” he said. winds collide and rise to create still, humid at the house for the trek to conditions at the surface. But the shoreline. Partin valued centuries ago in Peru, fishtheir mechanical insights, ermen noticed that someUplifting history honed by countless hours jurytimes, around Christmas A tropical archipelago of 83 islands, Vanuatu was formed by spurts rigging truck engines to keep and the arrival of “the boy of volcanism triggered by the Australian tectonic plate sliding beneath the New Hebrides Plate. A mountain on the sea floor has them running in the salty air. child,” their coastal waslipped into the subduction zone and pushed up Espiritu Santo, “They see everything,” he says. ters would grow unusually the country’s largest island, faster than rising sea levels. Every few days, the crew rowarm, disrupting the antated, allowing Kalmanmele chovy harvest. to spread the daily pay—2000 Then, in 1957, a strong El PACIFIC PLATE vatu, or $17—among the vilNiño struck the Pacific during lage’s four main families. The the International GeophysiEspiritu chief joined for the day, along cal Year, when researchers Santo AUSTRALIAN with Kariesen, an older villager fanned out worldwide to study PLATE who served as the crew chief. the planet. Shipboard data “New crew today,” Partin said. showed the warm waters off NEW HEBRIDES PLATE “There will be a learning curve.” Peru extended west past HaTanovusvus The hike to the drill site waii, and that, over the centook them past an old coconut tral Pacific, the trade winds V Subduction zone plantation, over several barbed that normally blow west A wire fences, and, in a final along the equator had weakscramble, down a steep coral ened. In several papers in the slope. There, glinting against mid-1960s, Jacob Bjerknes, a New Hebrides the azure sea and sky, was the meteorologist at the UniverTrench 6-meter-tall aluminum rig. sity of California, Los Angeles, Hissing hydraulic hoses powproposed that the two anomaPacific ered by a diesel engine snaked lies were entwined, and that Ocean Coral off to mechanical control panthey regularly interacted to Sea els and motors; sea spray would influence not just the Pacific, 0 200 corrode anything electric. This but the whole world, through morning, Partin discovered “teleconnections” in the atkm that the previous crew had mosphere. “That was the big forgotten to grease a clamp. bang of El Niño history,” says Overnight it had grown a patina of rust. As the crew unscrewed piping to reveal Michael McPhaden, an oceanographer at The reef terrace was beautiful but treachthe cored rock inside, Kariesen gave a the National Oceanic and Atmospheric erous terrain. The razor-sharp corals chew whistle of excitement, yelling in Bislama, Administration. through shoes and provide no even surface “Nambawan! Nambawan!” (“Number one! At their core, El Niño events occur when for resting; within a day, leg muscles turn to Awesome!”). Partin scrambled over to look. the Pacific swings out of balance. Durfire. One wrong step into one of the meter“Yeah? I told you it was white. I told you it ing typical years, the trade winds stir up deep voids in the rock could mean a trip to was white!” deep, cold water and nutrients and bring a doctor, hours away. But the crew was at They had hit dead center on a coral, exthem to the surface off Peru’s coast—the home, jumping between rocks in flip flops tracting a core nearly 1 meter long. It wasn’t reason anchovies normally flourish there as they learned their roles: operating the a Porites, but it was a Diploastrea, another and also why penguins find a happy home winch, running a pump to keep the drill luclimate recorder. “That’s probably 40 or in the Galápagos. The winds confine warm bricated with water, and screwing and un50 years,” Partin said. And the depth sugwaters to the west. But when these winds screwing sections of steel pipe. gested it came from more than 16,000 years falter, the warm waters can roll eastward. Drilling presented hazards, too. Later ago. “So now I have to learn how to sample As the temperature difference between west that day, as a core was being winched out this,” Domeyko said. The team boxed it up and east shrinks, the trade winds weaken of the fist-size hole, a cable jumped out of to carry back to the house. further, allowing more warmth to spread its groove. A sharp crack rang out, and the The good fortune propelled everyone east, and, along with it, bands of moisture twisted metal of a shattered winch guide back up the hill that afternoon. Partin and and storms. Researchers later realized this SCIENCE science.org

2 FEBRUARY 2024 • VOL 383 ISSUE 6682

475

Judson Partin (third from left) works with Tanovusvus’s residents to operate a lightweight drill, built for the rugged reef environment.

Bjerknes feedback, as it’s called, could run in reverse: Trade winds stir up the eastern Pacific cold water, temperature differences between east and west grow and strengthen the winds, and warmth and storms are pushed farther west. They named that pattern La Niña. With their additions and subtractions to global temperatures, El Niño and La Niña events complicated early efforts to recognize human-driven global warming. That climate signal has long since emerged. But its influence on El Niño remains mysterious. Some researchers see hints of stronger “super” El Niños and multiyear La Niñas in the past 80 years. But most believe temperature records are too short and noisy to discern any trend, says Amy Clement, a climate scientist at the University of Miami. “The null hypothesis is very hard to reject— that it’s all random variability and [El Niño] comes and goes.” Climate models haven’t offered much help either. Rendering El Niño is a challenge, requiring models to correctly capture the mysterious dynamics of cloud formation, ocean eddies, and the enormous underwater waves that ferry Pacific water back and forth. “The simulation of El Niño in models is still not great,” DiNezio says. Some models find that super El Niños and longer La Niñas grow more common in a warmer climate. Many others do not. One reason for this confusion is that the atmosphere and the ocean are in a tug-ofwar for influence. Many models suggest climate change will cause air over the east476

2 FEBRUARY 2024 • VOL 383 ISSUE 6682

ern Pacific to warm faster than in the west, weakening the large-scale air circulation that is responsible for the trade winds. With weaker winds, strong El Niño events should occur more often. But Clement and others have proposed a countervailing effect from the frigid water in the eastern Pacific. As global temperatures rise, that water should warm more slowly than the western Pacific, heightening the temperature contrast along the equator. That should put the ocean into a lasting La Niña–like state that would suppress El Niño events. Clement concedes that climate simulations don’t show the trend. “It’s fair to say models are missing this.” Yet current observations seem to support the idea. Temperatures in the eastern Pacific have fallen relative to the west since 1980. For years, researchers chalked it up to natural variability. But it’s gone on long enough that that option seems less viable, DiNezio says. “Every year that passes this explanation is becoming more shaky.” Wenju Cai, a climate scientist at Australia’s Commonwealth Scientific and Industrial Research Organisation cites another ocean factor, this one favoring stronger El Niños. Right now, he says, the slowly warming surface waters in the eastern Pacific are still heating up faster than its deep waters, intensifying the temperature stratification that separates surface waters from the cold depths. That leaves less surface water for the trade winds to push around and amplifies their ability to generate strong El Niño and La Niña events. “That’s what fundamentally

drives increased variability,” Cai says. McPhaden is optimistic that the models will eventually reveal the factors that matter most among the welter of proposed mechanisms. In fact, he says, many of the models that render El Niño and La Niña events well are converging. They suggest both extremes will grow stronger and more frequent in the warming world, he says. “It’s fuzzy, but the stars are aligning.” Clement isn’t so sure. Even amid all the proposed mechanisms, she worries that researchers are still missing something fundamental. “When you have to think too hard about something,” she says, “that’s a bad sign.” Partin thinks his team’s corals could offer some ground truth. FOR NOW, the tropical cyclone that would

eventually be named Lola has put his quest on hold. There was no cell service in Tanovusvus, so to check forecasts, Partin and Domeyko had to hike to the shoreline, where they could hold their phones up and receive a signal from a nearby island. Weather models showed Lola forming north of Vanuatu and curling southwest, directly hitting the island in a little over 3 days. Although El Niño normally suppresses cyclones in the region, it strengthens winds that push storm formation northeast toward the equator, resulting in Lola’s unusual starting point. At the drill site, Partin had the luxury of making video calls back home to his wife and young son during lunch breaks. “I went ahead and told the family, ‘Hey this is comscience.org SCIENCE

N E WS | F E AT U R E S

ing. We know about it. We’re prepared, if you hear it in the news.’” It seemed certain his stay in Tanovusvus would be extended. “The world is conspiring to have me miss Thanksgiving,” he said. In the meantime, the crew could try for another core. But today only debris came up out of the hole, and at 10 meters depth, the drill stalled. “Drilling deep holes is such a pain,” Partin said. It was late, so he decided to move to a new spot on the reef, only a few meters away. Doing so required several hours of disassembly. The hike home was less ebullient than the day before. Tomorrow was Sunday, a “spell” day, leaving little time before the storm was due—perhaps only enough to move the equipment to higher ground. Although the Diploastrea core they pulled up was promising, it wasn’t quite the gold mine they were hoping for. “The treasure hunting aspect of this can get tiring,” he said. “You can go weeks and come up with nothing.” The next day, Taylor arrived, bringing steak and wine to help the team ride out the storm. By the day after, Lola had become a category 5 storm, and Kalo, back in Vanuatu’s capital, sent a warning that it looked like it could be a direct hit. Spelled by Taylor and another assistant, Domeyko escaped to the airport in the bed of a beat-up pickup truck along roads that might soon be washed out. Rain fell in the predawn light as the truck passed uniformed children walking to school. Soon the villagers would begin to prep for the cyclone, covering and bracing their homes with large palm fronds. They should be OK, Domeyko said. They’ve survived many similar storms. But the nature of the team’s work might change, he said. “I just hope it doesn’t become a humanitarian mission.”

PHOTO: P. VOOSEN/SCIENCE

ALTHOUGH IT IS a remarkable portal, Tano-

vusvus is not the only way to Big Signal Land. As it happened, while Lola was bearing down on Vanuatu, an 80-meter-long research ship was going after coral from the last ice age near Hawaii’s Big Island. Unlike Vanuatu, Hawaii is slowly sinking under the weight of rock freshly erupted from its volcanoes. Although that puts the reefs deep underwater, the slow subsidence means the corals are preserved in stairlike terraces that are easy to target by age, says Christina Ravelo, a paleoclimatologist at the University of California, Santa Cruz and co-lead of the cruise. She and her colleagues aboard the MMA Valour were weeks into drilling fossil reefs on the island’s west flank. Using a robotic drill, the team recovered more than 400 meters of coral from 35 holes. El Niño and La Niña don’t change weather conditions in Hawaii as much as they do in Vanuatu, which will make it harder SCIENCE science.org

to tease out a signal. But what the corals lack in signal strength they make up in record length. In those 400 meters of coral, Ravelo believes, the team has captured records stretching back as far as 500,000 years ago—covering not just the peak of the last ice age, but three earlier ones as well. If Partin and his colleagues succeed in their mission, the two records from disparate places could hold revelations, she says. “It will be really interesting to see what they get.” Half a world away, in Antarctica, other El Niño events could be preserved not in coral, but in ice. Last year, a team led by Christo

A drilled core of a Diploastrea coral that likely lived more than 16,000 years ago.

Buizert, a glaciologist at Oregon State University, showed in the journal Climate of the Past that variations in noble gases contained within microscopic bubbles in ice cores can capture an increase or decrease in storminess in a region. Before layers of snow get compacted into ice, heavier elements, such as krypton, diffuse down through air pockets, eventually reaching an equilibrium. But when a low-pressure storm moves in, those elements get shaken up, like a snow globe. “It’s almost like the whole snowpack is breathing,” Buizert says. After such shaking, krypton takes longer to fall back into balance than other gases. By comparing levels of krypton and another gas in the ice’s bubbles, researchers can tell you how vigorous—or stormy—that shaking was. Those Antarctic storms may seem a local phenomenon, but the weather in West Antarctica, especially, is closely tied to changes in the Pacific through shifts in the westerly winds. “It really ties Antarctica to the global circulation,” Buizert says. He plans to mea-

sure several more ice cores, to better tease out El Niño’s stormy signal. If he and his colleagues can do that, they could uncover a record going back hundreds of thousands of years. AS LOLA APPROACHED, Partin made periodic

trips out of the village to check the forecast. On the last trip, seven other villagers came with him. “Everyone was worried,” he says. This time they got good news: The weather models showed the cyclone just missing Espiritu Santo. A tense wait began. The chief ordered the villagers to stay in their homes. Partin piled his belongings into a corner and covered them with a tarp. He and Taylor kept the doors open, for air flow, waiting for the crescendo of the storm. But it never arrived. Lola devastated several other islands, but Tanovusvus only had to contend with heavy rain and gusty winds. The next day, they were back drilling. Holes three and four went quickly. The crew was clearing 7 or 8 meters a day. As they drilled the fourth hole, the crew insisted on moving the engines up for fear of the tide, but the researchers wanted to keep drilling. Soon the tide rushed up between the coral boulders, inundating the drill legs. “Holy cow,” Partin said. “Trust the locals.” On a final two holes, Partin decided to push the team, and they ended up drilling down more than 15 meters, hitting several intact corals. “That was completely unexpected,” Partin says. “We were hitting corals way deeper than we thought.” The team needs dating to confirm it, but Partin believes these corals come from the depths of the ice age. “This is the sweet spot,” he says. They were gleaming, white as can be—a healthy reef environment. But none was quite long enough for a good centuryspanning climate record. “It’s so close,” he says. “As good as it can be without it being amazing.” Burnt out, his feet bruised, Partin decided to keep his original flight home and come back in May. He wouldn’t miss Thanksgiving after all. Taylor, meanwhile, had decided this would be his last visit to the reef: It was too dangerous, and he was too unsteady on his feet. But after decades of exploration, they now knew the exact stretch of reef that held El Niño’s past. The treasure had been found. It just needed to be dug up. Tanovusvus hosted a goodbye party. Buckets of kava flowed and keyboards played reggae music into the night. The next morning, everyone loaded the boxed up drill rig into pickup trucks that would take it to a storage site in Santo Espiritu’s capital. Saying goodbye, Partin promised the chief that, after the cyclone season, he’d be back. j 2 FEBRUARY 2024 • VOL 383 ISSUE 6682

477

INSIGHTS

PERSPECTIVES ECOLOGY

When function, not origin, matters Native and introduced megaherbivores similarly affect plant diversity and abundance By Yvonne M. Buckley1 and Andrew Torsney2

B

oth extinctions and introductions of large mammalian herbivores (>45 kg) affect biodiversity and ecosystem function. An important question is whether the origin of herbivorous megafauna predicts impacts on an ecosystem that are specific to introduced species in general, more severe than those of native species, or both. On page 531 of this issue, Lundgren et al. (1) report that functional traits, rather than the origin (introduced or native), of large herbivores are correlated with native plant diversity. Together with pre-

478

2 FEBRUARY 2024 • VOL 383 ISSUE 6682

vious contradictory findings on the ecological effects of introduced and native herbivores (2), the study raises the question of whether herbivore size and dietary specialism are general determinants of invasion impact, origin effects, or both. Large mammalian herbivores that range in size from red kangaroos (Osphranter rufus) and goats (Capra hircus) to African bush elephants (Loxodonta africana) have profound effects on ecosystems through their functions as consumers and modifiers of plant biomass (3), dispersers of seeds (4), and, when dead, concentrated resources for scavengers and decomposers (5). The largest adult mega-

herbivores (>1000 kg; e.g., Rhinoceros spp., Hippopotamus amphibius, and elephants) are relatively resistant to predation pressure, which has important consequences for the flow of energy through ecosystems and landscape heterogeneity (6). However, large animals are also vulnerable to local and global extinctions (7). This is attributed to a mixture of intrinsic traits, such as low reproductive output, and extrinsic exposure to human impacts. Although large mammalian herbivores have been lost in many habitats, some ecosystems have gained them through humanmediated introductions to areas where they did not previously occur. science.org SCIENCE

Plant biodiversity is related to the abundance of large herbivores, such as the white rhinoceros, whose muzzle width allows for a less selective diet.

Feral goats, buffalo, and horses (wild populations of formerly domesticated species) are high-impact invasive species, particularly in regions where humans have caused the decline and extinction of other large herbivores or where large herbivores with traits similar to the introduced species never occurred. Indeed, goat (C. hircus), fallow deer (Dama dama), and wild boar (Sus scrofa) are included in the list of 100 of the world’s worst invasive species (8). Goats that were introduced to islands have proven to be particularly problematic as a driver of local species extinctions (9), and their eradication from islands is a conservation success story (10). Lundgren et al. examined the datasets of 221 studies (spanning six continents, although biased to Europe, North America, and Australia) to determine whether herbivore nativeness or other factors accounted for the effect of large herbivores on plant abundance and diversity. The authors found no evidence that herbivore origin correlates with effects. This was true irrespective of several conditions. Coevolutionary history, such as an herbivore sharing a native range with plant species, was not a factor. Neither was remote island status, where plants may not have coevolved with large herbivores. Phylogenetic or functional novelty of large herbivores relative to closely related or functionally similar prehistoric native megafauna with which the plants had coevolved also had no effect. The finding of Lundgren et al. is surprising given that it is in contrast with a previous metanalysis of vertebrate herbivores of all sizes, which found opposing effects of native and introduced herbivores on plant diversity and abundance, albeit from a narrower range of species and regions (mostly from North America) (2). Lundgren et al. did, however, find a relationship between the diversity of plant species and both body size and muzzle width of large mammalian herbivores. Muzzle width is an indicator of dietary selectivity. Larger muzzled species (such as rhinoceros) are less selective of what they eat. This agrees with the general finding that body size of herbivores is a critical determinant of their relationship with other species in the ecosystem, including plants (11). In addition, Lundgren et al. found similar effects on the diversity or abundance of native plants by invasive and feral herbivores compared with native herbivores. This is 1Co-Centre for Climate + Biodiversity and Water, School of

Natural Sciences, Trinity College Dublin, Dublin 2, Ireland. 2School of Natural Sciences, Zoology, Trinity College Dublin, Dublin 2, Ireland. Email: [email protected]; [email protected] SCIENCE science.org

unexpected because invasive species, by definition, are expected to have bigger impacts than noninvasive introduced species. However, very few invasive species were assessed (invasives, n = 3; feral, n = 6). The native or non-native origin of a species can be a proxy for one or more of four underlying processes. These include the strength of a species’ association with humans, disjunct evolutionary history (when native and introduced species do not share an evolutionary association), an expanded species pool that increases the probability of encountering those with high performance, and superior adaptation to, or tolerance of, environmental conditions encountered in the habitat to which the species is introduced (12). Therefore, origin effects, if they occur, may manifest in different ways, depending on the underlying process they are a proxy for, the functional ecology of the introduced species, and the condition of the recipient ecosystem. A better understanding of how traits influence the likelihood of species being introduced through different pathways, and the effects of those traits on different recipient ecosystems, is likely to lead to more powerful insights on the ecology of introduced species compared with a focus on species origin alone. For example, the spatially extensive (and replicated) herbivore exclusion experiment in the Nutrient Network (NutNet) project has shown that the environment mediates the effects of herbivory on grassland communities through the effect of herbivores on ground-level light conditions (13). Where light is limiting, such as in productive grasslands, herbivores increase opportunities for the coexistence of plant species by removing biomass and disturbance, thus reducing competition for light. The use of functional traits, such as body size or muzzle width, together with the recipient environmental context, would enable more specificity in explaining and ultimately predicting the effects of large herbivores in different ecosystems. This has considerable value for land management, conservation, and restoration. There is a wide range of introduced species within many ecosystem contexts, and only some of these combinations are harmful. There is a separate question about the desired versus unwanted effects of native and introduced herbivores. For example, removal of biomass or litter through herbivory may suppress fire risk, but whether this is good or bad depends on the current and desired state of the ecosystem that is affected. Although Pleistocene communities (2.6 million to 11,700 years ago) contained many more largebodied herbivores than contemporary ecosystems, particularly in the Americas and Europe, profound changes in ecosystems

throughout the Anthropocene mean that there are no well-understood, well-defined, or even achievable “natural” reference conditions for many ecosystems. The desired state of a contemporary ecosystem can take many forms, including being closer to a historical reference state, being more diverse in terms of a particular group of species, or exhibiting desired ecosystem functions, composition of species, or ecosystem services. If the desired state is a historical reference state, in which only native species are present (a “wild” or prehuman ecosystem), then the presence of any introduced species moves the community away from that desired state. However, if the desired state is one in which the ecosystem function of large mammalian herbivores is restored or maintained, then introduced species could provide that function. A functional traits approach could be useful in reconciling Lundgren et al.’s findings with previous meta-analyses and known effects of problematic invasive herbivores. The search for predictive traits of problematic invaders is urgent because resources for ecosystem management are scarce. It is promising that some of the variance in the effects of native and introduced herbivores on plants can be resolved through functional traits of the herbivores. However, more evidence is needed on the broader role of herbivore body size, dietary selectivity, and other functional traits in combination with, not instead of, origin. j RE FE REN C ES AN D N OT ES

1. E. J. Lundgren et al., Science 383, 531 (2024). 2. J. D. Parker, D. E. Burkepile, M. E. Hay, Science 311, 1459 (2006). 3. C. Fløjgaard, P. B. M. Pedersen, C. J. Sandom, J.-C. Svenning, R. Ejrnæs, J. Appl. Ecol. 59, 18 (2022). 4. K. Bunney, W. J. Bond, M. Henley, Biotropica 49, 395 (2017). 5. P. S. Barton et al., Trends Ecol. Evol. 34, 950 (2019). 6. E. le Roux, G. I. H. Kerley, J. P. G. M. Cromsigt, Curr. Biol. 28, 2493 (2018). 7. M. Cardillo et al., Science 309, 1239 (2005). 8. Invasive Species Specialist Group (ISSG), Global Invasive Species Database, Version 2015.1 (2015); www. iucngisd.org/gisd/ [accessed 8 January 2024]. 9. Intergovernmental Science-Policy Platform on Biodiversity and Ecosystem Services (IPBES), “Thematic assessment report on invasive alien species and their control of the Intergovernmental Science-Policy Platform on Biodiversity and Ecosystem Services,” H. E. Roy, A. Pauchard, P. Stoett, T. Renard Truong, Eds. (IPBES Secretariat, 2023). 10. K. Campbell, C. J. Donlan, Conserv. Biol. 19, 1362 (2005). 11. R. M. Pringle et al., Curr. Biol. 33, R584 (2023). 12. Y. M. Buckley, J. Catford, J. Ecol. 104, 4 (2016). 13. E. T. Borer et al., Nature 508, 517 (2014). AC KN OW LE DG M E N TS

The Co-Centre for Climate + Biodiversity and Water is managed by Science Foundation Ireland (SFI), Northern Ireland’s Department of Agriculture, Environment and Rural Affairs (DAERA), and UK Research and Innovation (UKRI) and is supported through the UK’s International Science Partnerships Fund (ISPF) and the Irish government’s Shared Island initiative. 10.1126/science.adn4126 2 FEBRUARY 2024 • VOL 383 ISSUE 6682

479

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

MEDICINE

Degradation trumps mutations in cancer Redirecting targeted proteins for degradation can overcome acquired drug resistance By R. Eric Davis and Jason R. Westin

A

mino acid sequence determines a protein’s behavior, including drug susceptibility and its interacting proteins. Genomic instability enables cancer development but can also produce variants of proteins that resist targeted therapy, sometimes by changing the protein’s function and protein partners. On page 496 of this issue, Montoya et al. (1) show that a small molecule can be used therapeutically to redirect a targeted protein, even with drugresistant mutations, to endogenous degradation pathways in patients with chronic lymphocytic leukemia (CLL).

both irreversible (covalent) and reversible, are well-tolerated and effective against various B cell cancers and are now approved as initial therapy for CLL. However, their efficacy is limited by acquired resistance (2, 3). Mutations that commonly confer resistance to covalent BTKis affect predictable residues in the adenosine triphosphate (ATP)–binding pocket of BTK’s tyrosine kinase domain (TKD): Cys481 Ser (the target of covalent linkage by irreversible BTKis) or Thr474 Ile (the “gatekeeper” residue of the pocket). Other resistance mechanisms maintain the downstream effects of BTK despite loss of its kinase activity. For example, changes in epigenetic regulation (4) and/or cell signaling

Degrading drug-resistant mutants BCR signaling results in phosphorylation of Tyr551 in BTK. In the presence of wild-type BTK (left), dimerized BTK is autophosphorylated at Tyr223 and directly phosphorylates PLC-g2 at Tyr1217. In B cells with mutations that inactivate the TKD and resist reversible inhibitors (right), phosphorylation of Tyr223 does not occur; instead, mutant BTK binds HCK, leading to activation of PLC-g2. Both wild-type and mutant BTK are degraded by CRBN-engaging degraders, along with CRBN neosubstrates. Kinase-inactive TKD

Wild-type BTK BCR

BTK

PIP3 BTK

PI3K

Tyr223

Tyr223

SYK and SFTK

Tyr551

Tyr551

PIP3

HCK

Tyr1217 PLC-g2

Tyr551 Calcium flux, NF-kB activation

TKD

IKZF1, IKZF3, others

BTK degrader CRBN

BCR

TKD'

PI3K SYK and SFTK

IKZF1, IKZF3, others

Ubiquitination and degradation of BTK and CRBN neosubstrates BCR, B cell receptor; BTK, Bruton’s tyrosine kinase; CRBN, cereblon; HCK, hematopoietic cell kinase; IKZF1, Ikaros; IKZF3, Aiolos; NF-kB, nuclear factor kB; PI3K, phosphatidylinositol 3-kinase; PIP3, phosphatidylinositol 3,4,5-trisphosphate; PLC-g2, phospholipase C–g2; SFTK, Src family tyrosine kinase; TKD, tyrosine kinase domain.

The B cell receptor (BCR) signals through downstream kinases to drive B cell development, maintenance, and immune function. Many B cell cancers also depend on BCR signaling, implicating downstream kinases as therapeutic targets. Among these, Bruton’s tyrosine kinase (BTK) is especially attractive, owing to its restricted expression and function in B cells and myeloid cells, limiting the toxicity of on-target, off-tumor BTK inhibitors (BTKis). Highly specific BTKis, Department of Lymphoma and Myeloma, MD Anderson Cancer Center, Houston, TX, USA. Email: [email protected]

480

2 FEBRUARY 2024 • VOL 383 ISSUE 6682

(5) promote activation of the downstream effector, phospholipase C–g2 (PLC-g2), which is normally phosphorylated at Tyr1217 by BTK, or PLC-g2 may acquire activating mutations. Surprisingly, resistance to reversible BTKis comes from other TKD mutations (Leu528 Trp, Val416 Leu, and mutations of Cys481 to Phe or Tyr) that impair BTK kinase activity yet maintain downstream consequences of BCR signaling, such as PLC-g2–dependent calcium flux and nuclear factor kB (NF-kB) activation (6, 7). This suggests such mutations are neomorphic, changing BTK from a kinase to a scaffold protein (see the figure). Montoya et al. used a cell line model to show that the

Leu528 Trp mutation changes the binding of BTK to new proteins that include hematopoietic cell kinase (HCK) (8) and integrin-linked protein kinase (ILK), on which cells become dependent for viability and BCR signaling. Kinase-inactive BTK-Cys481 mutants activate HCK, which then phosphorylates PLC-g2 on Tyr1217 to a degree equivalent to that of cells with wild-type BTK (8). How Leu528 Trp and Val416 Leu mutations activate HCK is less well understood, because phosphorylation of PLC-g2 at Tyr1217 is reduced (6). Montoya et al. used NX-2127, a proteolysis-targeting chimera (PROTAC), to degrade BTK. The foremost class of “molecular glues,” PROTACs comprise one component that binds a protein of interest (POI), and another that engages a predetermined new protein partner (an E3 ubiquitin ligase complex), leading to POI ubiquitination and proteasomal degradation (9). Thus, PROTACs (degraders) catalyze POI elimination and are recycled for more activity after bound POIs are degraded. The requirements of PROTACs for POI binding are less specific than for conventional inhibitors, making many “undruggable” proteins targetable. Using cell lines expressing wild-type BTK or various BTKi-resistant mutants, Montoya et al. found that NX-2127 could eliminate all forms of BTK protein and prevent the calcium response of BCR signaling. This confirmed that reversible BTKi-resistant, kinaseinactive mutants have scaffold function, and suggested that BTK-targeting PROTACs may overcome the multitude of resistance mechanisms that plague clinical use of BTKis. Preclinical studies showed that NRX-0492, a BTK-targeting PROTAC similar to NX-2127, was as effective against treatment-naïve CLL cells as the BTK-Cys481–targeting irreversible inhibitor ibrutinib and was highly effective against Cys481 Ser-mutant cells that are resistant to ibrutinib (10). NX-2127 is now in a clinical trial (NCT04830137) for relapsed or refractory B cell malignancies, including CLL. Montoya et al. report that NX-2127 treatment rapidly degraded both wild-type and BTKi-resistant BTK in CLL patients’ blood cells. NX-2127 was safe and well-tolerated, and 11 of 14 CLL patients (with diverse BTK mutations) had reduction in anatomically measurable disease. Interpretation and extrapolation for this study must consider that NX-2127 binds, as do NRX-0492 and other BTK degradscience.org SCIENCE

ers, through an imide moiety to cereblon (CRBN), the substrate receptor of an E3 ubiquitin ligase complex. CRBN-binding, imide-containing drugs cause ubiquitination and degradation of a set of “neosubstrate” proteins, including the transcription factors Ikaros (IKZF1) and Aiolos (IKZF3), which underlies the activity of lenalidomide and related immunomodulatory drugs (IMiDs) against multiple B cell–derived cancers (9). PROTACs that target BTK are “specific” in that they add only BTK to CRBN’s set of neosubstrates, which remain targeted but can vary: NRX-0492 and NX2127 catalyze the degradation of IKZF1 and IKZF3, but another BTK degrader, NX-5948 (currently being tested in NCT05131022), does not (11). Therefore, it is uncertain how much of NX-2127’s activity is due to degradation of BTK, versus other CRBN neosubstrates, especially because the combination may be synergistic (12). Whether as a kinase or scaffold, functional BTK protein is needed in B cell cancers and therefore degraders are likely to be effective, even after BTKi resistance develops. The study of Montoya et al. predicts a promising clinical future for degraders, and further development of PROTACs is ongoing, such as for increased specificity of neosubstrates (13). However, many questions remain about BTK degraders. Will they be best used after BTKi resistance develops, or should they be used first, especially given that CRBN-modulating degraders have other desirable targets for treating B cell cancers? It is unknown whether resistance to CRBNmodulating BTK degraders will emerge, as for IMiDs in myeloma treatment (14), such as through mutations in BTK at the PROTAC binding site or in PLC-g2. Will BTK degraders replace BTKis and/or IMiDs in targeted combinations that are showing promise in clinical trials (15)? Patients and oncologists look forward to the answers. j RE F ER E NC ES AND NOTES

1. S. Montoya et al., Science 383, eadi5798 (2024). 2. C. I. E. Smith, J. A. Burger, Front. Immunol. 12, 689472 (2021). 3. A. Chirino et al., Genes 14, 2182 (2023). 4. A. L. Shaffer III et al., Blood Cancer Discov. 2, 630 (2021). 5. J. Rip et al., Eur. J. Immunol. 51, 2251 (2021). 6. E. Wang et al., N. Engl. J. Med. 386, 735 (2022). 7. J. Qi et al., Blood Adv. 7, 5698 (2023). 8. K. Dhami et al., Sci. Signal. 15, eabg5216 (2022). 9. M. Békés et al., Nat. Rev. Drug Discov. 21, 181 (2022). 10. D. Zhang et al., Blood 141, 1584 (2023). 11. M. Aryal et al., Sci. Signal. 15, eabn8359 (2022). 12. Y. Yang et al., Cancer Cell 21, 723 (2012). 13. T. M. Nguyen et al., Nat. Chem. (2023). 14. S. Bird, C. Pawlyn, Blood 142, 131 (2023). 15. J. Westin et al., J. Clin. Oncol. 41, 745 (2023). AC KN OW LE D GMENTS

J.R.W. has advised or received clinical research support from Abbvie, ADC Therapeutics, AstraZeneca, Bristol-Myers Squibb, Genentech, GenMab, Kite, Janssen, Kymera, Merck, Morphosys/Incyte, Novartis, Regeneron, and SeaGen. 10.1126/science.adn4945 SCIENCE science.org

MATERIALS SCIENCE

Plastics that lose their temper on demand Multiple properties can be programmed into a single dynamic material by using heat By Haley P. McAllister and Julia A. Kalow

A

s humans continue to explore harsher and less predictable settings, tools designed for specific use cases are not sufficient. Furthermore, survey vessels for oceanography and spacecraft for planetary exploration have limited space and thus must prioritize equipment on the basis of adaptability and versatility. The tools developed for sea and space exploration are deliberately unspecialized (1). A truly utilitarian invention warrants a single material that can be fashioned with many different properties. In a resource-scarce environment, it must also be reusable. Imagine an astronaut performing wheel repair on a rover by converting a single material into a wrench, adhesive, and a patch. On page 545 of this issue, Boynton et al. (2) report a class of polymeric materials with reversible, temperatureprogrammable mechanical properties that could be envisioned in such a scenario. Using temperature to determine the mechanical properties of materials is not a new phenomenon. Thousands of years ago, through trial and error, blacksmiths developed tempering processes involving heating and rapid cooling to tune the properties of steel, including hardness, strength, and ductility (3). As a result, steel can be used for diverse applications, from knives to springs to structural beams. Compared with steel, plastics are lightweight and can be processed at lower temperatures, making them ideal for supporting the exploration of resource-scarce environments. Processes such as chemical cross-linking can transform thermoplastic polymers (those that can be readily melted and recast) into tough, elastic polymer networks. Traditional cross-linking, however, is based on irreversible bond formation, and plastic “tempering” requires reversible changes in structure. In covalent adaptable networks, reversible covalent bonds allow the network of constituent polymer chains to rearrange. This ability underlies materials that are Department of Chemistry, Northwestern University, Evanston, IL, USA. Email: [email protected]

dynamic and responsive to stimuli (4). Many types of reversible covalent chemical reactions are used in adaptable networks, including disulfide exchange, transesterification, and Diels–Alder cycloaddition. The thermodynamics of the reversible crosslink determine the topology, and thus the mechanical properties, of the network. Boynton et al. used a dynamic cross-link based on the reversible addition of thiols to benzalcyanoacetates, known as a Michael addition (5). This reaction does not require a catalyst and is activated by mild heating. Like most dissociative reactions, the equilibrium constant for carbon-sulfur bond formation is sensitive to temperature, with higher temperatures favoring the dissociated state and lower temperatures favoring the bound state. The bound state is further favored by electron-withdrawing groups on the acceptor molecule. Therefore, when a donor molecule that bears multiple thiol groups (a multi-arm thiol) is combined with linkers bearing benzalcyanoacetate acceptors, the resulting network is tightly cross-linked at room temperature—that is, a stiff thermoset. When heated, the material softens and becomes extensible before completely disassembling into monomers at around 140°C. Qualitatively, such behavior is typical of a covalent adaptable network and is not suitable for “tempered” multi-use plastics because distinct mechanical properties are only achieved at specific temperatures. Under service conditions, the stiff thermoset is reformed. The key advance that allows Boynton et al. to lock in the desired properties achieved at elevated temperatures is dynamic reaction–induced phase separation (6), which is proposed to kinetically freeze the network topology formed at a higher temperature. Thus, materials may be tempered at temperatures between their glass transition temperature (Tg; the temperature at which an amorphous glassy polymer material becomes rubbery) and the upper transition temperature at which disassembly occurs (TUT). To lower the Tg and thereby widen the tempering window, Boynton et al. used a mixture of tetra- and bifunctional thiols. For a single formula2 FEBRUARY 2024 • VOL 383 ISSUE 6682

481

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

tion, the material was held at temperatures between 60° and 120°C for 24 hours before being cooled on a metal block. Depending on tempering temperature, the resulting material ranged from a hard, brittle thermoset similar to polypropylene, to a tough thermoplastic with stiffness comparable with that of high-density polyethylene, to a soft, extensible polymer that can be used as an adhesive (see the figure). Overall, for a single material, a greater than 10-fold change in Young’s modulus (a measure of stiffness) and 20-fold change in strain at break were achieved through tempering. The mechanism for temperature-induced changes in mechanical properties proposed by Boynton et al. is supported

ferential tempering to create swords with tough spines and sharp edges. Boynton et al. patterned their material by subjecting stiff polymer network films, tempered at lower temperatures, to heated conductive surfaces. The resulting selectively softened region of the material underwent preferential deformation or failure. The material also exhibited shape memory behavior, allowing simple actuation, such as picking up and lifting a heated object. The material developed by Boynton et al. is described as “pluripotent,” analogous to a stem cell, which can differentiate into different cell types. Although researchers have sought to achieve reversible changes in material properties, the study

Pluripotent plastics A pluripotent plastic material can be heated to temperatures between 60° and 120°C to achieve distinct properties, which are maintained upon cooling. The ratio of bonded to disassembled linkers is determined by the equilibrium constant of a reversible reaction, which is sensitive to temperature. Above 140°C, the bonds are completely disassembled, making the material circular in addition to versatile. Soft adhesive

Pluripotent material

More disassembled Mixture

60°C

145°C

145°C

O

HS

O

O O

HS

O O

O HS

O O

O

SH

Thiol

SH

SH

SH

2 FEBRUARY 2024 • VOL 383 ISSUE 6682

Bonded state

Benzalcyanoacetate O

O2N

through simultaneous Raman spectroscopy and mechanical measurements, which showed that as the material softens at higher temperatures, the fraction of bound cross-links decreases. Furthermore, these changes are fully reversible, so a single sample may be interconverted repeatedly between different states. Notably, the properties associated with each state were maintained for at least a month at room temperature but may be entirely “reset” by heating above the critical temperature, 140°C. Although many of the experiments were performed in a low-humidity, oxygenfree environment, these materials were also shown to be compatible with ambient atmosphere and did not form detectable amounts of disulfide. To achieve the remarkable strength and toughness of hierarchically assembled biological materials (such as tendons or bone), researchers have created multimaterial structures through patterned cross-linking (7) or crystallization (8). Although patterning is often achieved with the application of light, heat may also be applied with spatial control. Historically, bladesmiths used dif482

More bonded

Disassembled state

Tetrathiol linker Dithiol linker

Hard plastic

110°C

CN

Keq

S

OR O2N

O CN

OR

of Boynton et al. demonstrates the value of arresting a material’s properties in distinct states. A better understanding of the mechanism by which properties are “locked in” could further extend the lifetime of differentiated states under harsher conditions. The findings establish a new design goal for the materials science field and will inspire the development of other pluripotent polymers with an even wider range of useful properties accessed from a single material. j R E F E R E N C ES A N D N OT ES

1. NASA, “Hubble Space Telescope—Tools,” fact sheet (2023); https://science.nasa.gov/wp-content/ uploads/2023/04/hst-tools-fact-sheet.pdf. 2. N. R. Boynton et al., Science 383, 545 (2024). 3. G. R. Speich, W. C. Leslie, Metall. Trans. 3, 1043 (1972). 4. V. Zhang, B. Kang, J. V. Accardo, J. A. Kalow, J. Am. Chem. Soc. 144, 22358 (2022). 5. Y. Zhong, Y. Xu, E. V. Anslyn, Eur. J. Org. Chem. 2013, 5017 (2013). 6. K. M. Herbert et al., Chem. Sci. 11, 5028 (2020). 7. N. D. Dolinski et al., Adv. Mater. 30, 1800364 (2018). 8. A. K. Rylski et al., Science 378, 211 (2022). AC K N OW L E D G M E N TS

H.P.M. and J.A.K. receive support from the National Science Foundation under award CHE-1847948. 10.1126/science.adn3980

PHYSIOLOGY

Arterial pulses link heart-brain oscillations A central baroreceptor monitors arterial pressure to modulate brain activity By Owen P. Hamill

I

n 1942, the electrophysiologist E. D. Adrian published recordings from the hedgehog olfactory bulb (OB) indicating three basic classes of electrical oscillation: a respiration-nasal-airflow–related oscillation (RRO), a sensory-odor–induced oscillation, and an oscillation considered intrinsic to the local neural network (1). These oscillations are important because by moving the neuronal resting membrane potential toward and away from the spike threshold, they can synchronize spike activity in local and remote neural networks. All three classes of oscillation are dependent on synaptic transmission; have been recorded in other brain regions, most notably in the human cerebral cortex and hippocampus; and are considered fundamental to how the brain normally processes information (2). On page 494 of this issue, Jammal Salameh et al. (3) report a fourth class of oscillation—a heartbeat-related oscillation (HRO) evoked by arterial pressure pulsations and transduced by central baroreceptors. Jammal Salameh et al. focused on understanding the mechanism underlying a prominent local field potential (LFP) oscillation of ~4 Hz recorded in the rat OB mitral cell (MC) layer. LFPs are electrical signals generated in the local extracellular space surrounding cells and include the combined activity of excitatory and inhibitory synapses as well as any spike activity. As with the studies of Adrian, the large rodent OB provided a highly accessible region of the brain for recording. But, in the case of the study of Jammal Salameh et al., the nature of the nose-brain preparation (NBP) rat model (4) meant that respiratory- and sensory-evoked mechanisms could be excluded because of the lack of lungs, nasal airflow, and ascending respiratory pathways. Also, the absence of the heart Department of Neurobiology, The University of Texas Medical Branch at Galveston, Galveston, TX, USA. Email: [email protected] science.org SCIENCE

GRAPHIC: A. MASTIN/SCIENCE

mmHg

in this model would seem to exclude ral assemblies associated with new Pressure pulses link breathing, heartbeat, a role for heartbeat-induced presmemories by reigniting them during and brain rhythmicity sure pulsatility (~4 Hz for the rat). sleep, thereby countering any loss. Heartbeat- and respiratory-related intracranial pressure (ICP) pulsations However, somewhat serendipitously, It has also been proposed that brain are transduced by Piezo-type mechanosensitive ion channels (PIEZO2 the peristaltic pump used to perfuse and body oscillations are linked (10, channels) to generate electrical oscillations. These oscillations may the NBP with oxygenated solution 11) and even harmonically related, show coupling to the major brain oscillations (d, u, a, b, and g) that are consisted of eight individual rollers, forming a binary hierarchy of center relatively conserved in rats and humans, in contrast to their heart rate so for a typical pump rotation rate frequencies in which the fundamenand breathing frequencies. Center frequencies (shown in parentheses) of 30 rpm (0.5 Hz), pressure pulsatal frequency is 1.25 Hz, correspondsupport a binary hierarchical brain-body model in humans in which tions of 4 Hz were generated. By ing to the healthy human heart rate heart rate is the fundamental frequency. Polysynaptic pathways also recording the LFP oscillations and of 75 bpm (see the figure). In these contribute to lung-heart, lung-brain, and heart-brain coupling. the measured pressure pulsations siproposals, the HRO may now provide multaneously, the authors identified the mechanism for mechanoelectriCardiorespiratory ICP pulsations (human) Electrical oscillations a perfect match between the fundacal coupling of heartbeat pulsations (human) 40 mental frequency of the slow LFP oswith brain electrical oscillations. d 1–4 Hz (2.5) Brain oscillations cillation and the pressure pulsation The heartbeat does not behave as 30 frequency over a pulsation range of a metronome but rather is highly u 4–6 Hz (5.0) ~2 to 4 Hz. Furthermore, multiple variable. Apart from in response to 20 a 8–12 Hz (10) lines of evidence indicated that the energetic and emotional demands, LFP was neither a pulsatility artifact the heart rate is modulated by norb 15–25 Hz (20) 10 nor was it mediated by axonal-synmal breathing in a phenomenon aptic transmission, as in the case of called respiratory sinus arrythmia g 30–50 Hz (40) 0 the other three oscillations. Instead, (i.e., the heart rate increases during 0 5 10 15 20 25 30 35 40 45 they investigated whether a central inspiration and decreases during exTime (seconds) baroreceptor was at play in which piration) (12). This is most obviously pressure pulsations were transduced seen with resonance breathing, when by pressure-activated channels, such breathing frequency is reduced to 0.1 as Piezo-type mechanosensitive ion Hz (6 bpm) so that the baroreceptor channel component 2 (PIEZO2) reflex (involving a feedback loop beRat brain (5), which is expressed in MCs (6). tween peripheral baroreceptors and Notably, Jammal Salameh et al. were brainstem regulatory nuclei), blood able to exactly simulate the LFP pressure changes, and respiration Breath rate (Hz) Heart rate (Hz) Human 0.2–0.3 Human 1–1.5 (1.25) waveform and the observed peak are synchronized (i.e., they resonate). Rat 1.2–2.5 Rat 4–8.2 increase in spontaneous MC spiking Resonance breathing, particularly inby considering the specific pressure volving nasal inspiration, is used in sensitivity and gating properties of meditative practices, and it is recogthe PIEZO2 channels. nized to reduce anxiety and alleviate Why was the HRO not detected by panic attacks as well as other psychiAdrian or in any of the many studies atric disorders (12). These effects are Rat lungs Rat heart of brain oscillations carried out over dependent on cardiac and arterial the past 80 years? Jammal Salameh afferent spike activity transduced by et al. found that heartbeat-modulated spikstates, from the highly aroused to the deepest PIEZO channels and transmitted mostly via ing occurred only in a small subset (~15%) sleep state. It is interesting that the spiking of the vagus nerve to specific brain regions (13). of OB neurons, involved only a modest inmany (#60%) specialized tactile receptors in It is also expected that the large sinusoidal crease (~10%) in firing rate, and peaked 20 the human fingertips and stretch receptors in fluctuations in heart rate associated with ms after the heartbeat in recordings from human muscle are either phase locked (i.e., resonance breathing would increase intracraawake, head-fixed mice. By contrast, nasal one spike follows every heartbeat) or modunial pressure pulsatility and HRO to play an breathing–induced RRO entrained a higher lated by heartbeat-induced arterial pressure added beneficial role in brain function. j percentage of OB neurons (~94%), involved a pulsations generated in the surrounding RE FE REN CES A ND N OT ES larger increase in firing (~30%), and peaked tissue (7). This is consistent with the higher 1. E. D. Adrian, J. Physiol. 100, 459 (1942). 150 ms after inspiration. Without a specific density of PIEZO2 channels in peripheral ver2. G. Buzsáki, M. Vöröslakos, Neuron 111, 922 (2023). focus, these subtle heartbeat-related effects sus central neurons (6). Although the brain 3. L. Jammal Salameh et al., Science 383, eadk8511 (2024). 4. F. Pérez de los Cobos Pallarés, D. Stanić, D. Farmer, M. could be missed or misinterpreted as pulmay ignore the incoming spikes as noise, it Dutschmann, V. Egger, J. Neurophysiol. 114, 2033 (2015). satility artifacts. Notably, equally subtle is possible that the barrage of afferent spikes 5. B. Coste et al., Science 330, 55 (2010). heartbeat-related responses were recorded arriving within a narrow time window (i.e., 6. J. Wang, O. P. Hamill, J. Integr. Neurosci. 20, 825 (2021). 7. I. Birznieks, T. W. Boonstra, V. G. Macefield, PLOS ONE 7, by the authors in the mouse hippocampus 0.05, one-sided one-sample t test (H0: m = 0). Green dot is the mean, and orange line is the median. (M) Percentages of neurons exhibiting three projection modes: ipsilateral-only (purple); contralateral-only (green), and bilateral (blue) among all neurons projecting to each of 11 target areas.

dinal orientations (low T-L index values; Fig. 6C, top, magenta) within both ipsilateral and contralateral CA3 region but became preferentially orientated in transverse directions in CA1 regions (high T-L index values; Fig. 6D, top, green). Similar but smaller differences in T-L indices were also found for Schaffer collateral axonal arbors of ventral CA3 neurons (Fig. 6, C and D, bottom). The differences in T-L indices between dorsal and ventral CA3 neurons were significant, as shown by the box plots of the average T-L indices for Schaffer collateral axons in CA3 and CA1 regions (twoway ANOVA, P < 0.01 for all; Fig. 6H). These results indicate that Schaffer collateral axons of CA3 neurons first traverse predominantly in the longitudinal direction in CA3 regions and then project transversely to innervate CA1 neurons. Furthermore, dorsal CA3 neurons generally exhibit a wider span of Schaffer axon arbors along longitudinal axis than ventral CA3 neurons (Fig. 6J). This supports the notion that transverse and longitudinal projections of Schaffer collaterals are both extensive, and there is a more prominent intraHIP information flow from dorsal to ventral regions. Next, we examined transverse preference of DG mossy fiber axons, which were predominantly ipsilateral projections. Examples of mossy fiber projections at various longitudinal locations are shown in Fig. 6E. Heatmaps of T-L indices showed that mossy fiber axonal arbors were mainly orientated transversely in both DG and CA3 regions, with a slight longitudinal orientation in the dorsal CA3 region (Fig. 6F). Mossy axonal arbors of dorsal DG neurons exhibited more longitudinal orientations than those of ventral DG neurons (Fig. 6I), consistent with a wider longitudinal spread of mossy axonal arbors of dorsal DG neurons (Fig. 6K). This suggests that dominant D-V information flow occurs early in the HIP core circuit at the stage of DG-to-CA3 projections. Thus, whereas intra-HIP axons of DG and CA3 neurons largely project transversely, dorsal HIP neurons exhibit substantially more longitudinal spread of axonal arbors than ventral HIP neurons (Fig. 6G).

2 February 2024

Axonal arbor distribution depends on transverse locations of the soma

Previous studies have shown that CA3 neurons within the transverse plane of HIP exhibit heterogeneity in their connectivity and function (15, 38, 43, 44). We systematically examined the heterogeneity of axon projection patterns of HIP neurons with their soma located within the same transverse plane along the pyramidal cell layer (transverse axis) in CA fields. We defined the soma locations of HIP neurons within the transverse plane by their distance from the DG along the P-D axis (Fig. 7A and materials and methods). First, we examined the distribution of Schaffer axon tips of individual CA3 neurons in CA1. We further defined the depths of Schaffer axon tips of CA3 neurons by their locations in CA1 layers, with stratum oriens as the deepest layer (Fig. 7A and materials and methods). We found that the center-of-mass of Schaffer axon tips from distal CA3 neurons were preferentially located close to the stratum pyramidale and stratum oriens (Fig. 7B), whereas those from proximal CA3 neurons were mostly distributed in the stratum radiatum (Fig. 7C). We found a significant correlation of soma locations of CA3 neurons with the locations of their ipsi- and contralateral axon tips along P-D axis (n = 765, Pearson correlation, P < 0.0001 for all; Fig. 7D) and with the depth in CA1 layers (n = 765, Pearson correlation, P < 0.0001 for all; Fig. 7E). This is consistent with previous findings (38, 43, 44). Such topographic relationship between soma location and Schaffer collateral axon tips in the transverse plane could support the functional mapping of CA3 to CA1 neuronal activities in spatial information processing. Overall, the above topographic correlations exhibited D-V differences, which were stronger with P-D locations (Fig. 7D, inset) but weaker with axon-tip depths in CA1 layers for dorsal than ventral CA3 neurons (Fig. 7E, inset). On average, the Schaffer axon tips of individual axons covered an area with a P-D width of 330 ± 5 mm and a depth width of 146 ± 2 mm in CA1 (n = 765 neurons). Our results support the notion that CA3 neurons along the P-D axis (transverse axis) form selective connections 8 of 15

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

A Axon in target areas

CA1-enriched target patterns

SUB-enriched target patterns

SUBr

ProS-enriched target patterns

Soma

Fraction in subregion

ISOC OLFC HPFC CTXspC ISOR OLFR HPFR CNUR HYR THR MBR 0 CA1 SUB ProS SUBr

50

20

2

E

1 mm

H PF R

C

H PF C

R

IS O

C N U

R

H YR

TH

R

O LF R

IS O

C

C TX SP

CNUC OLFC CTX CTX ISOR OLFR CNUR HPFR ISOC HPFC THR HYR MBR

R

C

SUB

Caudal

1

0

Fraction in subregion

HPFR OLFR CTX CNUR HYR CTX CNUC HPFC OLFC ISOR THR ISOC MBR

C R

I

PRE POST MEC

R

1

C

-1

J Collateral index

THR CNUR HYR MBR HPFC HPFR OLFC CTX C CNUC ISOC CTX R ISOR OLFR

HYR MBR ISOR CNUR OLFR CTX R CNUC OLFC CTX C HPFR THR ISOC HPFC

R

1 mm

Axons from SUBr

CNUC OLFC CTX C CTX R ISOR OLFR CNUR HPFR ISOC HPFC THR HYR MBR

HYR MBR ISOR CNUR OLFR CTX CNUC OLFC CTX HPFR THR ISOC HPFC

C

LEC

1 mm

CTXspC

Axons from ProS

Axons from SUB

CA SUB PAR PRE POST ProS LEC MEC

8

CA1

SUB

ProS SUBr

LEC

6

1 mm

4 2 0 LEC

CA

ProS

SUB

MEC

POST

MEC

PRE

Neurons with preferred intra-HPF target

MEC

0

PRE

-0.4

R

MBR

ISOR

CTX

THR

HYR

OLFR

CNUR

C

POST

Pearson correlation

LEC

CNUC

0

0.6

ProS

OLFC

1

OLFR

OLFR

CA SUB

CTX

Rostral

Caudal

ISOC OLFC CTX C CNUC ISOR OLFR HPFR CTX R CNUR THR HYR MBR

L Correlation of proj. strength

ISOC

V

K Intra-HPF targets

CA1 SUB ProS SUBr

D

Proj. strength (Nor.)

Density

Intra-HPF targets

CNUC

CTXspC

HPFR OLFR CTX R CNUR HYR CTX C CNUC HPFC OLFC ISOR THR ISOC MBR

OLFC

1 mm

CNUC

OLFC

Pearson correlation

CTXspC

ProS CA

CNUC

Proj. strength (Nor.)

CNUC CTXspC

THR CNUR HYR MBR HPFC HPFR OLFC CTX CNUC ISOC CTX ISOR OLFR

Soma

G

ProS

SUB

Axons from CA1

Extra-HPF targets

Target areas

SUBr

CA1

Preferred Hintra-HPF target

F

M BR

0

0

Target number

OLFC

ProS SUBr

4

1 2 3 4 5 6 7 8 9 10

D

SUB

6

C

Target number = 3 Collateral index = 2

40

CA1 8

O LF C

Area2

60

R

Area1

10

CA1 SUB ProS SUBr

C N U

Area3

C

80

C TX SP

Soma

Collateral index

B

Neuron number (%)

100

LEC 1 mm

LEC

CA CA HYR

long arbors

HYR

short arbors

Extra-HPF targets 1 3 4 5 6 7 8 2 9 10 11 12 13 14 15 16 17 18 20

Fig. 5. Projection profiles of CA1, SUB, and SUBr neurons. (A) Top, dot plot showing distinct target patterns of neurons enriched in the CA1, SUB, ProS, and SUBr regions (top 15 neuron numbers in each region). Bottom, bar graph showing percentages of CA1 (magenta), SUB (purple), ProS (cyan), and SUBr (orange) neurons with each target pattern. (B) Left, schematic diagram showing the definition of collateral index for an axon or target areas receiving inputs from this axon based on the number of its co-projected targets. Right, the percentage of CA1 (magenta), SUB (purple), ProS (cyan), and SUBr (orange) neurons with projections to different numbers of targets. Chi-square test between two groups of neurons in CA1, SUB, ProS, and SUBr subregions was performed, and Qiu et al., Science 383, eadj9198 (2024)

2 February 2024

P < 0.0001 for all. (C) Summary (mean ± SEM) of collateral indices for individual axons in 13 primary target areas for CA1 (magenta), SUB (purple), ProS (cyan), and SUBr (orange) neurons, with the target areas ordered by the mean axon collateral indices of all four groups of neurons. Lines indicate two-sided Wilcoxon rank sum test, P < 0.05. (D to G) Top, examples of three neurons (arrows) located in CA1 (D), SUB (E), ProS (F), and SUBr (G) regions showing coprojection pattern for OLF, CNU, and CTXsp through the caudal pathway (OLFC, CNUC, and CTXspC) or co-projection patterns for CNUC and CTXspC. Bottom, Pearson correlation coefficient (color-coded) of the axonal arbor lengths for projection strength between different target pairs for neurons in CA1 (D), SUB (E), ProS (F), 9 of 15

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

and SUBr (G). The 13 target areas are HPFC, HPFR, ISOC, ISOR, OLFC, OLFR, CTXspC, CTXspR, CNUC, CNUR, THR, HYR, and MBR, which were sorted by hierarchical clustering analysis of correlation coefficients. Solid squares indicate target patterns shared by CA1, SUB, ProS, and SUBr groups of neurons; dashed squares are group-specific target patterns. (H) Heatmap of averaged total arbor lengths in intraHPF (top) and extra-HPF (bottom) targets in each subgroup (numbered below) after clustering analysis of arbor lengths of individual axons in all intra-HPF targets (see fig. S11A). Preferred intra-HPF targets are labeled in the white boxes above. The axonal arbor lengths were normalized by the maximal value found among intraHPF and extra-HPF targets, respectively. Middle, fraction of neurons with soma located in CA1, SUB, ProS, and SUBr regions (color-coded) and soma distributions along D-V axis. (I) Summary (mean ± SEM) of collateral indices for neurons with various preferred intra-HPF targets: CA1 (magenta), SUB (purple), ProS (cyan), and SUBr (orange) neurons. Lines indicate two-sided Wilcoxon rank

with CA1 neurons in a manner that depends on both the soma location and dendritic domain of CA1 neurons. In other words, more proximally located CA3 neurons (relative to DG) send projections to more distally located CA1 neurons at more superficial domains of their dendrites, as depicted schematically in Fig. 7F. Next, we investigated whether axon arborization of HIP CA1 neurons in target areas depends on the location of their soma along the P-D axis. We used PCA to characterize the distribution of axon tips at target areas for all CA1 neurons mapped in this study. Examples of CA1 axonal arbors in the endopiriform nucleus (EP) and ENT are shown in Fig. 7, G to J. The correlation between axon arbor location and its soma location was maintained across LEC and MEC, suggesting coordinated projections to the entire ENT (Fig. 7G). The soma locations and cross-hemispheric CA1-to-CA1 axon tips were topographically organized in a symmetric manner (Fig. 7, K and L), suggesting a highly organized interhemispheric circuit for coordinated physiological functions of CA1 neurons. Overall, axonal arbor distributions of CA1 neurons in many target areas correlated with the P-D positions of their soma in CA1 (Fig. 7M). We also examined the relationship between the soma depth of CA1 neurons and their axon arborizations in target areas. The projection strength in septum and ENT exhibited a negative correlation with the soma depth (fig. S12), implying a sublayer difference in HIP functions such as theta oscillations (20). These results revealed the topographic organization of the CA3 and CA1 neuron soma within the HIP and their coordinated axon arbor distribution in intra- and extra-HPF areas. Axonal arbor distribution depends on the longitudinal location of soma

Along the longitudinal axis, HIP neurons are known to exhibit heterogeneity in connectivity and function (9–12). We performed correlation analysis between the longitudinal location of the soma and the distribution of axon arbor terminals in various target areas. As shown Qiu et al., Science 383, eadj9198 (2024)

sum test, P < 0.05. (J) Examples of two neurons with LEC (top, axon in LEC is shown in red) and MEC (bottom, axon in MEC is shown in green) as the preferred intra-HPF target showing many and few targets, respectively. (K) Pearson correlation between axonal lengths in various pairs of intra- and extra-HPF targets. Each row presents the results from a subgroup of neurons with a preferred intra-HPF target: CA, SUB, PRE, POST, ProS, LEC, and MEC, as shown in (I). Only significant correlations are shown and are color-coded. (L) Example neurons illustrating axonal projections to a pair of intra- and extra-HPF targets showing positive (arrows the with same direction) and negative (arrows with opposite directions) correlations in their arbor lengths within co-projected target areas, corresponding to solid squares in (K). Top, two neurons sending axonal projections to LEC and OLFR. Bottom, two neurons sending axon projections to CA and HYR. Axonal arbors in the target areas are shown in color.

by two MEC-projecting HIP neurons with distinct soma locations (Fig. 8A), we observed a strong correlation between axon-tip locations (represented by the normalized principal component, NPC) in ENT and their soma locations along the D-V axis (Fig. 8, B and C). Such a correlation was found in many target areas, especially for intra-HPF target areas (Fig. 8D). For extra-HPF projections, neurons that innervated more targets were found to show narrower soma distribution along the D-V axis (as expressed by the longitudinal span in Fig. 8E). For CA1 and SUBr neurons projecting to multiple targets, their soma tended to localize ventrally (Fig. 8F). Thus, the topography of soma location along the longitudinal axis was preserved in their axon-tip distribution in the target areas. Neurons that projected to many (e.g., more than eight) primary targets, which may underlie brain-wide coordination, had their soma highly localized at specific longitudinal locations. The schematic diagram in Fig. 8G summarizes the number of neurons in seven major HIP subregions that projected to various targets along the rostral and ventral pathways, together with their longitudinal soma locations. Overall, the longitudinal location of neuronal soma is a key factor in determining the heterogeneity of extra-HPF axon projections. Discussion

By combining fMOST technology with singleaxon tracing procedures, we have mapped 10,100 single-neuron projectomes for the whole HIP at micrometer resolution and created a comprehensive three-dimensional single-cell atlas for the soma, dendrites, and axon arbors of HIP neurons along longitudinal and transverse axes, as well as the spatial transcriptome map of the CA1 region. Both projectome and transcriptome data were visualized online (https://mouse. digital-brain.cn/hipp). Our study revealed (i) 43 projectome subtypes with specific patterns of intra- and extra-HPF projection targets; (ii) axon projections exclusive to intra-HPF, cortical or subcortical area, or a combination of these areas; (iii) target pattern–dependent preferential

2 February 2024

soma locations in selective longitudinal subdomains of the HIP, (iv) bilaterally projections to a single pair of homologous targets; and (v) topographic correlation between axon arbor distribution in the target area and soma location along longitudinal and transverse axes. These findings provide unprecedented information on the brain-wide connectivity patterns of HIP neurons at single-cell resolution. Each HIP neuron sends its axon collaterals to various target areas along the rostral pathway, the caudal pathway, or both, pointing to a highly sophisticated and organized brain-wide axon projections at the single-neuron level. We have identified many previously unknown patterns of bilateral axon projections. For instance, longrange axon projections from mossy cells to extra-HPF targets (such as lateral and medial septum) may be related to their specific physiological functions distinct from those of granule cells (45, 46) and exhibit a large functional impact through brain-wide connectivity. Because GABAergic neurons in CA1 have long-range projections (47–51), we identified GABAergic projection neurons based on their soma locations and dendritic morphologies and found their axon projections to ENT. MSC-projecting GABAergic neurons sent axon projections to additional targets such as hypothalamic subregions (fig. S10J). Many septum-projecting GABAergic neurons exhibited similar dendritic morphology as TORO (theta-OFF, ripple-ON) cells (52). Many pyramidal cells projected to MSC, and their functional relevance remains to be determined. Most HIP axon projections to their downstream targets were either bilateral only or ipsilateral only, but rarely contralateral only. One of the rare targets that receive substantial contralateral-only projection was the thalamic nucleus ATN, and the functional relevance of such contralateral projection remains to be examined. We also observed the symmetric feature of CA1-CA1 bilateral connections, which may contribute to the auto-associative network in the HIP involving bilateral CA3-CA3 projections (38). These axon projections to bilateral CA1 regions from CA3 and CA1 neurons may 10 of 15

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

Qiu et al., Science 383, eadj9198 (2024)

2 February 2024

Transverse plane

Span (%)

Span (%)

T-L index (Nor.)

T-L index (Nor.)

Ventral (n=2318)

T-L index

Dorsal T-L index (n=741)

Ventral (n=544)

T-L index

T-L index

T-L index

Dorsal T-L index (n=352)

Ventral

Dorsal

Fig. 6. Longitudinal spread of A B Spa n Schaffer collaterals and mossy fiber axons. (A) Left, an example CA3 neuron sending Longitudinal axis Soma axon to CA3 (blue) and CA1 Axon (red). Black dots indicate soma. Soma Longitudinal Right, schematic illustration n n component (l) showing an axon arbor in a HIP Longitudinal axis L = (l) Transverse bin (100 bins in total along the component (t) Transverse plane T = (t) longitudinal axis) projected onto the longitudinal axis to compute T -L the longitudinal component (l) T-L index = Soma T+L CA1 CA3 and onto the transverse plane to compute the transverse CA1 component (t). Equation for T, L, CA3 and T-L index is shown. C D CA3 neurons CA3 neurons (B) Example bilateral axon proAxons in ipsi. CA1 Axons in contra. CA1 Axons in ipsi. CA3 Axons in contra. CA3 Dorsal 0.1 jections of a dorsal CA3 neuron 0.2 0.05 Dorsal 0.1 Ventral Ventral (top) and a ventral CA3 neuron 0 0 0 0 (bottom) into both CA3 and CA1 D D regions. Black dots indicate 0.8 0.8 soma. (C and D) T-L index analyzed in 100 bins along the longitudinal axis. Top, line chart -0.8 -0.8 showing the averaged T-L index V V for CA3 neurons in the dorsal -4 -2 0 2 4 -4 -2 0 2 4 -4 -2 0 2 4 -4 -2 0 2 4 (blue, n = 352) and ventral Longitudinal distance to soma (mm) Longitudinal distance to soma (mm) (orange, n = 544) halves of the D-V axis. Bottom, heatmaps DG neurons G E F showing the averaged T-L index Axons in ipsi. DG Axons in ipsi. CA3 (color-coded) in each bin for Intra-HPF axon arborization 0.25 Dorsal Ventral axonal arbors of CA3 neurons in 0.05 0 0 ipsilateral and contralateral CA3 CA1 D region (C) and CA1 region (D). 0.8 Neurons were sorted by their CA3 soma locations from dorsal to ventral. (E) Examples showing DG -0.8 axon arbors of five DG neurons DG CA3 V V D in DG region (orange) and CA3 -4 -2 0 2 4 -4 -2 0 2 4 region (blue) at various locations Longitudinal distance to soma (mm) along the D-V axis. Black dots indicate soma. (F) T-L index CA3 neurons DG neurons CA3 neurons DG neurons H I J K analyzed in 100 bins along the Axons in Axons in Axons in Axons in Axons in Axons in Axons in Axons in longitudinal axis. Top, averaged DG CA3 CA3 CA1 DG CA3 CA3 CA1 T-L index for DG neurons in the ** **** 100 1 1 15 30 **** dorsal (blue, n = 741) and ventral **** **** **** **** **** **** **** 80 **** **** **** **** **** (orange, n = 2318) halves along **** 10 20 the D-V axis. Bottom, heatmaps 60 0 0 showing the averaged T-L index 40 5 10 (color-coded) in each bin for 20 axonal arbors of DG neurons in -1 -1 0 0 0 ipsilateral DG region and CA3 DV DV DV DV D V D V DV DV DV DV DV DV ipsi. contra. ipsi. contra. ipsi. ipsi. ipsi. contra. ipsi. contra. ipsi. ipsi. region. Neurons were sorted by their soma locations from dorsal to ventral. (G) Schematic diagram illustrating intra-HIP axon projections along the longitudinal and transverse directions. Left, a DG neuron and a CA3 neuron in the dorsal HIP. Right, a DG neuron and a CA3 neuron in the ventral HIP. Dashed line indicates the transverse plane. (H) Box plots showing the T-L index for axon arbors of dorsal versus ventral CA3 neurons in the ipsilateral and contralateral CA3 and CA1 regions. (I) Box plots showing the T-L index for axonal arbors of dorsal versus ventral DG neurons in the ipsilateral DG and CA3 regions. (J) Box plots showing the span range along D-V axis covered by the axonal arbors of CA3 neurons in the ipsilateral and contralateral CA3 and CA1 regions. (K) Box plots showing the span range along the D-V axis covered by the axonal arbors of DG neurons in the ipsilateral DG and CA3 regions. In (H) to (K), two-sided Wilcoxon rank sum test was used, **P < 0.01, ****P < 0.0001. 11 of 15

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

B

l Dista

xim al Pr o

E

CA3 to CA1

0

V

CA1

Distal

1

0

Proximal

D

V

Distal

1

H P

D

P M

D

M

1 mm

A

Axon tip locations (NPC)

1 mm 1

****

1

A V

h

pt De

L

J CA1 to EP

CA1 D

P

M D

1 mm

M A 1 mm

A

0 Proximal

0

EP

Distal 1

P V M

Distal

r = 0.625 p < 0.0001

Pearson correlation

Proximal

1

0 0

Proximal

Distal

Soma locations in CA1

Fig. 7. Dependence of axonal arbor distribution on the transverse location of soma. (A) Schematic diagram of the HIP transverse plane defining P-D axes in CA3 and CA1 and the depth of CA1 perpendicular to the P-D axis. (B and C) Single-cell projectome examples showing the soma (red dots), axonal arbors (white lines), and axonal arbor terminals (cyan dots) in a 300-mm transverse HIP section overlaid with PI-stained images. The stratum (str.) layers of CA1, including stratum pyramidale, stratum oriens, stratum radiatum, and stratum lacunosum-molecular (l-m) are labeled. (B) The axonal arbor terminals of two CA3 neurons showing distinct preferences in the depth of CA1. (C) The axonal terminals of two CA3 neurons showing distinct preferences in the location along P-D axis of CA1. (D and E) Correlation of the soma location along the P-D axis of CA3 with center-of-mass of axon-tip locations along the P-D axis (D) and depth location (E) of CA1 (n = 765). Dashed line indicates linear regression. The normalized longitudinal locations of soma are color-coded with scale on the right. The two insets show the correlation coefficient (r) in transverse sections at various longitudinal locations. Arrows indicate the longitudinal location of transverse sections shown in (C) and (B), respectively. (F) Schematic illustration of topographic correlation between soma locations in CA3 P-D axis and depths and P-D locations of axonal arbor terminals in both the ipsilateral and 2 February 2024

0 Proximal

Distal 1

M

Proximal

Distal

r = 0.256 p = 0.0011

Soma locations in CA1

CA1 to contralateral CA1 Axon tip locations in CA1

Posterior

1

0

Soma locations in CA1

L

Qiu et al., Science 383, eadj9198 (2024)

Distal

DG

pth

D

LEC MEC

1 mm

De

r = 0.257 p < 0.0001

0

K

DG

I CA1 to ENT

MEC

Axon tip locations (NPC)

D

Soma

Proximal

Soma locations in CA3

G LEC

CA3

V

Soma locations in CA3 CA1

CA1

r

0

Proximal

Proximal

Soma

D

1

HIP

RHP

CA

D

0

Distal

r = 0.5 p < 0.0001

Deep

Axon tip depths

Proximal

V r

Contralateral

F

CA3 to CA1

1 D

CA3

Superficial

Distal

Proximal 0.4 mm

CA3

ENT POST PRE SUB ProS

Distal Axon tip locations

DG

P

im ox Pr

r = -0.353 p < 0.0001

Distal

ad. str. r str. l-m

DG

al

1

Axon tip Soma

. str. pyr

.

str. ori

CA1

V

DG

Anterior

Deep

str. l-m L

CA3

D

. str. pyr ad str. r

.

str. ori

CA1

Depth

Extra HPF

0.6 0.4 0.2 0.0 PL ORB RSP AON TT TR EP BLA STRd LSX PALm PVR MBmot

CA1

C Axon tip Soma

Axon tip locations (NPC)

A

Downstream target areas

contralateral CA1 regions. (G) Spatial distributions of soma in CA1 (left) and their corresponding axon tips in LEC and MEC (right). The soma locations along the P-D axis and their corresponding axon tips are color-coded with the scale on the top. The box plots show significant difference between the NPC of axon-tip locations in LEC and MEC. (H) Correlation between the soma location along the P-D axis of CA1 and NPC of axon tip locations in ENT (n = 1633). Dashed line indicates linear regression. (I) Spatial distribution of soma in CA1 and corresponding axon tips in EP presented similarly as in (G). (J) Correlation between the soma location along the P-D axis of CA1 and NPC of axon tip locations in EP (n = 160). Dashed line indicates linear regression. (K) Examples of three CA1 neurons with distinct soma locations along the P-D axis projecting to contralateral CA1 with axonal arbor terminals at mirrored locations. Dots indicate soma. (L) Summary of all CA1-to-CA1 contralateral projecting neurons (n = 89) showing correlation between the soma location along the P-D axis of CA1 and center-of-mass of axon tip locations in contralateral CA1 region along the P-D axis. (M) Bar graphs summarizing correlation coefficients for various downstream areas where the soma location of CA1 neurons along P-D axis significantly correlated with normalized principal component of axon-tip locations. M, median. L, lateral. 12 of 15

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

B MEC

P longitudinal axis

LEC

Dorsal

1 mm

HPFR

THR CNUR

E

MBR HYR 1

CA1 CA2 CA3 DG SUB ProS SUBr

0

Caudal pathway

Ventral 1

F 8

CA1 **** SUB **** ProS ** SUBr n.s.

6

4

2

0

3

CA1 SUB ProS SUBr

4

**** n.s. ** *

5

6

7 2

Downstream target areas

G

Dorsal

Soma in hippocampus

Soma locations in D-V axis (mm)

ISOR OLFR

Longitudinal span (mm)

CTXspC

Pearson correlation

CTXspR

CNUC HPFC

0

Axon tips in ENT

AUD VIS AI RSP TEa ECT PIR TR CA DG ENT PAR POST PRE SUB ProS HATA EP BLA BMA PA sAMY ACA PL ILA ORB TT DP CA ENT PAR PRE SUB CLA STRd STRv LSX sAMY PALv PALm PALc DORsm PVZ PVR MEZ LZ MBmot

Soma locations

OLFC

r = 0.80 p < 0.0001

0

Ventral

Soma in hippocampus

ISOC

1

A V

Soma

D

C

MEC

D

Axon tip locations in ENT (NPC)

A

4

6

8

10

2

Target number

4

6

8

10

Target number

Rostral pathway

Soma

DG

D

V

CA3

D

V D CA2

Neuron number (%) 20 40 60 80

V

CA1

D

V

SUB

D

V

ProS

D

V

SUBr

D

CNU

CTXsp

HPF

OLF

FRP MO ACA PL ILA ORB AI GU MOB AOB AON TT DP PIR NLOT HIP ENT PAR POST PRE SUB ProS HATA APr CLA STRd STRv LSX sAMY PALd PALv PALm PALc DORsm DORpm PVZ PVR MEZ LZ ME MBsen MBmot MBsta

PALd sAMY PA BMA BLA LA EP CLA APr HATA ProS SUB PRE POST PAR ENT HIP TR PAA COA NLOT PIR ECT PERI TEa PTLp RSP VIS AUD VISC SS AI

V

ISO

Fig. 8. Dependence of axonal arbor distribution on the longitudinal location of soma. (A) Examples showing that two CA1 neurons with distinct longitudinal soma locations (dorsal and ventral) exhibited axonal arborization in distinct MEC domains. Black dots indicate soma. (B) Spatial distribution of HIP neuronal soma (left) and corresponding topography of their axonal arbor terminals in LEC and MEC (right). The soma location along the D-V axis and their corresponding axonal arbor terminals are color-coded, with the scale shown in the middle. (C) Correlation between the soma location of HIP neurons along the D-V axis and NPC of axon tip locations in ENT [same color code as in (B); Pearson correlation, r = 0.80, P < 0.0001, n = 3177 neurons]. (D) Heatmap summarizing color-coded correlation coefficients in various downstream areas for soma locations of neurons in seven HIP subregions along the D-V axis. Only significant correlation coefficients are shown. (E) Correlation between the Qiu et al., Science 383, eadj9198 (2024)

2 February 2024

ISO

OLF

HPF

CTXsp

CNU

TH

HY

MB

number of primary targets and the soma distribution along the D-V axis (expressed as longitudinal span) for neurons in CA1, SUB, ProS, and SUBr. Pearson correlations: CA1, r = –0.88, ***P < 0.001; SUB, r = –0.94, ****P < 0.0001; ProS, r = –0.84, **P < 0.01; and SUBr, r = –0.68, P > 0.05. (F) Correlation between the number of primary targets (mean ± SEM) and the soma location along the D-V axis for neurons in CA1, SUB, ProS, and SUBr. Pearson correlations: CA1, r = 0.94, ****P < 0.0001; SUB, r = 0.46, P > 0.05; **ProS, r = 0.78, P < 0.01; and SUBr, r = 0.91, *P < 0.05. (G) Schematic diagram illustrating the number of neurons with longitudinal soma locations (eight bins in the D-V axis) in seven HIP subregions projecting to various downstream areas within primary targets: ipsilateral and contralateral HPF, ISO, OLF, CTXsp, CNU, TH, HY, and MB (see full names of all abbreviations in table S3). For each downstream target, the number of HIP neurons in each bin was normalized by the maximal neuron number projecting to that target. 13 of 15

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

jointly contribute to fast memory generalization (39). Although PRE neurons are known to project bilaterally to the ENT (53), we found that these ENT-projection neurons did not send collaterals outside the HPF. Such dedicated intra-HPF projection may be crucial for the hippo-entorhinal communication that is of particular importance for spatial navigation (40, 41). Because LEC could modulate learninginduced gamma synchrony in multiple cortices (54), the coordinated intra-HPF projection to LEC and widespread extra-HPF projections may provide a basis for exerting an efficient entrainment of brain-wide oscillatory activity. HIP neurons in the dorsal and ventral areas are known to have distinct anatomical and physiological functions (9, 10, 36, 55). We identified longitudinal subdomains that may represent the anatomical basis underlying distinct physiological functions of HIP neurons located at various D-V locations, as exemplified by the finding of NAc- and mPFC-projecting and NAc- and BA-projecting neurons involved in emotional processing for valence versus salience, respectively (35). With the availability of intersectional subtype-specific probes (35), neuronal activity within these subdomains could be monitored and manipulated, allowing us to determine how various projectome subtypes perform their specific physiological functions. Our finding that CA3 axons project longitudinally in the CA3 region and then transversely in CA1 region provides further details of axon patterning in both longitudinal and transverse directions. The extensive longitudinal and cross-hemispheric projections inside the HIP suggest a fundamental role of these projections in HIP information processing. In conclusion, our results provide a comprehensive database with which to elucidate organization principles for HIP axon projections. Our single-neuron projectome analyses have revealed the complexity and inherent order of brain-wide axon collaterals. These analyses helped to define HIP subtypes that are based on axon or dendrite morphology, axon arborization at target areas, as well as the soma location along the HIP axes. Such knowledge could be further combined with gene expression data, preferably at singlecell resolution, to pave the way for future functional dissection of neural circuits involving the HIP. Methods summary Single-cell projectome data acquisition and analysis

Spatial transcriptomic data acquisition and analysis

Brain sections containing HIP were laid on Stereo-seq chips for mRNA collection followed by gene sequencing. The spatial matrix of gene expressions in CA1 was binned with 50 × 50 DNA nanoballs (25 µm diameter) by aggregating the transcripts of the same gene within each bin. The Seurat (v.4.0.2) and Tidyverse (v.1.3.1) packages in R (v.4.1.2) were used for basic data processing and visualization. Cell clusters and subclusters were classified by the “FindClusters” function using the shared nearest neighbor modularity optimization with a clustering resolution of 0.1. The marker genes in each cluster or subcluster were identified by differential expression analysis using “FindAllMarkers” function. The correlation between numbers of transcriptomeidentified subclusters and projectome-defined subtypes were determined by the Spearman correlation coefficient. RE FERENCES AND NOTES

Male C57BL/6J mice (8 to 10 weeks of age) were group housed before surgeries. Animals were anesthetized by isoflurane and AAV was injected into a variety of sites in HIP to cover the whole range as much as possible. The fMOST imaging was prepared as previously described (26, 28). The EGFP channel was used for neurite tracing, and the PI channel was used for image Qiu et al., Science 383, eadj9198 (2024)

registration. Neurons were reconstructed by merging the tracing from two independent human tracers and validated by a third human tracer. All validated neurons were registered to the Allen Mouse Brain Common Coordinate Framework (CCFv3). Further analysis was performed by customized Python scripts. The primary target areas of HIP neurons were relatively large brain areas including ISO, OLF, HPF, CTXsp, CNU, TH, HY, and MB. To determine that a single HIP neuron has significant axon arborizations within a primary target, the total length of its axon arbors in this primary target was at least 1 mm long and had at least one terminal point and one branch point. The projection strength in a target area can be represented by either the number of axonal terminals or the total length of axon arbors. Axon morphologies were reconstructed by knowledge-based reference templates of axon trajectories. All single-neurons projectomes in our database were classified into 341 target patterns and then further clustered into 43 subtypes. The longitudinal axis for HIP was obtained to quantify the soma location along the D-V axis. When considering the selectivity of axon projections in two target areas, the lengths of axon arbors in two areas were denoted as A and B. The selectivity index for each neuron was then calculated by (A – B)/(A + B).

1. L. Luo, Architectures of neuronal circuits. Science 373, eabg7285 (2021). doi: 10.1126/science.abg7285; pmid: 34516844 2. P. Andersen, R. Morris, D. Amaral, T. Bliss, J. O’Keefe, The Hippocampus Book (Oxford Univ. Press, 2009). 3. W. B. Scoville, B. Milner, Loss of recent memory after bilateral hippocampal lesions. J. Neurol. Neurosurg. Psychiatry 20, 11–21 (1957). doi: 10.1136/jnnp.20.1.11; pmid: 13406589 4. J. O’Keefe, L. Nadel, The Hippocampus as a Cognitive Map (Oxford Univ. Press, 1978).

2 February 2024

5. P. G. Henke, Hippocampal pathway to the amygdala and stress ulcer development. Brain Res. Bull. 25, 691–695 (1990). doi: 10.1016/0361-9230(90)90044-Z; pmid: 2289157 6. N. Tottenham, M. A. Sheridan, A review of adversity, the amygdala and the hippocampus: A consideration of developmental timing. Front. Hum. Neurosci. 3, 68 (2010). doi: 10.3389/neuro.09.068.2009; pmid: 20161700 7. P. Andersen, T. V. Bliss, K. K. Skrede, Lamellar organization of hippocampal excitatory pathways. Exp. Brain Res. 13, 222–238 (1971). doi: 10.1007/BF00234087; pmid: 5570425 8. J. P. Aggleton, K. Christiansen, The subiculum: The heart of the extended hippocampal system. Prog. Brain Res. 219, 65–82 (2015). doi: 10.1016/bs.pbr.2015.03.003; pmid: 26072234 9. M. S. Fanselow, H. W. Dong, Are the dorsal and ventral hippocampus functionally distinct structures? Neuron 65, 7–19 (2010). doi: 10.1016/j.neuron.2009.11.031; pmid: 20152109 10. B. A. Strange, M. P. Witter, E. S. Lein, E. I. Moser, Functional organization of the hippocampal longitudinal axis. Nat. Rev. Neurosci. 15, 655–669 (2014). doi: 10.1038/nrn3785; pmid: 25234264 11. L. W. Swanson, W. M. Cowan, An autoradiographic study of the organization of the efferent connections of the hippocampal formation in the rat. J. Comp. Neurol. 172, 49–84 (1977). doi: 10.1002/cne.901720104; pmid: 65364 12. A. Pitkänen, M. Pikkarainen, N. Nurminen, A. Ylinen, Reciprocal connections between the amygdala and the hippocampal formation, perirhinal cortex, and postrhinal cortex in rat. A review. Ann. N. Y. Acad. Sci. 911, 369–391 (2000). doi: 10.1111/ j.1749-6632.2000.tb06738.x; pmid: 10911886 13. M. S. Bienkowski et al., Integration of gene expression and brain-wide connectivity reveals the multiscale organization of mouse hippocampal networks. Nat. Neurosci. 21, 1628–1643 (2018). doi: 10.1038/s41593-018-0241-y; pmid: 30297807 14. M. S. Cembrowski et al., Dissociable structural and functional hippocampal outputs via distinct subiculum cell classes. Cell 173, 1280–1292.e18 (2018). doi: 10.1016/j.cell.2018.03.031; pmid: 29681453 15. H. Lee, C. Wang, S. S. Deshmukh, J. J. Knierim, Neural population evidence of functional heterogeneity along the CA3 transverse axis: Pattern completion versus pattern separation. Neuron 87, 1093–1105 (2015). doi: 10.1016/ j.neuron.2015.07.012; pmid: 26298276 16. M. S. Cembrowski, N. Spruston, Heterogeneity within classical cell types is the rule: Lessons from hippocampal pyramidal neurons. Nat. Rev. Neurosci. 20, 193–204 (2019). doi: 10.1038/ s41583-019-0125-5; pmid: 30778192 17. Y. Li et al., A distinct entorhinal cortex to hippocampal CA1 direct circuit for olfactory associative learning. Nat. Neurosci. 20, 559–570 (2017). doi: 10.1038/nn.4517; pmid: 28263300 18. J. C. Jimenez et al., Anxiety cells in a hippocampalhypothalamic circuit. Neuron 97, 670–683.e6 (2018). doi: 10.1016/j.neuron.2018.01.016; pmid: 29397273 19. S. H. Lee et al., Parvalbumin-positive basket cells differentiate among hippocampal pyramidal cells. Neuron 82, 1129–1144 (2014). doi: 10.1016/j.neuron.2014.03.034; pmid: 24836505 20. K. Mizuseki, K. Diba, E. Pastalkova, G. Buzsáki, Hippocampal CA1 pyramidal cells form functionally distinct sublayers. Nat. Neurosci. 14, 1174–1181 (2011). doi: 10.1038/nn.2894; pmid: 21822270 21. J. Winnubst et al., Reconstruction of 1,000 projection neurons reveals new cell types and organization of long-range connectivity in the mouse brain. Cell 179, 268–281.e13 (2019). doi: 10.1016/ j.cell.2019.07.042; pmid: 31495573 22. M. M. Gergues et al., Circuit and molecular architecture of a ventral hippocampal network. Nat. Neurosci. 23, 1444–1452 (2020). doi: 10.1038/s41593-020-0705-8; pmid: 32929245 23. Y. Han et al., The logic of single-cell projections from visual cortex. Nature 556, 51–56 (2018). doi: 10.1038/nature26159; pmid: 29590093 24. J. Ren et al., Single-cell transcriptomes and whole-brain projections of serotonin neurons in the mouse dorsal and median raphe nuclei. eLife 8, e49424 (2019). doi: 10.7554/ eLife.49424; pmid: 31647409 25. R. Muñoz-Castañeda et al., Cellular anatomy of the mouse primary motor cortex. Nature 598, 159–166 (2021). doi: 10.1038/s41586-021-03970-w; pmid: 34616071 26. H. Gong et al., High-throughput dual-colour precision imaging for brain-wide connectome with cytoarchitectonic landmarks at the cellular level. Nat. Commun. 7, 12142 (2016). doi: 10.1038/ ncomms12142; pmid: 27374071 27. Q. Zhong et al., High-definition imaging using line-illumination modulation microscopy. Nat. Methods 18, 309–315 (2021). doi: 10.1038/s41592-021-01074-x; pmid: 33649587

14 of 15

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

28. L. Gao et al., Single-neuron projectome of mouse prefrontal cortex. Nat. Neurosci. 25, 515–529 (2022). doi: 10.1038/ s41593-022-01041-5; pmid: 35361973 29. Q. Wang et al., The Allen Mouse Brain Common Coordinate Framework: A 3D Reference Atlas. Cell 181, 936–953.e20 (2020). doi: 10.1016/j.cell.2020.04.007; pmid: 32386544 30. L. A. Cenquizca, L. W. Swanson, Spatial organization of direct hippocampal field CA1 axonal projections to the rest of the cerebral cortex. Brain Res. Rev. 56, 1–26 (2007). doi: 10.1016/ j.brainresrev.2007.05.002; pmid: 17559940 31. C. Xu et al., Distinct hippocampal pathways mediate dissociable roles of context in memory retrieval. Cell 167, 961–972.e16 (2016). doi: 10.1016/j.cell.2016.09.051; pmid: 27773481 32. Y. Zhou et al., A ventral CA1 to nucleus accumbens core engram circuit mediates conditioned place preference for cocaine. Nat. Neurosci. 22, 1986–1999 (2019). doi: 10.1038/s41593-0190524-y; pmid: 31719672 33. S. Ciocchi, J. Passecker, H. Malagon-Vina, N. Mikus, T. Klausberger, Selective information routing by ventral hippocampal CA1 projection neurons. Science 348, 560–563 (2015). doi: 10.1126/science.aaa3245; pmid: 25931556 34. J. Jin, S. Maren, Fear renewal preferentially activates ventral hippocampal neurons projecting to both amygdala and prefrontal cortex in rats. Sci. Rep. 5, 8388 (2015). doi: 10.1038/srep08388; pmid: 25669753 35. H. S. Chen et al., An intein-split transactivator for intersectional neural imaging and optogenetic manipulation. Nat. Commun. 13, 3605 (2022). doi: 10.1038/s41467-022-31255-x; pmid: 35739125 36. H. W. Dong, L. W. Swanson, L. Chen, M. S. Fanselow, A. W. Toga, Genomic-anatomic evidence for distinct functional domains in hippocampal field CA1. Proc. Natl. Acad. Sci. U.S.A. 106, 11794–11799 (2009). doi: 10.1073/pnas.0812608106; pmid: 19561297 37. A. Chen et al., Spatiotemporal transcriptomic atlas of mouse organogenesis using DNA nanoball-patterned arrays. Cell 185, 1777–1792.e21 (2022). doi: 10.1016/j.cell.2022.04.003; pmid: 35512705 38. N. Ishizuka, J. Weber, D. G. Amaral, Organization of intrahippocampal projections originating from CA3 pyramidal cells in the rat. J. Comp. Neurol. 295, 580–623 (1990). doi: 10.1002/cne.902950407; pmid: 2358523 39. H. Zhou et al., The interhemispheric CA1 circuit governs rapid generalisation but not fear memory. Nat. Commun. 8, 2190 (2017). doi: 10.1038/s41467-017-02315-4; pmid: 29259187 40. J. S. Taube, The head direction signal: Origins and sensorymotor integration. Annu. Rev. Neurosci. 30, 181–207 (2007). doi: 10.1146/annurev.neuro.29.051605.112854; pmid: 17341158 41. E. I. Moser et al., Grid cells and cortical representation. Nat. Rev. Neurosci. 15, 466–481 (2014). doi: 10.1038/nrn3766; pmid: 24917300 42. D. G. Amaral, M. P. Witter, The three-dimensional organization of the hippocampal formation: A review of anatomical data. Neuroscience 31, 571–591 (1989). doi: 10.1016/03064522(89)90424-7; pmid: 2687721 43. M. P. Witter, Intrinsic and extrinsic wiring of CA3: Indications for connectional heterogeneity. Learn. Mem. 14, 705–713 (2007). doi: 10.1101/lm.725207; pmid: 18007015

Qiu et al., Science 383, eadj9198 (2024)

44. X. G. Li, P. Somogyi, A. Ylinen, G. Buzsáki, The hippocampal CA3 network: An in vivo intracellular labeling study. J. Comp. Neurol. 339, 181–208 (1994). doi: 10.1002/cne.903390204; pmid: 8300905 45. H. E. Scharfman, The enigmatic mossy cell of the dentate gyrus. Nat. Rev. Neurosci. 17, 562–575 (2016). doi: 10.1038/ nrn.2016.87; pmid: 27466143 46. D. GoodSmith et al., Spatial representations of granule cells and mossy cells of the dentate gyrus. Neuron 93, 677–690.e5 (2017). doi: 10.1016/j.neuron.2016.12.026; pmid: 28132828 47. M. He et al., Strategies and tools for combinatorial targeting of GABAergic neurons in mouse cerebral cortex. Neuron 91, 1228–1243 (2016). doi: 10.1016/j.neuron.2016.08.021; pmid: 27618674 48. L. Acsády, T. J. Görcs, T. F. Freund, Different populations of vasoactive intestinal polypeptide-immunoreactive interneurons are specialized to control pyramidal cells or interneurons in the hippocampus. Neuroscience 73, 317–334 (1996). doi: 10.1016/ 0306-4522(95)00609-5; pmid: 8783252 49. L. Acsády, M. Pascual, N. Rocamora, E. Soriano, T. F. Freund, Nerve growth factor but not neurotrophin-3 is synthesized by hippocampal GABAergic neurons that project to the medial septum. Neuroscience 98, 23–31 (2000). doi: 10.1016/ S0306-4522(00)00091-9; pmid: 10858608 50. R. Francavilla et al., Connectivity and network state-dependent recruitment of long-range VIP-GABAergic neurons in the mouse hippocampus. Nat. Commun. 9, 5043 (2018). doi: 10.1038/s41467-018-07162-5; pmid: 30487571 51. T. Klausberger, P. Somogyi, Neuronal diversity and temporal dynamics: The unity of hippocampal circuit operations. Science 321, 53–57 (2008). doi: 10.1126/science.1149381; pmid: 18599766 52. G. G. Szabo et al., Ripple-selective GABAergic projection cells in the hippocampus. Neuron 110, 1959–1977.e9 (2022). doi: 10.1016/j.neuron.2022.04.002; pmid: 35489331 53. M. Caballero-Bleda, M. P. Witter, Projections from the presubiculum and the parasubiculum to morphologically characterized entorhinal-hippocampal projection neurons in the rat. Exp. Brain Res. 101, 93–108 (1994). doi: 10.1007/ BF00243220; pmid: 7843307 54. W. Luo et al., Acquiring new memories in neocortex of hippocampal-lesioned mice. Nat. Commun. 13, 1601 (2022). doi: 10.1038/s41467-022-29208-5; pmid: 35332120 55. R. Lorente De Nó, Studies on the structure of the cerebral cortex. II. Continuation of the study of the ammonic system. J. Psychol. Neurol. 46, 113–177 (1934). 56. S. Qiu et al., Data for: Mouse hippocampal single-neuron projectome and spatial transcriptome data analysis and visualization, Brain Science Data Center, Chinese Academy of Sciences (2023); https://doi.org/10.12412/BSDC.1697181607.20002. 57. S. Qiu et al., Code for: Mouse hippocampal single-neuron projectome and spatial transcriptome data analysis and visualization, Dryad (2023); https://doi.org/10.5061/dryad.wh70rxwv4.

computing center at CEBSIT, Chinese Academy of Science, for data analysis; and the imaging facility and animal facility at CEBSIT, Chinese Academy of Sciences, for technical support. Funding: This work was supported by the National Key R&D Program of China (grant 2020YFE0205900 to C.X. and grant 2019YFA0709504 to C.L.); the Shanghai Municipal Science and Technology Major Project (grant 2018SHZDZX05 to C.X.); STI2030-Major Projects (grant 2022ZD0205000 to C.X., grant 2021ZD0204404 to Y.S., grant 2021ZD0201000 to H.G., and grant 2021ZD0203601 to C.L.); the Strategic Priority Research Program of the Chinese Academy of Sciences (grant XDB32010105 to C.X.); The National Natural Science Foundation of China (grants 31771180 and 91732106 to C.X., grant 32221003 to Y.S. and C.L., grant 61890953 to H.G., and grants 31827803 and 32161133024 to C.L.); the Shanghai Science and Technology Committee (grant 2019-78677 to C.X.). Author contributions: Conceptualization: C.X., S.Q.; Data curation: D.W., Y.C., X.W.; Data interpretation: C.X., M.P., L.X.; Formal analysis (neurite tracing): B.R., X.S., Y.C., X.W., L.H., Y.L., J.Y.; Formal analysis (visualization): S.Q., Y.H., T.G., X.S., T.X., Y.H., Y.S.; Formal analysis (website): L.Q., C.J., J.H., W.D.; Investigation (fMOST imaging): H.G., Q.L., A.L., T.J., X.L., X.J., J.Y.; Investigation (virus injections): L.D., L.Q., D.L., C.Z., Z.L., C.T., H.Y., W.Q.; Project administration: D.W., Y.C., X.W.; Software: J.Y.; Supervision: C.X., Y.S.; Writing – original draft: C.X., Q.S.; Writing – review & editing: C.X., M.P., L.X., Y.S., S.Q. Competing interests: The authors declare no competing interests. Data and materials availability: The mouse HIP-related data visualization can be navigated at https://mouse.digital-brain.cn/hipp. All of the reconstructed neurons and an atlas with color-coded longitudinal coordinates can be interactively searched, visualized, and downloaded at https://mouse.digital-brain.cn/projectome/hipp. The spatial transcriptome data in CA1 can be interactively queried and visualized at https://mouse.digital-brain.cn/spatial-omics/CA1 and downloaded from Brain Science Data Center (https:// braindatacenter.cn) of the Chinese Academy of Sciences (56). The transcriptomic cluster and subclusters and related scripts for this study are available on Dryad (57). The FNT software can be downloaded from https://zenodo.org/record/5981001. Customwritten codes used in this study are available upon reasonable request from the corresponding authors. 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

ACKN OWLED GMEN TS

science.org/doi/10.1126/science.adj9198 Materials and Methods Figs. S1 to S12 Tables S1 to S4 References (58–62) MDAR Reproducibility Checklist

We thank G. Buzsáki and Z. Nusser for valuable comments on early versions of the manuscript; W. Li and J. Wei at the data and

Submitted 27 July 2023; accepted 19 December 2023 10.1126/science.adj9198

2 February 2024

15 of 15

RES EARCH

RESEARCH ARTICLE SUMMARY

◥

NEUROSCIENCE

Blood pressure pulsations modulate central neuronal activity via mechanosensitive ion channels Luna Jammal Salameh*, Sebastian H. Bitzenhofer, Ileana L. Hanganu-Opatz, Mathias Dutschmann, Veronica Egger*

INTRODUCTION: Neural electrical oscillations are considered fundamental to how the brain processes information. Various modes of oscillation reflect processing in local or brain-wide networks and occur spontaneously or are associated with sensory and cognitive processing. Accumulating evidence suggests that such neural oscillations can also be modulated by the interoception of body rhythms, such as respiration or the heartbeat. Interoception is the sensing of internal body signals—as opposed to the sensory perception of the outer world— and thus informs the brain about the state of the organism. RATIONALE: To investigate the mechanisms of local oscillations within a restricted network, we had developed a semi-intact preparation of the rat olfactory bulb—the first station of olfactory processing in the brain, noted for its strong oscillatory activity. In this type of reduced prep-

Heart

aration, there is no heart, lung, or input from other brain areas, and the vasculature of the bulb is perfused with artificial blood by a peristaltic pump. This pump generates pressure pulsations within the cerebral vascular system that, in our setup, happen to fit within the physiological range of heartbeat-induced pulsations of intracranial pressure in vivo. Notably, these pump-induced mechanical pulsations were precisely followed by local electrical field oscillations that originated from mitral cells, the principal neurons of the olfactory bulb. On the basis of recent evidence for the expression of mechanosensitive ion channels in subsets of principal neurons across the brain, we then hypothesized that these neurons may be capable of directly sensing the vascular blood pressure pulsations associated with the heartbeat.

CONCLUSION: The role of interoception in RESULTS: The pump-induced pressure pulsa-

tions provided an adequate stimulus for a

Brain

Mitral cell

Mitral cell

Mechanoreceptors

∆p Olfactory bulb ECG

Vascular pressure pulsation ∆p

Modulation mitral cell Vm

Ensemble spiking activity

50 ms

How neurons can feel the pulse within the brain. (Left) Rat heart (top) and schematic rat electrocardiogram (ECG) (bottom). (Middle) Schematic rat brain and coronal section through the olfactory bulb and its blood vessels with exemplary mitral cell (top) and intracranial pressure pulsations (Dp) caused by the heartbeat (bottom). (Right) Mitral cell with mechanoreceptors (top); weak excitatory modulation of the mitral cell membrane potential Vm by mechanosensitive ion channel currents, with the ion channels (most likely Piezo2) gated by deflections of the mitral cell membrane caused by the pulsations Dp (middle); and the ensuing subtle modulation of the timing of spontaneous spikes in an ensemble of mitral cells (bottom). [Created in part with www.biorender.com] Jammal Salameh et al., Science 383, 494 (2024)

mechanosensory transduction pathway within mitral cells. This transduction was mediated by fast excitatory mechanosensitive ion channels— most likely Piezo2—that are present in a subset of mitral cells. In many other body tissues, Piezo2 has been shown to contribute to the detection of vibrations within similar-frequency regimes. Its gating properties could well underlie the transformation of the sinusoidal waveform of the pressure stimulus into the more complex waveform of the local field oscillations observed in our preparation. Although this fast transduction pathway did not involve synaptic transmission, the vascular pressure pulsation rhythm entrained the spontaneous spiking activity of the mitral cells. Thus, the mechanosensory transduction exerted a direct modulatory influence on spike timing. Can this pathway allow the brain to sense the heartbeat in vivo? In awake mice, we found that neuronal spiking activity was in fact modulated by the heartbeat, with ~15% of olfactory bulb neurons being entrained by this rhythm, mostly within 20 ms. This effect was considerably weaker than the known coupling of neuronal activity to the respiration rhythm, which explains why it has not been observed until now. We observed similar heartbeat-induced modulations of neuronal activity also in the hippocampus and prefrontal cortex.

2 February 2024

brain function is one of the major challenges in current neuroscience. In humans, recent experimental evidence supports the modulation of autonomous and conscious perception and cognition by the cardiac cycle. Although this modulation is partially mediated by the classical ascending multisynaptic pathway originating from aortic baroreceptors, the present results reveal that heartbeat-induced pulsations of cerebral blood vessels can directly affect central neuronal activity through the activation of mechanosensitive channels. Although currently the function of this immediate pathway is a matter of speculation, we propose that a brain-wide network of “heartbeat sentinel neurons” mediates interoceptive modulation of cognition, mood, and autonomic status. For example, the occurrence of certain states of arousal might correlate with activation of this network. Our finding adds a fast transmission line to the interoceptive bodybrain axis, whereby central neurons can feel the pulse within the brain.

▪

The list of author affiliations is available in the full article online. *Corresponding author. Email: [email protected] (L.J.S.); [email protected] (V.E.) Cite this article as L. Jammal Salameh et al., Science 383, eadk8511 (2024). DOI: 10.1126/science.adk8511

READ THE FULL ARTICLE AT https://doi.org/10.1126/science.adk8511 1 of 1

RES EARCH

RESEARCH ARTICLE

◥

NEUROSCIENCE

Blood pressure pulsations modulate central neuronal activity via mechanosensitive ion channels Luna Jammal Salameh1*, Sebastian H. Bitzenhofer2, Ileana L. Hanganu-Opatz2, Mathias Dutschmann3, Veronica Egger1* The transmission of the heartbeat through the cerebral vascular system causes intracranial pressure pulsations. We discovered that arterial pressure pulsations can directly modulate central neuronal activity. In a semi-intact rat brain preparation, vascular pressure pulsations elicited correlated local field oscillations in the olfactory bulb mitral cell layer. These oscillations did not require synaptic transmission but reflected baroreceptive transduction in mitral cells. This transduction was mediated by a fast excitatory mechanosensitive ion channel and modulated neuronal spiking activity. In awake animals, the heartbeat entrained the activity of a subset of olfactory bulb neurons within ~20 milliseconds. Thus, we propose that this fast, intrinsic interoceptive mechanism can modulate perception—for example, during arousal—within the olfactory bulb and possibly across various other brain areas.

B

oth spontaneous and sensory-evoked neuronal network oscillations are a hallmark of olfactory systems across phylae (1). To uncover the origins and underlying mechanisms of these oscillations within the olfactory bulb (OB), we have developed a semiintact preparation of the rat olfactory system [nose-brain preparation (NBP)] (2). In this type of preparation (3), the tissue of interest is perfused with oxygenated artificial cerebrospinal fluid (ACSF) by a peristaltic pump (Fig. 1A). Initially, the perfusion pressure is adjusted by increasing the pump frequency until rhythmic bursting of the phrenic nerve activity (PNA) resumes, which indicates sufficient brain tissue oxygenation. The pump generates pressure pulsations within the cerebral vascular system that are likely to fit within the physiological range of heartbeat-induced pulsations of the intracranial and cerebral vascular pressure in vivo (4, 5). We have previously reported that in the NBP, spontaneous slow oscillations are detectable in the OB local field potential (LFP), in the absence of respiration and ascending inputs (2). We aimed to reveal the origin of these oscillations and uncovered a correlation with the pump-induced pressure pulsations. On the basis of the finding that subsets of principal neurons within the OB and other brain regions express fast mechanosensitive ion channels (6), we generated the hypothesis that specific neurons of

1

Neurophysiology Group, Zoological Institute, Regensburg University, 93040 Regensburg, Germany. 2Institute of Developmental Neurophysiology, Center for Molecular Neurobiology, Hamburg Center of Neuroscience, University Medical Center Hamburg-Eppendorf, 20251 Hamburg, Germany. 3Division of Pulmonary, Critical Care, and Sleep Medicine, Department of Medicine, Case Western Reserve University, Cleveland, OH 44106, USA. *Corresponding author. Email: [email protected] (L.J.S.); [email protected] (V.E.)

Jammal Salameh et al., Science 383, eadk8511 (2024)

the brain may have the capacity for direct interoception of heartbeat-induced pressure pulsations. Interoception plays an important yet so far not-fully-known role in providing rhythmic sensory feedback on bodily activity that in turn modulates neuronal activity (7). For instance, the respiratory modulation of neural activity within the olfactory system during odor perception and at resting states (8–11) is linked to mechanoreceptors of the nasal mucosa that sense nasal airflow (12, 13). Another example is cardiovascular activity that provides interoceptive feedback via aortic, carotid sinus, and cardiac baroreceptors that sense blood pressure and heartbeat (14). Such ascending modulations of forebrain functions are associated with cortical heartbeat-evoked potentials (HEPs) that, in humans, occur at the earliest ~300 ms after the R peak of the electrocardiogram (ECG) (15), which is in line with the known multisynaptic ascending modulatory pathway via the nucleus tractus solitarius (NTS) as the underlying mechanism for interoception (16). We propose an additional faster pathway for heartbeat interoception. Slow LFP oscillations in the NBP are associated with pressure pulsations induced by the peristaltic pump

LFP oscillations within a frequency range of 1.5 to 4 Hz (n = 82 NBPs; fig. S1A) were recorded at the depth of the mitral cell layer (MCL) (Fig. 1B and Materials and methods) and were observed in most NBP preparations that showed regular PNA (mean burst frequency in the 82 NBPs was 14.5 ± 5.2 bursts per minute; fig. S1A). In contrast to previous in vivo findings (17, 18), MCL oscillations in the NBP were not coupled to the respiratory theta rhythm because of the absence of nasal 2 February 2024

airflow and the absence of ascending respiratory pathways due to decerebration. Thus, the perfused OB was devoid of any respiratory modulation, even though rhythmic PNA was generated by the brainstem and served as an important readout for the viability of the NBP (2, 19) (Fig. 1C and Fig. 2A). Spectral analysis further revealed harmonics of this fundamental rhythm (P0) in 53 of the 83 analyzed experiments, with on average 2.0 ± 1.0 detectable harmonic peaks (P1, P2, etc.; Fig. 1D). The individual frequency peaks were very narrowly tuned [Fig. 1D and fig. S1A; P0 average full width at half maximum (FWHM) = 5.5 ± 3.6 mHz]. Because slow LFP oscillations, such as hippocampal theta and respiration-related OB oscillations, show far larger spectral widths (20–22), we investigated whether the slow oscillations were related to the perfusion system. Although at optimal settings of perfusion pressure, the pump was operated well below the LFP frequency range (rotation frequency < 1 Hz), we observed that the periodic LFP signal was entrained to the pressure pulsations induced by the eight rollers of the peristaltic pump (fig. S1, C and D). The broad distribution of the P0 frequency (fig. S1A) was thus a result of the initial adjustment of the pump frequency (see Materials and methods). To establish the correlation between the LFP oscillations and perfusion pressure pulsations, we recorded OB-LFPs and perfusion pressure simultaneously (Materials and methods and Fig. 1A). Perfusion pressure recordings displayed pulsations with an amplitude of ~2 to 4 DmmHg on top of the static pressure established by the pump (~40 to 80 mmHg at the entry point to the aorta; Fig. 1E). The pulsations matched the fundamental frequency of the slow LFP oscillation (mean ratio of frequencies, 1.0 ± 0.0; n = 13 NBPs). Inserting an extra Windkessel reservoir into the perfusion circuit to dampen pulsations reduced the pulsation amplitude to ~25% of its initial value and resulted in complete cessation of the slow LFP oscillation (Fig. 1F; n = 3 NBPs). Thus, pressure pulsations generated oscillatory LFPs. Pulsation-induced slow LFP oscillations are neurogenic and depend on a nonsynaptic baroreceptive mechanism

To exclude a pulsation artifact resulting from pulsatile arterioles close to the recording electrode, we tested for the neurogenic origin of the oscillatory LFP by interfering with the viability of the tissue. We investigated the impact of hypoxia on slow LFP oscillations using two types of manipulation (n = 12 NBPs in total). We recorded baseline LFP with the standard oxygenated ACSF (95% O2, 5% CO2) for at least 10 min before we either lowered pump frequency from 15 to 30 rotations per minute (rpm) down to 3 rpm (0.05 Hz, n = 7) or used nonoxygenated ACSF (95% N2, 5% CO2) instead 1 of 12

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

A

C

E

B

D

F

Fig. 1. Slow oscillations in the OB-LFP in a semi-intact perfused brain preparation due to pressure pulsations. (A) The decerebrated rat preparation is perfused via the cannulated descending aorta. Spontaneous OB-LFPs and PNA are recorded in parallel. Created with www.biorender.com. (B) Recording location of the LFP within the OB. (Left) Staining of recording electrode location (Materials and methods). (Right) OB-LFP activity is highest in the MCL. (C) Exemplary recording of rhythmic PNA and OB-LFP within the MCL. (Bottom) Low-pass filtering of the LFP reveals a slow oscillatory component. (D) Lower end of the Fourier spectrum of the above OB-LFP recording. Note the harmonic

(n = 5) to evoke progressive hypoxia for 10 min until we returned to the initial pump frequency or oxygenated ACSF, respectively. Parallel PNA recordings were used to monitor the progression of hypoxia. In all experiments, the PNA displayed a characteristic biphasic response to the hypoxic condition (23) that finally resulted in the cessation of PNA [Fig. 2A; baseline PNA = 15.7 ± 6.6 beats per min (bpm), hypoxic PNA = 0.2 ± 0.5 bpm]. The LFP oscillations in the OB were virtually abolished in all experiments (Fig. 2A). Thus, the LFP oscillations must be of neuronal origin because they depend on proper oxygenation of the OB tissue. A contribution of glial cells cannot be excluded at this point (but see Discussion). We further investigated the origin of OB-LFPs through the analysis of harmonics. Harmonics are a common hallmark of neuronal network oscillations (24). LFP waveforms related to theta oscillations often resemble a slanted, sawtooth-like pattern because of the asymmetric time courses of the underlying mechanisms, mainly the faster rise and slower decay of postsynaptic currents (24, 25). The slow LFP waveforms in NBPs also showed a characteristic pattern, reflected in their harmonics (Fig. 2B). The average power amplitude of harmonic Jammal Salameh et al., Science 383, eadk8511 (2024)

peaks P1 and P2 at multiples of the fundamental P0 and the narrow peak width, as evident from the magnified insets. (E) Parallel recordings of perfusion pressure and LFP oscillations. (Left) Individual example experiment, overlay of pressure and OB-LFP recording FFTs (scaled for matching P0 power amplitudes). (Right) Cumulative data of P0 frequency ratio LFP/pressure. (F) (Top left) Individual experiment with baseline pressure recording (yellow) and added Windkessel effect (gray). (Bottom left) Respective FFTs of the OB-LFP recorded in baseline (blue) and with additional Windkessel (+Wk) (gray). (Right) Cumulative data of change in (P0 + P1) power amplitude.

peaks relative to the next lower peak was 56 ± 67% for P1/P0 and 77 ± 42% for P2/P1 (Fig. 2C; n = 37 NBPs). The average fast Fourier transform (FFT) based on these ratios can be inverted to reconstruct the average slow oscillatory component of the LFP (Fig. 2D, top). This LFP waveform clearly differed from the sinusoidal pressure pulsation waveform of the stimulus (fig. S2, A and B, and Fig. 2D, bottom). Still, the waveform showed a notable mirror symmetry in time (see the filtered examples in Fig. 2B). Because of this symmetry, the involvement of synaptic network interactions is highly unlikely. In contrast to the abovementioned signature of theta oscillations, the presence of harmonics in this case excludes synaptic network activity as a source of the oscillatory waveform. These observations, together with the narrow width of the FFT peaks, implied that a direct baroreceptive transduction mechanism was the most likely explanation for these slow MCL-LFP oscillations. Slow baroreceptive LFP oscillations in the NBP are robust and localized to the MCL

We performed control experiments to investigate the robustness of OB oscillations in the NBP by recording MCL-LFPs from the same 2 February 2024

location and with the same pump settings for more than 60 min (n = 10 NBPs). We then analyzed the LFP signals and their FFTs within 10 consecutive blocks of 6-min recordings. Although the oscillatory P0 power amplitude was variable across blocks [mean coefficient of variation (CV) across experiments = 0.38 ± 0.21], no systematic overall rundown in the baroreceptive response to pulsatile stimulation was detected (Fig. 3A). Only the spontaneous LFP activity, which was also variable (mean CV = 0.20 ± 0.15), slightly declined over time (fig. S3A and Materials and methods). We also investigated whether the pulsationinduced oscillations have a specific source within the OB. We recorded MCL-LFPs sequentially from up to nine spots arranged in a grid (Fig. 3B) on the dorsal surface of the OB (5 ± 2 spots per preparation, n = 17 NBPs). Again, although the variability in P0 power was high (mean CV = 0.60 ± 0.31), no significant local deviations from the normalized mean were observed. Vertical mapping (Materials and methods) of the localization of the slow LFP oscillations across the glomerular layer (GL), the external plexiform layer (EPL), the MCL, and the granule cell layer (GCL) at the central dorsal OB showed that P0 power was localized 2 of 12

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

A

B

C

Fig. 2. Neurogenic and nonsynaptic origin of LFP oscillations. (A) Hypoxia abolishes LFP oscillations. (Left) Example recording of PNA and FFT of OB-LFP before and after hypoxia induction by nonoxygenated ACSF for 10 min (FFT spectra normalized because recordings were of different duration). (Right) Cumulative data of all hypoxia induction experiments (n = 7: lower pump rate; n = 5: nonoxygenated ACSF). Mean power amplitude in hypoxic conditions 0.12 ± 0.12 of baseline (P = 0.0024, two-tailed Wilcoxon test). No difference in power amplitude values between hypoxia induction protocols (two-tailed

predominantly around the MCL (n = 9 NBPs; Fig. 3C). Other layers of the OB hardly displayed LFP oscillations. Finally, we evaluated the spatiotemporal stability of harmonics. No significant fluctuations in either the occurrence of harmonic peaks P1 and P2 or the P1/P0 power ratio over time (fig. S3B) or across mapped locations (fig. S3C) were observed. Baroreceptive LFP oscillations are generated independently of synaptic inputs and rely on mechanosensitive ion channels

We investigated the potentially baroreceptive mechanism for the generation of OB oscillations. Local ACSF microinjection yielded no significant changes in oscillatory power, estimated spike rate, or LFP activity (n = 6 NBPs; Fig. 4A, table S1, and Materials and methods). Subsequently, we blocked voltage-gated sodium (Nav) channels using local lidocaine microinjections (30 mM). Lidocaine caused significant Jammal Salameh et al., Science 383, eadk8511 (2024)

D

Mann-Whitney test, P = 0.52). (B) Example low-pass LFP recordings with respective FFTs to their right. Time scaled such that one period has the same length for all examples. (C) Cumulative histograms of relative amplitudes of LFP harmonics. (D) (Left) Average power spectra (P0 set at 2.5 Hz, top blue LFP, bottom yellow pressure pulsations) based on the average of the cumulative harmonics (see fig. S2 for cumulative pressure data). (Right) Inverted FFTs of the average spectra, which for the slow LFP amounts to its regeneration in the absence of spectral noise.

reductions in LFP activity and spike rate, whereas the mean oscillatory power was not significantly changed (n = 9 NBPs; Fig. 4B and table S1). Next, we tested whether pressure pulsations might be transduced through mechanosensitive ion channels because Piezo2 channel expression has recently been reported in a subset of mouse mitral cells (MCs) (6, 26). Although there is currently no selective antagonist for Piezo2 available, GsMTx4 (the spider venom Grammostola spatulata mechanotoxin 4) (27, 28) interferes with the gating of various cationic mechanosensitive ion channels, including Piezo1, Piezo2, and transient receptor potential type C (TRPC) channels (29, 30). Local injection of D-GsMTx4 (1.25 mM) resulted in a significant reduction or even complete block of slow oscillations in all experiments but no significant change in spike rate or LFP activity (n = 13 NBPs; Fig. 4C and table S1). The latter indicates that overall network activity was not 2 February 2024

affected, although antagonistic effects of D-GsMTx4 on voltage-gated channels (Nav and Kv) have been reported for higher concentrations [median inhibitory concentration (IC50) in the 10-mM range] (31). The insensitivity of spike rate to D-GsMTx4 instead implies that the oscillatory increase in excitability through mechanosensitive channel activation was subthreshold for spike generation (Discussion). To dissect a potential role of mechanosensitive TRPCs in slow LFP generation, we injected the broadband blocker SKF 96365 (10 mM) (32). Although several members of the TRPC channel family are expressed in MCs (33), SKF 96365 did not have any effect on oscillations, spike rate, or LFP activity (n = 12 NBPs; Fig. 4D and table S1). Piezo channel properties match symmetric LFP waveform

The pharmacological experiments together with the baroreceptive LFP waveform suggest that 3 of 12

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

A

B

Fig. 3. Pulsation-induced LFP oscillatory power over time, horizontal location, and the layers within the OB. Small italicized numbers next to axes in cumulative graphs indicate averaged number of data points recorded at the respective time, location, or depth interval. (A) Recording at MCL for up to 60 min, data binned into 6-min blocks and normalized to the mean P0 power amplitude within experiments. There was large variability within experiments but no systematic rundown (insignificant two-tailed Wilcoxon tests; comparison of data point distributions to 1). (B) Recording at MCL from up to nine spots across the OB surface within

synaptic transmission is not involved and that Piezo channels are a possible substrate for the intrinsic transduction of pressure pulsations in MC membranes. Piezo channels provide the basis for fast, differential mechanosensation in the skin and other body parts (29, 34–36) as well as in the walls of the aorta and carotid sinus, where they contribute to mediation of the baroreflex (37). Although the gating mechanisms of Piezo channels are not yet fully understood (38), many of their characteristics are known by now. Gating is very fast (activation, ≪1 ms) and involves an inactivated state (Piezo1 inactivation, pH > 2.2, PbAc2 reacted with HI to form weak acetic acid (HAc) (Eq. 3), and PbI2 began to precipitate (Eq. 4). If the reactant content was low, FAPbI3 could not form as it dissolved. By keeping pH between 1.9 and 2.2, PbI2 dissolved and converted intoPbIx
ð2þxÞ scaffolds (26). In this regime, the reaction in Eq. 5 dominated the system. When the pH was 3500 parts per million (ppm), and the control sample also exhibited an impurity concentration of ~380 ppm (table S4). However, the aqueous synthesized FAPbI3 reduced the impurity concentration to 230 g in 100 g of H2O, which was two orders of magnitude higher than in organic solvents such as ACN (6.99 g/100 g) and 2-ME (4.69 g/100 g). Thus, the concentration of Ca2+ (as well as that of Na+ and K+) in FAPbI3 microcrystals synthesized from aqueous solution were lower compared with those synthesized from ACN and 2-ME (fig. S6B). The FAPbI3 perovskite synthesized in aqueous solution exhibited a champion purity of 99.996% with an average value of 99.994 ± 0.0015%, surpassing that of those synthesized in organic solvents (99.831% in ACN and 99.815% in 2-ME) (fig. S6C). Additionally, we confirmed the absence of noticeable hydrogen-containing impurities in FAPbI3 microcrystals through the 1H nuclear magnetic resonance (NMR) spectrum (fig. S7). Furthermore, we ascertained the exceptional dryness of the synthesized microcrystals with a trace water content of 101.0 ± 15.0 ppm (table S7). We calculated the materials cost of the aqueous synthesized d-FAPbI3 microcrystals for kilogramscale synthesis. The materials cost of d-FAPbI3 microcrystals was as low as US$0.2259 per gram, which was nearly two orders of magnitude lower Zhu et al., Science 383, 524–531 (2024)

integrated intensity of a (100) peak versus annealing time extracted from the in situ GIWAXS results. (D and E) 2D GIWAXS images of perovskite films from control (D) and ASPM (E) precursors deposited on a silicon wafer substrate. The incidence angle is 1°. qz, out-of-plane scattering vector; qxy, in-plane scattering vector. (F) The integrated intensity plots azimuthally along the ring assigned to the a-FAPbI3 (100) lattice plane.

than the retail prices of commercial PbI2 and FAI (tables S8 and S9). The d-FAPbI3 microcrystals also showed six-month storage stability under ambient conditions (fig. S8). The thermogravimetric analysis (TGA) and differential thermal analysis (DTA) indicated a precise stoichiometric ratio of FAPbI3 (figs. S9 and S10). Perovskite film crystallization

To prepare a precursor solution for the fabrication of FA0.85MA0.1Cs0.05PbI3 perovskite film, we used a mixture of aqueous synthesized FAPbI3, MAPbI3, and CsPbI3 microcrystals and the mixture of commercial FAI, PbI2, CsI, and MAI in solvents for the ASPM and control precursor solution, respectively. As shown in the dynamic light scattering (DLS) measurement (Fig. 3A), the control precursor formed colloids with an average size of 134 nm, whereas the ASPM precursor contained larger colloids with an average size of 293 nm. The ASPM displayed a stronger typical Tyndall effect that also confirmed the presence of larger colloids. We used synchrotron-based in-situ grazingincidence wide-angle x-ray scattering (GIWAXS) to analyze the crystallization kinetics of the ascoated perovskite solution under annealing for 340 s. As shown in Fig. 3B, at the initial state, three peaks were found in control films located

2 February 2024

at scattering vector q = 5.21, 6.64, and 8.49 nm−1, which were assigned to the perovskite intermediate phase of DMSO adduct or the hexagonal perovskite polytype 2H (30). However, in the ASPM film, no obvious peak was observed at the initial state given the presence of the amorphous perovskite colloids. As the annealing time increased, the a-phase FAPbI3 (100) plane of both two films appeared at the same time. The control perovskite film phase formed from the intermediate phase, whereas the ASPM perovskite film formed directly from the FAPbI3 colloids. During annealing for ~250 s until the formation of the final state, the PbI2 peak appeared in the control perovskite film, indicating the occurrence of nonstoichiometric or unbalanced composition reactions in the perovskite film for the control precursor solution. However, the PbI2 peak was absent in the ASPM perovskite film because of the stoichiometric composition. For the ASPM film, the peak of a-FAPbI3 (100) became more intense, indicating the generation of high-quality a-FAPbI3 films (Fig. 3C). The SEM and AFM results showed that the surface morphology of the film fabricated from the ASPM precursor was compact and flat, whereas the one derived from the control precursor exhibited wider grain boundaries with some small crystals filled (fig. S11). 4 of 8

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

Fig. 4. Carrier transport mechanism. (A) Representative DFT calculation of Ca2+ in a-FAPbI3 lattices using 2×2×2 perovskite supercells. (B) The calculated destabilization energy of the three defect models (FA+ site substitution, Pb2+ site substitution, and interstitial) based on the 1×1×1, 2×2×2, and 3×3×3 perovskite supercell size. The absence of certain data means the results are unavailable because the cells were distorted and not converged correctly during the geometry optimization, which

The 2D GIWAXS patterns shown in Fig. 3, D and E, exhibited an intense and integrated DebyeScherrer ring at q = 10 nm−1, which corresponded to the a-FAPbI3 (100) lattice plane. To obtain azimuthal-dependent intensities, we extract 1D integrated intensity data for the a-FAPbI3 (100) plane in GIWAXS patterns (Fig. 3F). The preferred azimuthal angles 24° (156°) and 54° (126°) of ASPM film confirmed that the large-sized polyiodide colloids could trigger the preferred Zhu et al., Science 383, 524–531 (2024)

suggests highly unstable defects. (C) Steady-state PL spectra of the control and ASPM precursor–based perovskite films, HTL/perovskite films, and perovskite/C60 films. (D) The charge-carrier diffusion length of the control and ASPM precursor– based perovskite films. (E) Dark current (I)–voltage (V) data for hole-only devices based on control and ASPM precursors. VTFL, trap filled limit voltage. (F) tDOS obtained by thermal admittance spectroscopy for the control and ASPM devices.

orientation during film formation (fig. S12), in which two dominant preferred orientations could improve carrier transport compared with random distributions and reduced shallow traps (31). Charge transport

To find out how the presence of impurities affected the stability of the perovskite structure, we used density-functional theory (DFT) to

2 February 2024

calculate the destabilization energies associated with the placement of impurities, including Ca2+, Na+, K+, and vacancies, within the FAPbI3 perovskite lattice. The DFT was conducted by using three different models (A-site substitution, B-site substitution, and interstitial site) and varying perovskite supercell sizes (1×1×1, 2×2×2, and 3×3×3 perovskite supercells) (Fig. 4A, fig. S13, and table S10). The destabilization energy was the unstable energy for the defect states 5 of 8

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

Fig. 5. Characterization of the PSCs. (A) Cross-sectional SEM image of the p-i-n PSCs. (B) Current density–voltage (J-V) curves of control and ASPM precursor-based PSCs. (C) Statistics of the photovoltaic parameters of PSCs based on control and ASPM precursors. (D) EQE spectra and integrated JSC of

compared with that of the pure perovskite (0 eV) by the introduction of ions or vacancy, suggesting that the existence of defects contributes to the instability of the perovskite structure. As shown in Fig. 4B, the destabilization energy for Ca2+, Na+, K+, and vacancy was Ca2+ < K+ < Na+ < vacancy, regardless of the supercell size. We constructed various supercells with substitution ratios of 20, 2.5, and 0.7%. Nevertheless, the observed changes in the instability energy were negligible for this range of supercells, implying that even a small quantity of impurity can destabilize the perovskite structure and affect the charge-carrier transport. We conducted additional DFT calculations to evaluate the defect band energy associated with both Ca-introduced defects and intrinsic defects, alongside their defect formation energy within the context of three distinct I conditions for FAPbI3 film growth, as shown in figs. S14 Zhu et al., Science 383, 524–531 (2024)

the champion ASPM precursor–based PSCs. (E) The stabilized output of the champion device tracked at the MPP under standard AM 1.5G illumination. (F) Unencapsulated PSC performance tracking at the MPP under continuous AM 1.5G illumination for 1000 hours in N2 atmosphere.

and S15 and table S11. The results reveal that Ca2+ could serve as shallow traps in the FAPbI3, underscoring the necessity of eliminating Ca2+ in the perovskite. We observed the photogenerated carrier behavior of the perovskite film through the photoluminescence (PL) and time-resolved photoluminescence (TRPL) spectra. The PL spectra at 793 nm showed no peak shift between control and ASPM films, indicating similar optical bandgaps (Fig. 4C and fig. S16). The steady-state PL intensity of the ASPM film showed higher intensity than that of the control film, which was attributed to the lower concentration of impurities. TRPL was measured on the control and ASPM perovskite films with or without a carrier quenching layer (fig. S17). TRPL data demonstrated approximately two times longer average carrier lifetimes (tave) for the ASPM perovskite film compared with that

2 February 2024

of the control film (3163 ns versus 1675 ns) (table S12) (32). By conducting the temperaturedependent PL and TRPL (figs. S18 and S19 and table S13) as complementary analyses, we can reasonably infer that the shorter average carrier lifetimes may be attributed to the presence of quenching traps in the control perovskite films. We evaluated the charge-carrier diffusion length on the basis of the TRPL results of perovskite films with a carrier quenching layer {[2(9H-carbazol-9-yl) ethyl] phosphonic acid (2-PACz) and C60 for hole and electron quenching, respectively} according to a simplified 1D diffusion model (Eq. 7) (33): sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi LD 2 1 ¼
1 ð7Þ p tQ =t0 L where LD is the charge-carrier diffusion length, L is the thickness of the perovskite film, and tQ 6 of 8

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

and t0 are the time-resolved photoluminescence with or without the quenching layer, respectively. The result showed that the electron and hole diffusion lengths of control films were 981 and 799 nm, respectively, whereas ASPM showed electron and hole diffusion lengths of 2449 and 2390 nm, respectively, indicating a prolonged charge-carrier lifetime and balanced charge extraction (Fig. 4D). The ASPM films also displayed a lower density of trap states within the bandgap (figs. S20 and S21 and table S14). We estimated the trap densities (Ntrap) using the hole-only device structure of glass/ITO/ 2-PACz/perovskite/Au. The Ntrap of the ASPM film was calculated to be 2.84 × 10 15 cm −3 , which was lower than that of the control device (5.57 × 1015 cm−3) (Fig. 4E, fig. S23, and table S15). We also conducted temperature-dependent thermal admittance spectroscopy to distinguish the trap states in PSCs. The shallow trap states were more likely to originate internally within the perovskite (Band I), whereas the deep trap states primarily correlated with surface defects in the perovskite (Bands II and III) (34). In the ASPM, the trap density of states (tDOS) intensity decreased 10-fold in the Band I region (Fig. 4F, figs. S22 and S23, and table S16). The Band I region could be related to the impurities present in the perovskite films, particularly the Ca2+ impurity (fig. S24). These results indicated that a relatively high density of defects in ASPM perovskite was reduced by the removal of impurities. Photovoltaic performance and stability

We fabricated p-i-n inverted structure PSCs with the configuration of ITO/2-PACz/ FA0.85MA0.1Cs0.05PbI3/C60/BCP/Cu { ITO, indium tin oxide; 2-PACz, [2-(9H-Carbazol-9-yl) ethyl] phosphonic acid; FA, formamidinium; MA, methylammonium; BCP, bathocuproine} to confirm the performance of ASPM perovskite precursors (Fig. 5A and fig. S25). The PSCs based on the ASPM precursor had a champion efficiency of 25.6%, with an open-circuit voltage (VOC) of 1.19 V, a short-circuit current density (JSC) of 25.0 mA/cm2, and a fill factor (FF) of 0.860, whereas the control PSCs achieved a champion efficiency of 23.7%, with a VOC of 1.14 V, JSC of 24.9 mA/cm2, and an FF of 0.832 (Fig. 5B). We compared 40 devices in eight different batches to check the fabrication reproducibility (Fig. 5C). The ASPM devices exhibited good reproducibility, averaging a PCE of 25.0% and a VOC of 1.18 V, whereas the control devices yielded an average PCE of 23.1% and an average VOC of 1.13 V. The ASPM PSCs exhibited a prolonged carrier lifetime [4.29 ms in transient photovoltage (TPV)] and a shorter photocurrent decay [0.326 ms in transient photocurrent (TPC)] compared with those of control PSCs (2.85 ms in TPV and 0.667 ms in TPC), suggesting enhanced radiative recombination and better charge-carrier extraction (fig. S26). The integrated Zhu et al., Science 383, 524–531 (2024)

JSC (24.9 mA/cm2 for ASPM and 24.6 mA/cm2 for control) from the external quantum efficiency (EQE) matched well with the JSC from the J-V curve of 25.0 and 24.9 mA/cm2 for ASPM and control devices, respectively, within an error of 50% of total megafauna biomass. Deer include all Cervidae, and wild pigs include all Suidae (primarily introduced wild boar, Sus scrofa). Equids include all Equidae but primarily feral horses (Equus ferus caballus). Large, broad-muzzled bovids include the genera Bison, Bos, and Syncerus. (C) Native and introduced deer can reduce plant diversity by selectively browsing preferred plants (49, 50). [Photo: Murray Foubister] (D) Pigs are distinct for belowground foraging and are dietary generalists, despite their relatively narrow muzzles (51). Feral pigs often increase plant diversity, at times doubling native plant diversity by suppressing competitive dominants (52). [Photo: Valentin Panzirsch] (E) Feral horses (E. ferus caballus) appear to have mixed effects on local plant diversity. (F) Bulk-grazers, like cape buffalo (Syncerus caffer) and bison (Bison bison), tend to increase plant diversity (53). Our results suggest that this is driven by their inability to selectively feed, forcing them to consume the most abundant (i.e., competitively dominant) plants. [Photo: Stig Nygaard] 5. E. J. Lundgren et al., Proc. Natl. Acad. Sci. U.S.A. 117, 7871–7878 (2020). 6. C. P. Hedberg, S. K. Lyons, F. A. Smith, Glob. Ecol. Biogeogr. 31, 294–307 (2022). 7. E. J. Lundgren et al., Science 372, 491–495 (2021). 8. P. A. Werner, Austral Ecol. 30, 625–647 (2005). 9. D. Spear, S. L. Chown, J. Zool. 279, 1–17 (2009). 10. T. M. Blackburn et al., PLOS Biol. 12, e1001850 (2014). 11. A. D. Wallach, E. J. Lundgren, W. J. Ripple, D. Ramp, Conserv. Biol. 32, 962–965 (2018). 12. M. Rejmánek, D. Simberloff, Environ. Conserv. 44, 97–99 (2017). 13. J. N. Price et al., Nat. Ecol. Evol. 6, 1290–1298 (2022). 14. M. E. Soulé, Bioscience 35, 727–734 (1985).

Lundgren et al., Science 383, 531–537 (2024)

15. D. F. Owen, R. G. Wiegert, Oikos 36, 376–378 (1981). 16. U. Gélin et al., bioRxiv 2023.02.09.527903 [Preprint] (2023); https://doi.org/10.1101/2023.02.09.527903. 17. S. J. O’Hanlon et al., Science 360, 621–627 (2018). 18. Conference of Parties to the UN Convention on Biological Diversity, Kunming-Montreal Global Biodiversity Framework CBD/COP/15/L25 (2022); https://www.cbd.int/conferences/ 2021-2022/cop-15/documents. 19. Y. Rohwer, E. Marris, Conserv. Sci. Pract. 3, e411 (2021). 20. A. D. Wallach et al., Conserv. Biol. 34, 997–1007 (2020). 21. D. H. Janzen, Oikos 45, 308–310 (1985). 22. D. M. Wilkinson, J. Biogeogr. 31, 1–4 (2004). 23. M. Sagoff, Conserv. Biol. 34, 581–588 (2020).

2 February 2024

24. A. D. Wallach, W. J. Ripple, S. P. Carroll, Trends Ecol. Evol. 30, 146–153 (2015). 25. E. J. Lundgren et al., Functional traits—not nativeness—shape the effects of large mammalian herbivores on plant communities [Dataset], Dryad (2023); https://doi.org/10.5061/dryad.b5mkkwhj9. 26. Materials and methods are available as supplementary materials. 27. S. Lowe, M. Browne, S. Boudjelas, M. De Poorter, “100 of the world’s worst invasive alien species: a selection from the global invasive species database” (The Invasive Species Specialist Group, 2000); http://www.iucngisd.org/gisd/100_worst.php. 28. S. Grange, P. Duncan, J.-M. Gaillard, Proc. R. Soc. Lond. Ser. B 276, 1911–1919 (2009). 29. A. Zizka et al., Proc. Natl. Acad. Sci. U.S.A. 119, e2208629119 (2022). 30. D. Simberloff, B. Von Holle, Biol. Invasions 1, 21–32 (1999). 31. J. Cubas et al., Proc. R. Soc. London Ser. B 286, 20190136 (2019). 32. Plants of the World Online (POWO), Facilitated by the Royal Botanic Gardens, Kew; http://www.plantsoftheworldonline.org/. 33. S. Faurby et al., Ecology 99, 2626 (2018). 34. C. M. Janis, D. Ehrhardt, Zool. J. Linn. Soc. 92, 267–284 (1988). 35. A. Eskelinen, W. S. Harpole, M.-T. Jessen, R. Virtanen, Y. Hautier, Nature 611, 301–305 (2022). 36. N. A. Mungi, Y. V. Jhala, Q. Qureshi, E. le Roux, J.-C. Svenning, Nat. Ecol. Evol. 7, 1645–1653 (2023). 37. D. J. Augustine, S. J. McNaughton, J. Wildl. Manage. 62, 1165–1183 (1998). 38. R. Ø. Pedersen, S. Faurby, J.-C. Svenning, Glob. Ecol. Biogeogr. 32, 1814–1826 (2023). 39. W. J. McShea, H. B. Underwood, J. H. Rappole, Eds., The Science of Overabundance: Deer Ecology and Population Management (Smithsonian Institution Press, 1997). 40. J.-C. Svenning et al., Proc. Natl. Acad. Sci. U.S.A. 113, 898–906 (2016). 41. E. le Roux et al., Ecography 42, 1115–1123 (2019). 42. D. Boltovskoy et al., Hydrobiologia 848, 2225–2258 (2021). 43. J. Trepel et al., Nat. Ecol. Evol. 10.1038/s41559-024-02327-6 (2024). 44. E. J. Lundgren et al., J. Anim. Ecol. 91, 2348–2357 (2022). 45. E. S. Bakker et al., Proc. Natl. Acad. Sci. U.S.A. 113, 847–855 (2016). 46. R. M. Cowling, A. Kamineth, M. Difford, E. E. Campbell, Afr. J. Ecol. 48, 135–145 (2010). 47. D. J. Eldridge, J. Ding, S. K. Travers, Biol. Conserv. 241, 108367 (2020). 48. F. Cardou, M. Vellend, Biol. Conserv. 280, 109968 (2023). 49. C. W. Habeck, A. K. Schultz, AoB Plants 7, plv119 (2015). 50. J.-L. Martin, S. A. Stockton, S. Allombert, A. J. Gaston, Biol. Invasions 12, 353–371 (2010). 51. D. E. Wilson, R. A. Mittermeier, Eds., Handbook of the Mammals of the World, Volume 2: Hoofed Mammals (Lynx Edicions, 2011). 52. M. J. S. Hensel, B. R. Silliman, E. Hensel, J. E. K. Byrnes, Ecology 103, e03572 (2022). 53. C. E. Burns, S. L. Collins, M. D. Smith, Biodivers. Conserv. 18, 2327–2342 (2009). ACKN OW LEDG MEN TS

We thank R. Buitenwerf and J. Kerby for help in designing analyses. We thank A. D. Wallach, S. Archibald, and three anonymous reviewers for helpful feedback on earlier drafts. Funding: VILLUM FONDEN via the VILLUM Investigator grant 16549 (J.C.S.); Danish National Research Foundation via Center for Ecological Dynamics in a Novel Biosphere (ECONOVO) grant DNRF173 (J.C.S.); and Independent Research Fund Denmark–Natural Sciences via the MegaComplexity project, grant 0135-00225B (J.C.S.). Author contributions: Conceptualization: E.J.L., J.-C.S., S.M., and J.A.K. Methodology: E.J.L., J.B., M.T., R.Ø.P., J.-C.S., S.M., J.A.K., and J.T. Investigation: E.J.L., S.M., J.A.K., and J.T. Visualization: E.J.L., J.B., and E.l.R. Funding acquisition: J.-C.S. Project administration: J.-C.S. and E.J.L. Supervision: J.-C.S. Writing – original draft: E.J.L., J.B., J.-C.S., and E.l.R. Writing – review & editing: E.J.L., J.B., J.T., E.l.R., S.M., J.A.K., R.Ø.P., P.P., M.T., and J.C.S. Competing interests: The authors declare that they have no competing interests. Data and materials availability: All data and the core analysis scripts are provided in Dryad (24). 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.adh2616 Materials and Methods Supplementary Text Figs. S1 to S21 Tables S1 and S2 References (54–282) Submitted 21 February 2023; accepted 30 November 2023 10.1126/science.adh2616

6 of 6

RES EARCH

ORGANIC CHEMISTRY

Terminal C(sp3)–H borylation through intermolecular radical sampling Miao Wang, Yahao Huang, Peng Hu* Hydrogen atom transfer (HAT) processes can overcome the strong bond dissociation energies (BDEs) of inert C(sp3)–H bonds and thereby convert feedstock alkanes into value-added fine chemicals. Nevertheless, the high reactivity of HAT reagents, coupled with the small differences among various C(sp3)–H bond strengths, renders site-selective transformations of straight-chain alkanes a great challenge. Here, we present a photocatalytic intermolecular radical sampling process for the iron-catalyzed borylation of terminal C(sp3)–H bonds in substrates with small steric hindrance, including unbranched alkanes. Mechanistic investigations have revealed that the reaction proceeds through a reversible HAT process, followed by a selective borylation of carbon radicals. A boron-sulfoxide complex may contribute to the high terminal regioselectivity observed.

C

hemists have long sought to efficiently and sustainably transform unreactive feedstock alkanes into value-added fine chemicals through C–H bond functionalization (1–10). The emergence of transition metal catalysis has led to great advancements in the functionalization of C(sp2)–H bonds in aromatic compounds (5–8) and of C(sp3)–H bonds in substrates with directing groups (9). Nonetheless, the direct modification of inert alkanes in a selective manner remains a major challenge, with only a few pioneering studies that use precious metal catalysts having been reported (10–14). Furthermore, the reactivity of a C–H bond toward transition metal catalysts is typically dependent on the corresponding bond acidity (Fig. 1A) (15), with terminal alkynes and aromatic rings reacting preferentially to alkane fragments, resulting in substrate limitations for C(sp3)–H bond functionalization (10–14). By contrast, C–H bond activity toward homolysis depends primarily on the bond dissociation energy (BDE), with C(sp3)–H bonds displaying the lowest levels of BDEs (Fig. 1A) (16, 17). Thus, through an activated species, cleavage of C(sp3)–H bonds may proceed through a HAT (hydrogen atom transfer) procedure, enabling compatibility with arene, alkene, and even terminal alkyne moieties while the C(sp3)–H bond is cleaved to form an alkyl radical for further reaction. Despite HAT being a fundamental mechanism in various chemical, environmental, and biological processes (1, 3), traditional HAT reactions usually require the application of stoichiometric HAT precursors or harsh conditions (2, 18). However, the development of photocatalysts has enabled ketones (19); decatungstate anions (20–23); eosin Y (24); heteroatom radicals, such as oxygen (25–29),

Institute of Green Chemistry and Molecular Engineering, Lehn Institute of Functional Materials, School of Chemistry, Sun Yat-sen University, Guangzhou 510275, China. *Corresponding author. Email: [email protected]

Wang et al., Science 383, 537–544 (2024)

nitrogen (30–32), and sulfur (33); and halogencentered radicals (34–39) to serve as efficient HAT agents catalytically (1–4). Intramolecularly, the HAT strategy has been used for substrates containing directing groups that can generate radicals, facilitating the cleavage of remote C–H bonds (18, 40, 41). Nevertheless, achieving C(sp3)–H bond homolysis intermolecularly under mild conditions generally necessitates the use of highly reactive HAT reagents, which often manifest low regioselectivity and lead to product mixtures. The slight difference in BDE values among C(sp3)–H bonds (Fig. 1B) is another reason for the poor selectivity, although thermodynamically, methylene or methine groups represent the preferred reaction sites because of their lower bond strength. To address this long-standing issue, some ingenious studies have used substrates with C(sp3)–H bonds exhibiting different steric hindrance or distinct electronic properties to generate regioselective C(sp3)–H functionalization products (1, 3). However, achieving selective modification of simple alkanes, particularly the terminal functionalization of unbranched chains, poses a formidable challenge owing to the formation of relatively unstable primary carbon radical intermediates through HAT from stronger C–H bonds (Fig. 1B) (19–21, 27, 37, 38, 42–45). We considered two general pathways (Fig. 1C) to address the regioselectivity challenge that use n-hexane as a model molecule. The first pathway involves a regioselective HAT followed by functionalization (Fig. 1C, upper pathway). However, this process has thus far achieved only moderate terminal selectivity because of the previously noted difficulties. As an elegant example, Aggarwal and Noble have reported pioneering work on terminalmethyl selective borylation of sterically hindered substrates that uses an in situ–formed, chloride radical–boron “ate” complex as a selective HAT catalyst (43). This approach abstracted sterically unhindered C(sp3)–H bonds and achieved unusual selectivity. Nonetheless, the

2 February 2024

system exhibited unsatisfactory regioselectivity when applied to straight-chain substrates, such as pentane, resulting in an approximately 1:1 methyl to methylene ratio. Very recently, the Xia group (44) and Aggarwal group (45) independently reported intriguing C(sp3)–H borylation reactions that use base metal chlorides under light irradiation. The regioselectivity observed was similar to that of Aggarwal and Noble’s former metal-free work (43), thereby confirming the challenge of achieving selective borylation of unbranched alkanes. The second pathway begins with an unselective HAT step, leading to the formation of both primary and secondary radicals, followed by a radical sampling procedure (22) to realize the formation of terminal-functionalized product. The unreactive secondary radicals revert to the substrate molecules through the reverse HAT reaction (Fig. 1C, lower pathway). This strategy appears to be more reliable when the system can differentiate between primary radicals and secondary radicals on the basis of steric hindrance. Iron (III) chloride has been established as a promising precursor for generating chlorine radicals through a photoinduced ligand-to-metal charge transfer (LMCT) process (34). Our group and others have successfully applied this protocol in several photocatalytic reactions (34, 46). In this work, we applied the aforementioned strategy to achieve terminal C(sp3)–H borylation of unbranched alkanes and substrates with varying steric hindrance. The reaction is facilitated by photoinduced iron catalysis involving a chlorine radical–assisted reversible HAT process, followed by selective borylation of a primary radical intermediate at the terminal position. The observed regioselectivity is putatively regulated by an in situ–formed boron-sulfoxide species. Reaction optimization

Our initial study used cyclohexane and bis(catecholato)diboron (B2cat2) as the reactants to search for the optimal catalytic system for regioselective C(sp3)–H borylation. After exhaustive searching, dimethyl sulfoxide (DMSO) was found to be a good oxidant to realize the desired reaction, with iron chloride (FeCl3) as the catalyst under irradiation by purple light-emitting diodes (LEDs) at room temperature (figs. S1 to S3). Further screening studies led to optimized standard conditions of applying FeCl3 (10 mol%), tetraethylammonium chloride (TEACl) (50 mol%), and DMSO (1.4 equiv) in mixed MeCN/acetone (1:1), with B2cat2 as the limiting reagent (tables S1 to S8). This simple and inexpensive system produced the target cyclohexyl pinacol boronic ester (Bpin) in a yield of 67% after treatment with pinacol. Because sulfoxides are readily available oxidants with easily tunable substituents, we quickly realized that we might achieve high 1 of 8

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

Fig. 1. Challenges in achieving selectivity for C(sp3)–H bonds in unbranched alkanes through HAT. (A) Acidity and BDE trends of C–H bonds with different hybridization. (B) C–H BDEs of n-hexane (16) and the unsolved challenge of terminal C(sp3)–H bond functionalization through HAT. (C) General designs for terminal functionalization of C(sp3)–H bonds through HAT, illustrated with the example of

regioselectivity for the C(sp3)–H borylation of unbranched alkanes by choosing a suitable sulfoxide. Therefore, we selected n-hexane as the model substrate to test the performance of different sulfoxides under the standard conditions (Fig. 1D). Application of DMSO and diethyl sulfoxide resulted in moderate regioselectivity with a 42:58 ratio of methyl to methylene borylated products (entries 1 and 2). Sulfoxides with larger substituents, such as isopropyl, showed slightly improved terminal selectivity (entry 3). However, when di-tert-butyl sulfoxide was applied, the reaction was completely inhibited, likely because the bulky tert-butyl group hindered the catalytic cycle (entry 4). Tetramethylene sulfoxide was completely selective for the terminal borylated product, although only a 5% yield was observed (entry 5). Further investigation revealed that methyl sulfoxide also showed moderate performance with respect to selectivity and product yield (entry 6). We found that diphenyl sulfoxide generated the terminal borylated product 1 in a yield of 37% Wang et al., Science 383, 537–544 (2024)

n-hexane. (D) Selective C(sp3)–H borylation of n-hexane performed using different sulfoxides. Yields and selectivity were determined with GC-MS by using mesitylene as an internal standard unless otherwise noted. Experimental details are provided in the supplementary materials. B2cat2, Bis(catecholato)diboron; TEACl, tetraethylammonium chloride; r.t., room temperature.

without any other regioisomer (entry 7). Modification of the phenyl group with methyl or chloro substituents showed little difference in selectivity but lower product yields (entries 8 and 9). By increasing the amount of inexpensive and easily recyclable hexane to 20 equiv, we achieved a 55% isolated yield of hexyl 2,3dihydro-1H-naphtho[1,8-de][1,3,2]diazaborinine (Bdan) (1′, treated with 1,8-diaminonaphthalene 3instead of pinacol) with no loss of regioselectivity (entry 10). This success led to the regioselective C(sp3)–H borylation of unbranched alkanes under gentle photocatalytic conditions, a result previously unachievable without our sulfoxide-controlled photocatalytic procedure (43–45). Substrate scope study

Using the optimized reaction conditions, we conducted a compatibility test and found that the reaction succeeds with straight-chain substrates as well as those with varying degrees of steric hindrance (Fig. 2, 2 to 53). We

2 February 2024

employed 1,8-diaminonaphthalene (dan), Nmethyliminodiacetic acid (MIDA), or pinacol, respectively, to treat the intermediates formed from B2cat2, resulting in the production of stable boron products that were easily separable. Similarly to n-hexane, n-pentane (2) and n-heptane (3) generated terminal borylated products with excellent selectivity and moderate yields, with no regioisomers observed. Unexpectedly, longer-chain alkanes (4, 5) also exhibited outstanding selectivity, although lower product yields were observed. The alkane with two different types of methyl groups successfully produced the less hindered C(sp3)–H borylated compound as the only observed product (6). The reaction exhibited excellent regioselectivity for sterically hindered terminal methyl groups (7, 8), with the weaker methylene and methine positions remaining unreactive. Additionally, selective borylation of unbranched alkyl chlorides (9, 10) was successfully achieved. Electronic effects had a significant influence on the reactivity of the methyl 2 of 8

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

Fig. 2. Substrate scope of simple substrates. Reaction conditions: FeCl3 (10 mol%), TEACl (50 mol%), sulfoxide (1.4 equiv), alkane (5, 10, or 20 equiv), B2(cat)2 (0.2 mmol, 1 equiv), MeCN/acetone (1:1, 4 ml), 400 nm LEDs (20 W), 12 hours, r.t. Subsequently, 1,8-diaminonaphthalene (dan), N-methyliminodiacetic acid (MIDA), or pinacol was added. Complete experimental details can be found in the supplementary materials. The isolated yields are reported unless stated

group near electron-withdrawing substituents. For instance, 2-chloroheptane produced only the distal borylated product (11). When decanenitrile was employed, the terminal borylated product remained dominant, alWang et al., Science 383, 537–544 (2024)

otherwise. The ratios of the terminal borylated products and other regioisomers are shown in parentheses, as determined by 1H NMR. For cases in which only one product was observed, >95:5 is indicated. The positions of borylated regioisomers are indicated by red dots. *GC-MS analysis was performed for volatile products by using mesitylene as an internal standard. #Three-watt LEDs (400 nm) were used.

though a small fraction of regioisomers was observed (12). The reaction with ketones (13 to 18) and esters (19 to 25) provided a more complex illustration of the C(sp3)–H bond selectivity problem. Symmetrical substrates ex-

2 February 2024

hibited the same terminal regioselectivity as did unbranched alkanes, regardless of the chain length (13, 14, 25). Although radicals can be stabilized by carbonyl and oxygen atoms, the corresponding a positions did not 3 of 8

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

Fig. 3. Applications. Procedure details are found in supplementary materials. (A) Late-stage C(sp3)–H borylation of drug derivatives and complex molecules and the derivatization of the resulting alkyl boron compounds. Condition (a): Cs2CO3, 2,4,6-trimethyl-N'-nonylidenebenzenesulfonohydrazide, PhCl. (B) Large-scale reaction of cyclohexane and n-hexane in a flow reactor.

undergo reaction. For unsymmetrical substrates (15 to 24), the sterically unhindered methyl groups with weaker C(sp3)–H bonds located proximal to the carbonyls or oxygen atoms were found to have been deactivated. The terminal C(sp3)–H bonds remained unreactive even at the b positions, as exemplified by compounds 18 and 22. DMSO proved to be efficient for substrates with no regioselectivity challenges (26 to 30) and for compounds with steric hindrance and/or electron-withdrawing groups (31 to 53). A Wang et al., Science 383, 537–544 (2024)

wide range of simple feedstock chemicals could be transformed into value-added boron compounds through this inexpensive procedure. Moderate isolated yields were observed for simple alkanes (26 to 28), including ethane (26). As useful synthons, (borylmethyl)silanes could be directly produced by using various functionalized methyl silanes (29 to 36, 43, 44). The borylation of methylene and methine groups with weaker C(sp3)–H BDE values was found to be fully impeded in sterically more hindered environments or relatively electron-

2 February 2024

deficient positions (31 to 53). Even the active benzylic (35, 41) and allylic C–H bonds (36) were tolerated. Compared with noble metal– catalyzed C(sp3)–H borylation, another advantage of this method is the excellent compatibility of C(sp2)–H and C(sp)–H bonds. Phenyl (35, 41, 44, 50), terminal alkene (36), and terminal alkyne (37) groups were all well tolerated. A small degree of steric hindrance was sufficient to distinguish reactive positions (43 to 53). Short-chain substrates with isopropyl fragments all demonstrated excellent regioselectivity (38, 4 of 8

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

Fig. 4. Mechanism study. Standard conditions are shown in Fig. 1D, entry 1. The presented ratios of deuterated compounds are based on electron ionization HR GC-MS analysis unless otherwise noted. (A) Evidence of thioether and carbon radical formation. (B) Diene-mediated chlorine radical–trap reaction. (C) Parallel-mode KIE experiment. (D) Mixed-mode KIE experiment. (E) The evidence of reversible HAT performed using mixed n-octane (4a) and deuterated n-octane (4a-d18) as the substrates (in 1:1 mol ratio). H-D exchange of the substrates and the products are observed. (F) Study of hydrogen-atom donor for the reversible HAT procedure. (G) H-D exchange of deuterated substrate Wang et al., Science 383, 537–544 (2024)

2 February 2024

(65a-d7) and n-hexane (1) under the standard conditions. The presented hydrogen-atom ratios are based on the 1H NMR analysis of 65a-d′. (H) Reaction of n-hexane and electron-deficient olefin (66a) in the presence of diphenyl sulfoxide. (I) Reaction of n-hexane, electron-deficient olefin (66a), and B2cat2 in the presence of diphenyl sulfoxide. (J) The influence of sulfoxide or thioether on the regioselectivity of carbon radical–trap reaction of n-hexane with TEMPO. (K) 1H NMR and 11B NMR titration spectra of B2cat2 (1 equiv) with DMSO (0 to 4 equiv). (L) The influence of Ph2S and Ph2S(=O) on the regioselectivity of borylation of n-hexane under standard conditions (with 1.4 equiv of DMSO). 5 of 8

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

Fig. 5. Proposed mechanism.

39, 41, 42, 45 to 53). In the case of unsymmetrical ketones and esters (48 to 53), the a-terminal methyl groups of the carbonyl were also rendered inert (48, 49, 53), whereas b-methyl groups led to a small proportion of borylated regioisomers (51, 52). Moreover, this reaction showed versatility toward a wide range of functional groups, including chloro, bromo, ketone, ester, nitrile, amide, alkenyl, alkynyl, silyl, and boronic ester, thereby providing ample opportunities for further functionalization. In addition, the reaction can also occur under low-power (3-W) light irradiation, although it may lead to a decrease in reaction yield (28, 38, 42, table S9). To demonstrate the specific applications of this strategy in synthetic chemistry, it was applied to several complex molecules (Fig. 3, 54 to 61). The lorzone derivative, dicamba derivative, and methylparaben derivative were all selectively borylated (54, 55, 61). Additionally, the celecoxib derivative was borylated at the terminal methyl group instead of the benzylic methyl group, with excellent regioselectivity (56). Furthermore, branched alkanes and silanes could be selectively borylated by Wang et al., Science 383, 537–544 (2024)

using DMSO as an oxidant. The triclosan derivative with a pentan-3-yloxy group showed good methyl selectivity (59). The D-galactose derivative and Sandoz 58-035 precursor formed borylation products at the methyl silane groups successfully, leaving the other methyl groups untouched (57, 60). The memantine derivative, having an adamantane skeleton, also achieved high selectivity for methyl borylation (58). Subsequently, we tested some derivatization reactions of the Bpin group using 60 and 61 as substrates. The cholesterol acyltransferase inhibitor Sandoz 58-035 (60b) (47) can be easily generated from 60 in two steps. Furthermore, transformations of the Bpin group were tested by using the methylparaben derivative 61. Upon oxidation, alcohol 61b was efficiently generated (48). Treatment with vinylmagnesium bromide and I2 led to the vinylation product 61c in excellent yield (48). Modifying Aggrawal’s protocol and applying fluoroiodomethane and lithium diisopropylamide (LDA) led to the alkenyl fluoride 61d (49). Moreover, Pd-catalyzed Suzuki-Miyaura coupling of 61 with 4′-bromo4-cyano-biphenyl afforded the cross-coupling product 61e in a 92% yield (48). To showcase

2 February 2024

the method’s practicability, larger-scale reactions of cyclohexane and n-hexane were tested with a flow reactor (Fig. 3B and fig. S4). The cyclohexane reaction produced 16.43 g of cyclohexyl Bpin (28) at a rate of 1.8 g/hour. In the case of n-hexane, 9.82 g of n-hexyl boron MIDA (BMIDA) (1′′) was obtained as the sole product, with no regioisomers observed. The leftover iron in the products was tested with inductively coupled plasma mass spectrometry (ICP-MS), showing only trace amounts of iron impurities (table S10). The limiting factors for the borylation yield were investigated, revealing that the reaction ceases because of the consumption of B2cat2 and sulfoxide. In the presence of iron, these two compounds generate inactive 2,2′oxybis(benzo[d][1,3,2]dioxaborole (BcatOBcat) through a competitive reaction (figs. S5 to S8). The reaction can be reactivated with the further addition of B2cat2 and sulfoxide (table S11). Additionally, recycling experiments were conducted to demonstrate the stability of the catalyst system. The iron catalyst remains active even after being used for 6 cycles, resulting in a total turnover number (TON) of 33 (table S12). 6 of 8

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

Furthermore, efforts were made to achieve C(sp3)–H borylation by using alkanes as limiting reagents. On the basis of the aforementioned experiments, it was observed that simple alkanes can be used as a limiting reagent by adding B2cat2 and sulfoxide in portions, albeit with lower yields (tables S13 to S17). However, this strategy was not effective for complex substrates such as the lorzone derivative, triclosan derivative, and celecoxib derivative. Fortunately, although an excess amount of substrate was necessary for these complex substrates, the unreacted alkanes could be recovered in excellent yields because of the reversible HAT procedure. We conducted additional experiments in which both the products and the unreacted alkane substrates were isolated. We observed that these experiments resulted in minimal loss of substrates, with a maximum depletion of 5% (table S18). These findings provide evidence that no significant side reactions occur involving the alkane substrate during the course of the reaction. This conclusion is further supported by the analysis of the crude reaction mixtures performed with gas chromatography–mass spectrometry (GC-MS). Mechanistic studies

Mechanistic experiments were carried out to obtain more comprehensive insights. As previously reported (34, 46), ultraviolet-visible spectra of the individual reaction components as well as the reaction mixture were acquired to confirm the presence of the [FeCl4]− species as the photocatalyst (fig. S9). Sulfoxide was presumed to function as the terminal oxidant for the reaction. The application of diphenyl sulfoxide led to the production of the reduced product, diphenyl sulfide (62), with a yield of 83%, supporting the aforementioned hypothesis (Fig. 4A, left). By incorporating stoichiometric 2,2,6,6-tetramethylpiperidinoxyl (TEMPO), the carbon radical generated from cyclohexane was captured in 63 with a yield of 67% (Fig. 4A, right), indicating the formation process of a carbon radical. Hepta-1,6-diene, a typical radical acceptor, was used to capture the chlorine radical during the reaction with stoichiometric TEACl (1.5 equiv) under standard conditions (Fig. 4B). The resultant cyclic product containing chlorine (64) was isolated in a 58% yield, confirming the generation of chlorine radical. Moreover, the low quantum yield (0.0601, figs. S10 and S11) and the light on/off experiment (fig. S12) provide evidence that a radical chain mechanism is unlikely for our borylation reaction. To investigate the distinctive selectivity for unbranched substrates, a series of experiments was conducted with n-hexane. Kinetic isotopic effect (KIE) experiments were conducted with both cyclohexane and cyclohexane-d12 in parallel (Fig. 4C and fig. S13) and mixed modes (Fig. 4D and fig. S14), yielding the identical KIE value Wang et al., Science 383, 537–544 (2024)

of 1. These findings indicate that the HAT step is not rate-determining. During the mixed-mode KIE experiment, H-D–exchanged products 28-d9 and 28-d10 were detected with high-resolution GC-MS, indicating a reversible HAT process. To further demonstrate the reversibility of the HAT procedure for unbranched alkanes, a reaction using 4a and 4a-d18 (1:1) as substrates was performed. A range of H-D–exchanged substrates (4a-d1, 4a-d16-d18) and products (4-d1-d2, 4-d13-d17) were detected, indicating the reversible HAT procedure (Fig. 4E and supplementary materials 4.8.1). Additionally, the 4-d16/4-d17 ratio was much larger than the 4-d1/4-d0 ratio and was similar to the ratios between 4a-d1/4a-d0 and 4a-d17/4a-d18, illustrating a strong KIE between the alkyl radical and the hydrogen-atom donor in the reversible HAT process. This may also explain the absence of 28-d1 in the mixed-mode KIE experiment (Fig. 4D). Subsequently, to identify the hydrogen-atom donor involved in the reversible HAT, a series of control experiments were designed (table S19). It was observed that in the absence of B2cat2 or Ph2S(=O), H-D exchange stopped, indicating that both B2cat2 and Ph2S(=O) were necessary for the reversible HAT. A series of potential hydrogen-atom donors related or unrelated to B2cat2 and Ph2S(=O) were tested for reversible HAT (Fig. 4F and table S20). It was found that neither H2O, nor catechol, nor boric acid [B(OH)3] enabled efficient reversible HAT. However, significant H-D exchange was observed when benzo[d][1,3,2]dioxaborol-2-ol (BcatOH) was added. Furthermore, when BcatOH and Ph2S(=O) were applied simultaneously, an even stronger reversible HAT was demonstrated. Nuclear magnetic resonance (NMR) titration of BcatOH with Ph2S(=O) indicated interactions between the two compounds (figs. S15 and S16). These findings suggested that the BcatOHPh 2 S(=O) combination was the effective hydrogen-atom donor for the reversible HAT procedure in the reaction. To further verify the reversible HAT, a deuterium-labeled substrate (65a-d7) was used to react with hexane and B2cat2 under standard conditions, and H-D– exchanged 65a-d′ was isolated and confirmed by NMR (Fig. 4G and figs. S17 to S19). Additionally, the deuterium-labeled substrate 50a-d with unlabeled C(sp3)–H was also conducted under standard conditions, leading to the redistribution of the deuterium label in the substrate (50a-d', fig. S20), clearly indicating the reversible nature of the HAT process. When treated under the standard conditions of the borylation reaction with 2-benzylidenemalononitrile, a typical alkyl radical acceptor, in place of B2cat2, n-hexane unexpectedly produced Giese-type regioisomer mixtures (66) even in the presence of diphenyl sulfoxide (Fig. 4H). The regioisomeric ratios on different carbons were 20:59:21, demonstrating a representative chlorine radical–induced methylene-preferred selec-

2 February 2024

tivity. Further investigation using n-hexane, 2-benzylidenemalononitrile, and B2cat2 together showed that the major Giese-type products were methylene-functionalized regioisomers, with the terminal borylated hexane being the sole borylation product (Fig. 4I). The enhanced methylene selectivity of the Giese-type reaction is possibly a result of competition at the methyl positions with the borylation process. Overall, these results clearly indicate a “free” chlorine radical–induced HAT procedure without specific selectivity control, as well as a regioselective borylation of the resulting carbon radicals. The conclusion was further corroborated through reactions with n-hexane, TEMPO, and B2cat2, conducted under altered conditions (Fig. 4J). Regardless of the presence or absence of DMSO and the use of diphenyl thioether or diphenyl sulfoxide, the TEMPO-trapped product (67) consistently exhibited a characteristic “free” chlorine radical–induced regioselectivity. In conjunction with the optimization study with various sulfoxide species (Fig. 1D), it is reasonable to assert that the regioselectivity was established during the formation of the C-B bond and that the concomitant use of sulfoxide and B2cat2 played a pivotal role in achieving very good regioselectivity. Thus, it is plausible to propose that a complex of B2cat2 and sulfoxide (or the sulfide formed after reduction) with appropriate steric hindrance was generated and that it impeded the borylation reaction with sterically encumbered secondary and tertiary carbon radicals. Comparable B2cat2-DMAc (N,N-dimethylacetamide) (50, 51) and B2cat2-DMF (N,N′-dimethylformamide) (52) complexes have also been reported. To provide evidence for the formation of the complex, 1H NMR and 11B NMR titration experiments of B2cat2 with DMSO were undertaken (Fig. 4K). As the amount of DMSO was increased, a gradual shift in the proton peaks of B2cat2 was observed, along with the appearance of a two-peak signal instead of the singlet B2cat2 peak in 11B NMR spectra. These observations indicate an interaction between B2cat2 and DMSO. Titration experiments were also conducted on the B2cat2-DMS system to rule out any interaction, as no significant change was observed in both 1 H NMR and 11B NMR spectra (fig. S21). It was previously known that B2cat2 could interact with DMSO to form the oxidized BcatOBcat species at high temperature (53). We found that this procedure could be carried out at room temperature with iron under light irradiation (fig. S5). Further NMR studies with varied combinations of reagents supported the existence of the B2cat2-DMSO complex, along with some other boron species. (figs. S5 to S8). Moreover, diphenyl sulfide was introduced under standard conditions (in the presence of DMSO) with n-hexane, but no change in the borylation selectivity at the methyl position was observed (Fig. 4L and table S21). Conversely, even a small 7 of 8

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

amount (0.2 equiv) of diphenyl sulfoxide had a significant impact on enhancing the ratio of terminal borylation. These observations suggest that a B2cat2 complex with an appropriate sulfoxide, rather than sulfide, is necessary to regulate the regioselectivity. On the basis of mechanistic experiments and prior studies on photocatalysis that used FeCl3 (34, 46), we propose a mechanism for the selective borylation of n-hexane (Fig. 5). Specifically, the photoactive [FeCl4]− complex formed from FeCl3 was excited by the purple light, which initiated an LMCT process that released a chlorine radical. This procedure ultimately led to HAT with n-hexane to generate carbon radicals in an unselective manner. Upon interaction with sulfoxides, a B2cat2 molecule formed a complex (I), which subsequently reacted with alkyl radicals to produce the borylated alkane product. This radical-sampling procedure was performed in a regioselective manner dependent on steric hindrance. When diphenyl sulfoxide was used, the sterically unhindered terminal radical was preferentially borylated, whereas reactions on secondary positions were completely inhibited. A boron radical intermediate stabilized by sulfoxide (II) formed from I was decomposed into thioether and benzo[d][1,3,2]dioxaborol2-ol radical (BcatO·), and this radical acted as an oxidant, undergoing a single-electron transfer with Fe(II) to regenerate [FeCl4]− and produce BcatOH (detected, figs. S5 to S8). And the unreacted secondary alkyl radicals underwent reversible HAT processes with BcatOH-sulfoxide to regenerate alkanes. A competitive reaction occurred in which B2cat2 was consumed slowly with sulfoxide under the reaction conditions, generating a thioether and a BcatOBcat (detected, figs. S5 and S7). We have achieved the regioselective terminal C(sp3)–H borylation of unbranched alkanes and substrates with varied steric hindrance through FeCl3 photocatalysis. Unlike traditional metal-catalyzed procedures, the method demonstrates the capability to selectively functionalize C(sp3)–H bonds over C(sp2)–H and C(sp)–H bonds, displaying a broad functional group

Wang et al., Science 383, 537–544 (2024)

tolerance. This strategy presents a simple and convenient approach for achieving regioselective C(sp3)–H bond functionalization through HAT, particularly for challenging substrates with a small degree of steric hindrance. RE FERENCES AND NOTES

1. H. Cao, X. Tang, H. Tang, Y. Yuan, J. Wu, Chem Catal. 1, 523–598 (2021). 2. H. Yi et al., Chem. Rev. 117, 9016–9085 (2017). 3. L. Capaldo, D. Ravelli, M. Fagnoni, Chem. Rev. 122, 1875–1924 (2022). 4. Z. Zhang, P. Chen, G. Liu, Chem. Soc. Rev. 51, 1640–1658 (2022). 5. J. F. Hartwig, Acc. Chem. Res. 45, 864–873 (2012). 6. J. Wencel-Delord, F. Glorius, Nat. Chem. 5, 369–375 (2013). 7. J.-J. Li et al., Chem 9, 1452–1463 (2023). 8. F. Berger et al., Nature 567, 223–228 (2019). 9. B. Liu, A. M. Romine, C. Z. Rubel, K. M. Engle, B.-F. Shi, Chem. Rev. 121, 14957–15074 (2021). 10. J. F. Hartwig, M. A. Larsen, ACS Cent. Sci. 2, 281–292 (2016). 11. R. Oeschger et al., Science 368, 736–741 (2020). 12. K. Liao, S. Negretti, D. G. Musaev, J. Bacsa, H. M. L. Davies, Nature 533, 230–234 (2016). 13. K. Liao et al., Nature 551, 609–613 (2017). 14. K. Liao et al., Nat. Chem. 10, 1048–1055 (2018). 15. D. H. Ripin, D. Evans, pKa’s of inorganic and oxo-acids (provided by ACS Organic Division, updated 4 April 2022); https://organicchemistrydata.org/hansreich/resources/pka/ pka_data/evans_pKa_table.pdf. 16. E. T. Denisov, V. E. Tumanov, Russ. Chem. Rev. 74, 825–858 (2005). 17. S. J. Blanksby, G. B. Ellison, Acc. Chem. Res. 36, 255–263 (2003). 18. L. M. Stateman, K. M. Nakafuku, D. A. Nagib, Synthesis 50, 1569–1586 (2018). 19. Y. Li, M. Lei, L. Gong, Nat. Catal. 2, 1016–1026 (2019). 20. I. B. Perry et al., Nature 560, 70–75 (2018). 21. G. Laudadio et al., Science 369, 92–96 (2020). 22. K. Chen et al., Nature 620, 1007–1012 (2023). 23. H. Cao et al., Nat. Commun. 11, 1956 (2020). 24. X.-Z. Fan et al., Angew. Chem. Int. Ed. 57, 8514–8518 (2018). 25. E. Tsui, H. Wang, R. R. Knowles, Chem. Sci. 11, 11124–11141 (2020). 26. A. Hu, J. J. Guo, H. Pan, Z. Zuo, Science 361, 668–672 (2018). 27. Q. An et al., J. Am. Chem. Soc. 142, 6216–6226 (2020). 28. Q. An et al., J. Am. Chem. Soc. 145, 359–376 (2023). 29. H. Cao et al., Nat. Synth. 1, 794–803 (2022). 30. C. Le, Y. Liang, R. W. Evans, X. Li, D. W. C. MacMillan, Nature 547, 79–83 (2017). 31. M. A. Ashley et al., Angew. Chem. Int. Ed. 58, 4002–4006 (2019). 32. J. Li et al., Nature 574, 516–521 (2019). 33. J. D. Cuthbertson, D. W. C. MacMillan, Nature 519, 74–77 (2015). 34. S. Bonciolini, T. Noël, L. Capaldo, Eur. J. Org. Chem. 2022, e202200417 (2022).

2 February 2024

35. H. P. Deng, Q. Zhou, J. Wu, Angew. Chem. Int. Ed. 57, 12661–12665 (2018). 36. S. K. Kariofillis, A. G. Doyle, Acc. Chem. Res. 54, 988–1000 (2021). 37. S. M. Treacy, T. Rovis, J. Am. Chem. Soc. 143, 2729–2735 (2021). 38. Y. Jin et al., Green Chem. 23, 6984–6989 (2021). 39. Q. Yang et al., Science 372, 847–852 (2021). 40. J. C. K. Chu, T. Rovis, Nature 539, 272–275 (2016). 41. G. J. Choi, Q. Zhu, D. C. Miller, C. J. Gu, R. R. Knowles, Nature 539, 268–271 (2016). 42. T. J. Fazekas et al., Science 375, 545–550 (2022). 43. C. Shu, A. Noble, V. K. Aggarwal, Nature 586, 714–719 (2020). 44. J.-L. Tu, A.-M. Hu, L. Guo, W. Xia, J. Am. Chem. Soc. 145, 7600–7611 (2023). 45. R. Sang et al., J. Am. Chem. Soc. 145, 15207–15217 (2023). 46. C. Yin, M. Wang, Z. Cai, B. Yuan, P. Hu, Synthesis (Stuttg.) 54, 4864–4882 (2022). 47. A. C. Ross, K. J. Go, J. G. Heider, G. H. Rothblat, J. Biol. Chem. 259, 815–819 (1984). 48. J. Li, S. Qu, W. Zhao, Angew. Chem. Int. Ed. 59, 2360–2364 (2020). 49. V. Fasano, N. Winter, A. Noble, V. K. Aggarwal, Angew. Chem. Int. Ed. 59, 8502–8506 (2020). 50. A. Fawcett et al., Science 357, 283–286 (2017). 51. J. Wu, L. He, A. Noble, V. K. Aggarwal, J. Am. Chem. Soc. 140, 10700–10704 (2018). 52. Y. Cheng, C. Mück-Lichtenfeld, A. Studer, J. Am. Chem. Soc. 140, 6221–6225 (2018). 53. F. Takahashi, K. Nogi, H. Yorimitsu, Eur. J. Org. Chem. 2020, 3009–3012 (2020). AC KNOWLED GME NTS

We thank X. Zhao and W. Liu for helpful suggestions. We thank S. Yan (Instrumental Analysis and Research Center, Sun Yat-sen University) for assistance with HR GC-MS. We thank the reviewers for helpful suggestions. Funding: This research was supported by National Natural Science Foundation of China grant 21821003, Guangdong Science and Technology Department grant 2019QN01L151 (P.H.), a Sun Yat-sen University start-up grant (P.H.), and the GBRCE for Functional Molecular Engineering. Author contributions: Conceptualization: P.H., M.W., and Y.H. Methodology: M.W., Y.H., and P.H. Investigation: M.W. Funding acquisition: P.H. Project administration: P.H. and M.W. Supervision: P.H. Writing – original draft: M.W. and P.H. Writing – review and editing: M.W., Y.H., and P.H. Competing interests: The authors declare that they have no competing interests. Data and materials availability: All data are available in the main text or the supplementary materials. 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.adj9258 Materials and Methods Supplementary Text Figs. S1 to S21 Tables S1 to S21 References (54–63) Submitted 24 July 2023; accepted 2 January 2024 10.1126/science.adj9258

8 of 8

RES EARCH

MATERIALS SCIENCE

Accessing pluripotent materials through tempering of dynamic covalent polymer networks Nicholas R. Boynton1, Joseph M. Dennis2, Neil D. Dolinski1, Charlie A. Lindberg1, Anthony P. Kotula3, Garrett L. Grocke1, Stephanie L. Vivod4, Joseph L. Lenhart2, Shrayesh N. Patel1,5*, Stuart J. Rowan1,5,6* Pluripotency, which is defined as a system not fixed as to its developmental potentialities, is typically associated with biology and stem cells. Inspired by this concept, we report synthetic polymers that act as a single “pluripotent” feedstock and can be differentiated into a range of materials that exhibit different mechanical properties, from hard and brittle to soft and extensible. To achieve this, we have exploited dynamic covalent networks that contain labile, dynamic thia-Michael bonds, whose extent of bonding can be thermally modulated and retained through tempering, akin to the process used in metallurgy. In addition, we show that the shape memory behavior of these materials can be tailored through tempering and that these materials can be patterned to spatially control mechanical properties.

T

raditionally, polymer researchers have developed materials through iterative optimization of macromolecular structure [for example, (co)monomer type and sequence (1), architecture (2), and molecular weight (3)] to maximize specific performance metrics for a given application. This approach has worked well and led to the development of many different polymers that span a wide breadth of material properties. Inspired by pluripotent stem cells that have the ability to differentiate into various cell types (4–6), we have been interested in taking a different approach to polymer design by creating a pluripotent or “stem” polymer that has the ability to be differentiated into a range of different materials with disparate properties through exposure to an external stimuli. Such single-feedstock polymers could offer substantial advantages, for example, in resource-scarce areas (at sea, in space, or on the battlefield) where a single material could meet an evolving, complex set of demands and applications. Although chemical modification has been the main tool to alter polymer properties (7), a postsynthetic route to modify a single feedstock by using environmental or processing cues would be ideal for such resource-scarce areas. Perhaps the most accessible stimulus in most environments is heat, which has been used by materials scientists for centuries. For example, metallurgists use tempering—the isothermal heating of a material at a given tem1

Pritzker School of Molecular Engineering, University of Chicago, Chicago, IL 60637, USA. 2Sciences of Extreme Materials Division, Polymers Branch, US DEVCOM Army Research Laboratory, Aberdeen Proving Ground, MD 21005, USA. 3Materials Science and Engineering Division, National Institutes of Standards and Technology (NIST), Gaithersburg, MD 20899, USA. 4 NASA Glenn Research Center, Cleveland, OH 44135, USA. 5 Center for Molecular Engineering, Argonne National Laboratory, Lemont, IL 60439, USA. 6Department of Chemistry, University of Chicago, Chicago, IL 60637, USA. *Corresponding author. Email: [email protected] (S.J.R.); [email protected] (S.N.P.)

Boynton et al., Science 383, 545–551 (2024)

perature below a critical point (such as the melting point) before rapidly quenching (8, 9)— to expand the application of steel from knives to structural beams. Inspired by this, we wondered whether it would be possible to design a single polymeric material that could access a range of mechanical properties by simply tempering that material at different temperatures. To design such a material it was proposed to leverage the following three criteria: (i) reconfigurable bonds that allow for bulk property manipulation, (ii) sensitivity to accessible and tunable stimuli conditions to allow access to a range of material properties, and (iii) an ability to retain properties within the operational temperature window of interest (Fig. 1A). A wide range of reconfigurable dynamic bonds are accessible, which include supramolecular interactions (10, 11) and dynamic covalent chemistries (12–16) such as Diels-Alder (17), boronic acids (18), hindered ureas (19), urethanes (20), disulfides (21), and thiol-ene derivatives (22). Within polymeric materials, this array of dynamic chemistries has been exploited to alter the adaptive properties of the material and allow them to be used in applications such as nanogels (23), solid polymer electrolytes (24), and stress-adaptive suspensions (25). The reversibility of dynamic bonds under different environmental stimuli satisfies the first criterion. However, particular emphasis must be placed on materials that can be reconfigured at low processing temperatures ( 0.05, ***P < 0.001, binomial test. (E) Field selectivity is >1 during Light but not Control Sloin et al., Science 383, 551–558 (2024)

2 February 2024

120

***

n= 165 0.95ns

1.05***

Control trials

Light trials

1.4 1.2 1 0.8 0.6

0 No synthetic precession (n=173)

J Spike frequency relative to LFP

0

Spatial Temporal Either

trials. Here and in (J), ***P < 0.001, Wilcoxon’s test. (F) Focal activation induces synthetic precession. Top, example Control and Light trials. Middle, mean ± SEM firing rate. Bottom, theta phase versus animal position at each spike. **P < 0.01, permutation test. Inset, Spectrum of the observed spike train and of phase-randomized spikes. ns: P > 0.05, ***P < 0.001, constrained randomization test. (G) Illumination induces precession in one-third of the artificial fields. (H) Field selectivity in artificial fields. ns, P > 0.05, U test. (I) More than one-third of rate-amplified PYRs precess during Light trials. Horizontal lines show chance levels. Error bars indicate SEM. (J) Rate-amplified PYRs exhibit higher spike frequency during Light trials. 3 of 7

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

A

Synthetic

10 ms

Mean firing rate [spk/s]

5 0 Slope= -0.011** cyc/cm ηs= 1.18 Span= 0.52 cyc/field p y

4π 3π 2π π 40

*** 0.78 cyc

1

80 Position [cm]

Slope= -0.056** cyc/cm ηs= 1.37 y Span = 1.02 cyc/field

4π 3π 2π π 40

Whole field fit

0.5 n= 341 Spontaneous fields

Theta phase [rad]

80 Position [cm]

120

n= 49 Artificial fields

Whole field fit

Half-field fits

Half-field fits

4π n= 485 spk Slope= -0.011** cyc/cm ηs= 1.18

3π 2π π

n= 201 spk Slope= -0.027** cyc/cm ηs=1.23 n= 284 spk Slope= -0.0038ns cyc/cm ηs= 1.01

n= 52 spk Slope= -0.019ns cyc/cm ηs= 1.04

n= 72 spk Slope= -0.056** cyc/cm ηs= 1.37

3π 2π

n= 20 spk Slope= -0.064ns cyc/cm ηs= 1.31

π 0

0 35

Position [cm]

80

35

Position [cm]

80

107

Position [cm]

143 107

Position [cm]

143

0.5 1 2 Span [cyc/field]

D Precession slope

F Effect size

E

Spontaneous *** -0.02 -0.0078 cyc/cm cyc/cm

0.0003 cyc/cm

-0.019 cyc/cm

1

0.41*** *** 0.94ns

1.19

Fraction of fields

0

0

0.5

-0.1

-0.1 n =341 First half-field

n= 35 Second half-field

Synthetic ns 1.04

1

0.91***

ns

0.97ns

1.09

0.1

0.1

G

Spontaneous ***

Synthetic ns

First half-field

0 0.01

Second half-field

ηs

Fraction of fields

0

C

ns 0.96 cyc

0

Precession slope [cyc/cm]

2

120

4π

0.25

4

0

0 ns

B

6

Sloin et al., Science 383, 551–558 (2024)

2

1

1

n= 341 n= 35 0.1 1 10 Slope size ratio

Fig. 3. Synthetic precession spans a complete theta cycle and does not slow down or reverse direction. (A) Left, example CA1 PYR with spontaneous precession. Right, example PYR with synthetic precession. All conventions are the same as in Fig. 2F. (B) Synthetic precession span is not different from a full cycle. Spontaneous precession spans more than half a cycle. ns: P > 0.05, ***P < 0.001, Wilcoxon’s test comparing with a null of 1 and 0.5 for synthetic and spontaneous precession, respectively. (C) In the example PYRs, the first half-field exhibits higher slope and effect size during spontaneous but not synthetic precession. Gray and blue patches show the running mean ± SD

slowed down and diminished in the second half of the place field but synthetic precession did not (Fig. 3C). Overall, spontaneous precession slopes were more negative in the first half-field compared with the second (P = 5.4 × 10−11, Wilcoxon’s test; Fig. 3D, left). By contrast, synthetic precession slopes did not slow down and were –0.019 (IQR –0.047 to 0.0044) cycles/cm in the second half-field (Fig. 3D, right). Therefore, the ratio between the slope sizes in the second and first half-fields was smaller in spontaneous compared with synthetic precession (P = 0.0046, U test; Fig. 3E). Similar results were obtained for alternative definitions of place field boundaries (fig. S12). The difference between slope size ratios illustrates that spontaneous precession slows down, whereas synthetic precession does not.

2

100

n= 341 0.5

First half-field

0.5 Second half-field

n= 35 First half-field

Second half-field

0.5

0 0.25

n= 341 n= 35 1 ηs ratio

4

theta phase of spikes as a function of position. Black lines indicate circular-linear fits. (D) Precession slope in the two halves of spontaneous and synthetic fields. Here and in (F), ns: P > 0.05, ***P < 0.001, Wilcoxon’s test. (E) Precession slope in the second half-field divided by the first. Values 0.05, ***P < 0.001, Wilcoxon’s test compared with a unity null. (F) Effect size is smaller in the second half-field of spontaneous but not synthetic precession. (G) Spatial effect size ratio, the effect size for the second half-field divided by the effect size for the first half-field.

Both the excitation-inhibition summation and somato-dendritic competition models predict that precession will diminish in the second half-field (31, 34, 51) (Fig. 1, D and E, and figs. S3, C and E, and S4D). Whereas spontaneous fields indeed displayed smaller precession effect size in the second half-field (P = 2.8 × 10−15, Wilcoxon’s test), effect size did not decrease in the second half of artificial fields (P = 0.6, Wilcoxon’s test; Fig. 3F). Accordingly, the ratio between precession effect sizes in the second and first half-fields was different from a value of one for spontaneous precession (P = 1.7 × 10−10, Wilcoxon’s test) but not for synthetic precession (P = 0.5; Fig. 3G). Thus, precession slope and effect size do not diminish in the second half of the artificial field, at odds with inhibition-

2 February 2024

1.04 Fraction of fields

0.25 ms

10

Theta phase [rad]

100 μV

Theta phase [rad]

Mean firing rate [spk/s]

Spontaneous

excitation summation (Fig. 1D and fig. S3) and somato-dendritic competition models (Fig. 1E and fig. S4). Focal pyramidal cell activation slows preexisting spontaneous precession

Although artificial fields are not unmasked subthreshold fields (Fig. 2E and fig. S7), the synthetic precession observed during local activation of PYRs could be generated locally or inherited from an upstream source (Fig. 1A). To assess whether the generator of spontaneous precession resides in CA1, we investigated the effects of local activation on place cells with preexisting precession within the illuminated region (Fig. 4A). A priori, if precession is inherited (e.g., from CA3/EC), then local CA1 activation may reduce precession prevalence 4 of 7

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

Control

Inherited generator

CA3

Theta phase

ηLight < ηControl

ηLight < ηControl

Position

Temporal ***

0.8 199/279

No preexisting precession 42% (165)

Only temporal 29% (114)

Position

D

C Preexisting precession Spatial ***

Both 6% (62)

|SlopeLight| < |SlopeControl|

|SlopeLight| ≈ |SlopeControl|

Position

***

**

Either *** 338/408

4

1.6

1.7

1.4

4

1.53

222/313 133/225

61/111 0.6

78/176

2 ηs

Precession in Light trials

CA3

CA3 PG

|SlopeControl| ηControl

1

Only spatial 13% (49)

CA1 PG

CA1

CA1

tivation on precession speed is at odds with inheritance models (Figs. 1A and 4A, center), suggesting that the generator underlying spontaneous precession resides in CA1.

amplified B Rate n=390 PYRs

Local generator

0.4

ηt

A

2

1

0.2 n= 111 PYRs

0

*** -0.018*** cyc/cm

0

-0.02

-0.04

-0.06

1.12***

n= 111 PYRs Control trials

F

n= 176 PYRs Control trials

Light trials

*** 1

1.4 1.2 1 0.8 0.6

Light trials

1.07***

Light trials

Fraction of PYRs

-0.021*** cyc/cm

Spike frequency relative to LFP

E

Control trials

Rate amplified: Light vs. Control

**

Precession slope [cyc/cm]

1

0.5 Spontaneous: odd vs. even

n= 176 PYRs

Preexisting precession 0.93*** (176)

No preexisting precession 1.1*** (165)

0.5

0

Control trials

Light trials

0.5 1 2 Light/Control spike frequency ratio

Fig. 4. Focal pyramidal cell activation slows preexisting spontaneous precession. (A) Possible outcomes of activating PYRs with preexisting precession based on inheritance (e.g., from CA3) as opposed to local precession generation (PG). (B) Spontaneously active PYRs that exhibit a light-dependent rate increase include units with and without preexisting precession. (C) Illumination reduces precession prevalence among PYRs with preexisting precession. Rate-amplified Light versus Control, precession during Light versus Control for rate-amplified PYRs with preexisting precession. Spontaneous odd versus even, precession during odd versus even trials within spontaneous place fields. Top horizontal lines, ***P < 0.001, G-test. (D) Focal activation reduces the effect size of preexisting precession. Here and in (E), top horizontal lines, **P < 0.01, ***P < 0.001, Wilcoxon’s test. (E) Activation slows preexisting precession during Light compared with Control trials. Left, precession slope size is smaller. ***P < 0.001, Wilcoxon’s test compared with a zero null (one-tailed). Right, spike frequency relative to LFP is smaller. Here and in (F), ***P < 0.001, Wilcoxon’s test compared with unity null (one-tailed). (F) Activation exerts opposite effects on spike frequency of precessing (solid line) and nonprecessing rate-amplified PYRs (dashed line). Top horizontal line, ***P < 0.001, U test.

and effect size but would not modify precession speed, because the precession generator itself is not affected (Fig. 4A, middle). By contrast, if precession originates in CA1, then local activation may modify precession speed by directly biasing the generator (Fig. 4A, right). We found that 225 of 390 (58%) rate-amplified PYRs exhibited preexisting precession within light limits (Fig. 4B). The prevalence of precession was reduced during Light compared with Control trials (133 of 225; 59%; Fig. 4C, green), beyond the reduction observed in spontaneous place fields during no-light odd- versus even-numbered trials (338 of 408; 83%; P < 1.1 × 10−16, G-test; Fig. 4C, gray). Furthermore, preSloin et al., Science 383, 551–558 (2024)

cession effect sizes were lower during Light compared with Control trials (P < 9.1 × 10−3; Wilcoxon’s test; Fig. 4D). Activation reduced spatial precession slope (P = 0.0084, Wilcoxon’s test) and spike frequency relative to the LFP (P = 2.2 × 10−10, Wilcoxon’s test; Fig. 4E). Our findings (Fig. 4, D and E) contrast with increased spike frequency among rate-amplified PYRs with no preexisting precession (Fig. 2J). Specifically, activation increased spike frequency in PYRs without preexisting precession (P = 0.018, Wilcoxon’s test) but decreased spike frequency in PYRs with preexisting precession (P = 3.9 × 10−7), exerting opposite effects (P = 6.9 × 10−12, U test; Fig. 4F). The impact of ac-

2 February 2024

Focal PYR activation induces a rate increase without precession in CA1 INTs and parietal cortex PYRs

Various factors within CA1 may contribute to precession generation, including dynamic reduction of inhibition (47, 50, 52) and local circuit PYR-INT interactions (16, 36). Consistent with the circuit models, CA1 INTs may exhibit spontaneous precession (53, 54), and silencing INTs affects spontaneous precession (41, 55). We therefore investigated whether focal PYR activation induces synthetic precession in local circuit INTs. A total of 85 of 158 (54%) local INTs recorded on the illuminated shank exhibited a position-dependent firing rate increase (Fig. 5, A and B; fig. S13, B and C; and table S4). Only three of 158 (2%) INTs exhibited artificial fields (P = 0.99, binomial test; Fig. 5C and fig. S13, D and E), precluding the characterization of synthetic precession among INTs. A total of 68 local INTs were rate amplified (fig. S13F) and did not exhibit preexisting precession (Fig. 5D), of which only five (7%) exhibited precession during Light trials (P = 0.8, binomial test; Fig. 5E). Spike frequency relative to LFP of CA1 INT did not change between Control and Light trials (P = 0.78, Wilcoxon’s test; Fig. 5F). Likewise, nonlocal INTs exhibited increased firing rates but no precession (fig. S13, G to J). Thus, whereas a position-dependent rate increase is transferred from PYRs to INTs, synthetic precession does not necessarily involve precession in local INTs. In CA1, an imposed excitation of PYRs was sufficient to generate precession. To investigate whether a rate increase combined with theta may suffice for synthetic precession, we repeated the imposed excitation experiment in the posterior parietal cortex (PPC; n = 22 sessions in two CaMKII::ChR2 mice; Fig. 5G; fig. S14, A to C; and table S5). Hippocampal theta oscillations are volume conducted to the PPC, where neurons are phase locked to specific theta phases (56). We found that PPC neurons exhibited rate-modulated spatial activity (57) (fig. S15, C and D) and phase locking to specific theta phases (fig. S15E) but no spontaneous theta-phase precession (fig. S15G, left). Focal PYR activation in the PPC induced position-dependent firing rate increases (Fig. 5H), with 52 of 195 (27%) PPC PYRs showing artificial place fields (P < 1.11 × 10−16, binomial test; Fig. 5I). PPC PYRs exhibited theta-phase locking in artificial fields (fig. S15F), but phase precession was observed at chance level (P = 0.76, binomial test; Fig. 5J and fig. S15G). Among rate-amplified PPC PYRs (Fig. 5K), precession 5 of 7

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

B 10

2π 0

Rate amplified 52% (82)***

Light trials

4π

0

40

80 Position [cm]

0.2 5 ns 0.1

0

75 Position [cm]

H

20

0.98ns

0.98ns

Control trials

Light trials

1.4 1.2 1 0.8

Illumination

Rate difference (Light - Control)

10 160

0 4π

Control trials

2π

120

80

0 Light trials 40

2π 0

0 40

0

J

K

n= 52 PYRs

75 Position [cm]

120

80 Position [cm]

Artificial fields

n= 195 PYRs

L

Rate amplified

150

No preexisting precession

n= 107 PYRs

75 150 0.1 1 10 Difference Position [cm] [spk/s]

0.1 1 10 0 Power [µW]

M

ns

n= 95 PYRs 1.15**

No preexsting precession 89% (95)

No synthetic precession 92% (48)

Artificial fields 27% (52)***

Synthetic precession 8% (4)ns

N

Only spatial Both Only temporal 5% (5) 1% (1) 6% (6)

0.2

Spike frequency relative to LFP

Unchanged 18% (36)

Fraction of precession

0.3

Rate amplified 55% (107)***

150 0.1 1 10 Difference [spk/s]

ns

Light trials

PPC PYRs

Mean firing rate [spk/s] Theta phase [rad] Theta phase [rad]

0

0.6 Control trials

I Firing rate changes

F

n= 68 CA1 INTs

0

Both 1% (1)

4π

10 ms

150 0.1 1 10 Power [µW]

75 Position [cm]

No preexisting precession 0.3

No preexisting precession 83% (68)

Decreasing Artificial 2% (3)ns fields 2% (3)ns

100 μV 0.25 ms

40

120

E

Rate amplified n= 82 CA1 INTs

Only temporal 16% (13)

PPC

80

0 0

Unchanged 44% (70)

G

120

2π

D

Firing rate changes n= 158 CA1 INTs

Local-shank CA1 INTs

0 4π Control trials

100 μV 0.25 ms

C

Rate difference (Light - Control)

20

Spike frequency relative to LFP

10 ms

PYR illumination

30

Precession in Light trials

Theta phase [rad]

Theta phase [rad]

CA1: indirectly activated local INTs

Mean firing rate [spk/s]

A

12 ns

0.1

0 0 Control trials

1.05ns

1.4 1.2 1 0.8 0.6 Control trials

Light trials

Light trials

Rate amplified

Artificial fields

1

1

1

1

*** 52/195 ***

286/1,095 ***

*** 0.5 105/286 *** 4/52 ns

3/158 ns CA1 PYR

Local CA1 INT

occurred in 12 of 95 (13%) PYRs, not above chance (P = 0.21, binomial test; Fig. 5L). Furthermore, spike frequency relative to LFP of rate-amplified PPC PYRs did not increase between the Control and Light trials (P = 0.11, Sloin et al., Science 383, 551–558 (2024)

CA1 PYR

PPC PYR

107/195 ***

0.5 390/1,095 ***

0

0

0

82/158 ***

Local CA1 PPC PYR INT

*** *** 0.5

63/165 ***

ns 5/68 ns

12/95 ns

0 CA1 PYR

Local CA1 INT

PPC PYR

Wilcoxon’s test; Fig. 5M). Together, these findings (Fig. 5N) suggest that synthetic precession requires synaptic theta or depends on cellular network properties found in CA1 but not in the PPC.

2 February 2024

Fraction of precession

***

0.5

ns

*** Fraction of units

ns

Fraction of precession

*** Fraction of units

Fig. 5. Focal PYR activation induces a rate increase without precession in CA1 INTs and parietal cortex PYRs. (A) Position-dependent CA1 PYR illumination indirectly activates local INTs. (B) PYR activation (same trials as in Fig. 2C) increases firing rates of local INTs recorded on the illuminated shank. (C) PYR activation does not generate artificial fields in local INTs. (D) Precession prevalence during Control trials in rate-amplified INTs. (E) Rate-amplified INTs do not precess during Light trials. Here and in (L), ns: P > 0.0975, binomial test. (F) Local INT spike frequency relative to LFP is not higher during Light compared with Control trials. (G) In the PPC of CaMKII::ChR2 mice, focal closed-loop illumination activates PPC PYRs. (H) Illumination activates local PYRs. (I) Activation induces artificial fields in PPC PYRs. (J) PPC PYRs with artificial fields do not exhibit synthetic precession. (K) Precession prevalence during Control trials in rate-amplified PPC PYRs. (L) PPC PYR activation does not affect precession prevalence in rate-amplified PYRs. (M) PPC PYR spike frequency relative to LFP is not increased during Light trials. (N) Synthetic precession occurs in lightactivated CA1 PYRs but not in indirectly activated CA1 INTs or in light-activated PPC PYRs. ns: P > 0.05, ***P < 0.001, binomial test. Top horitontal line, ns: P > 0.05, *P < 0.05, **P < 0.01, ***P < 0.001, G-test.

CA1 PYR

Local CA1 INT

PPC PYR

Discussion

Imposing artificial place fields on CA1 PYRs generated synthetic precession, which is incompatible with spreading activation and dual-input models. Synthetic precession spanned a full 6 of 7

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

cycle of precession and did not diminish or reverse in the second half of the field, inconsistent with the inhibition-excitation summation and somato-dendritic competition models. Local activation slowed preexisting precession, suggesting that the generator of spontaneous precession is not inherited but rather resides in CA1. The synthetic precession mechanism did not involve precession of local circuit CA1 INTs but required properties absent in the PPC, e.g., synaptic arrival of theta, the circuit architecture of CA1, or biophysics specific to CA1 PYRs. Among the models proposed for precession generation, our results are consistent with dualoscillator models (1, 35–37), which require theta and faster oscillations. PYR membrane potentials may demonstrate faster-than-theta oscillations caused by interacting amplifying and resonant (e.g., persistent sodium and slowly activating potassium) currents. Alternatively, a PYR may exhibit fast oscillations arising from local circuit interactions (38, 39) or inherited from nearby INTs (16). Synthetic precession may originate from an as-yet-unidentified mechanism that meets the following constraints: rate dependence, completion of a full cycle, independence of local INT precession, and CA1 specificity. Although the processes underlying spontaneous and synthetic precession may differ, our findings constrain the mechanisms that may be supported by the hippocampal network. The existence of synthetic precession indicates that if ongoing theta (43, 58) and inputs from upstream regions (24, 28, 59, 60) remain intact, then local excitation of CA1 circuity suffices for generating precession, transforming a rate code into a temporal code. Synthetic precession carries implications for the functional role of precession in generating theta sequences and learning (22, 23). Whereas precession has been proposed to generate theta sequences (2, 16–18), others have suggested that both precession and sequences arise from the preconfigured connectivity of PYRs (29, 61, 62). We found that artificial field appearance could be predicted from the response to light off the linear track, and that precession could be generated by activating arbitrary CA1 PYRs at random positions along the track. Because the activated PYRs are not expected to have prior synaptic connections with one another, preconfigured connectivity is unlikely to suffice for generating synthetic precession. Synthetic precession may therefore

Sloin et al., Science 383, 551–558 (2024)

generate theta sequences between previously unassociated PYRs, potentially facilitating the learning of new sequences (18, 63). RE FERENCES AND NOTES

1. J. O’Keefe, M. L. Recce, Hippocampus 3, 317–330 (1993). 2. W. E. Skaggs, B. L. McNaughton, M. A. Wilson, C. A. Barnes, Hippocampus 6, 149–172 (1996). 3. T. Hafting, M. Fyhn, T. Bonnevie, M. B. Moser, E. I. Moser, Nature 453, 1248–1252 (2008). 4. M. W. Jones, M. A. Wilson, Hippocampus 15, 867–873 (2005). 5. M. A. A. van der Meer, A. D. Redish, J. Neurosci. 31, 2843–2854 (2011). 6. D. Bush, H. F. Ólafsdóttir, C. Barry, N. Burgess, Curr. Biol. 32, 64–73.e5 (2022). 7. T. Eliav et al., Cell 175, 1119–1130.e15 (2018). 8. O. Jensen, J. E. Lisman, J. Neurophysiol. 83, 2602–2609 (2000). 9. J. Huxter, N. Burgess, J. O’Keefe, Nature 425, 828–832 (2003). 10. A. P. Maurer, A. W. Lester, S. N. Burke, J. J. Ferng, C. A. Barnes, J. Neurosci. 34, 15022–15031 (2014). 11. H. Hirase, A. Czurkó, J. Csicsvari, G. Buzsáki, Eur. J. Neurosci. 11, 4373–4380 (1999). 12. E. Pastalkova, V. Itskov, A. Amarasingham, G. Buzsáki, Science 321, 1322–1327 (2008). 13. A. Shimbo, E. I. Izawa, S. Fujisawa, Sci. Adv. 7, eabd7013 (2021). 14. N. T. M. Robinson et al., Neuron 94, 677–688.e6 (2017). 15. S. Terada, Y. Sakurai, H. Nakahara, S. Fujisawa, Neuron 94, 1248–1262.e4 (2017). 16. A. Chadwick, M. C. van Rossum, M. F. Nolan, eLife 5, e20349 (2016). 17. E. T. Reifenstein, I. Bin Khalid, R. Kempter, eLife 10, e67171 (2021). 18. T. M. George, W. de Cothi, K. L. Stachenfeld, C. Barry, eLife 12, e80663 (2023). 19. G. Dragoi, G. Buzsáki, Neuron 50, 145–157 (2006). 20. D. J. Foster, M. A. Wilson, Nature 440, 680–683 (2006). 21. C. Drieu, R. Todorova, M. Zugaro, Science 362, 675–679 (2018). 22. J. Jaramillo, R. Kempter, Curr. Opin. Neurobiol. 43, 130–138 (2017). 23. C. Drieu, M. Zugaro, Front. Cell. Neurosci. 13, 232 (2019). 24. M. B. Zugaro, L. Monconduit, G. Buzsáki, Nat. Neurosci. 8, 67–71 (2005). 25. J. Jaramillo, R. Schmidt, R. Kempter, J. Neurosci. 34, 7715–7731 (2014). 26. N. Burgess, C. Barry, J. O’Keefe, Hippocampus 17, 801–812 (2007). 27. F. S. Chance, J. Neurosci. 32, 16693–703a (2012). 28. A. Fernández-Ruiz et al., Neuron 93, 1213–1226.e5 (2017). 29. O. Jensen, J. E. Lisman, Learn. Mem. 3, 279–287 (1996). 30. M. V. Tsodyks, W. E. Skaggs, T. J. Sejnowski, B. L. McNaughton, Hippocampus 6, 271–280 (1996). 31. M. R. Mehta, A. K. Lee, M. A. Wilson, Nature 417, 741–746 (2002). 32. A. Kamondi, L. Acsády, X. J. Wang, G. Buzsáki, Hippocampus 8, 244–261 (1998). 33. J. C. Magee, J. Neurophysiol. 86, 528–532 (2001). 34. K. D. Harris et al., Nature 417, 738–741 (2002). 35. A. Bose, V. Booth, M. Recce, J. Comput. Neurosci. 9, 5–30 (2000). 36. M. Lengyel, Z. Szatmáry, P. Érdi, Hippocampus 13, 700–714 (2003). 37. L. S. Leung, J. Neurosci. 31, 12282–12296 (2011). 38. H. R. Wilson, J. D. Cowan, Biophys. J. 12, 1–24 (1972). 39. E. Stark et al., Neuron 80, 1263–1276 (2013). 40. C. D. Harvey, F. Collman, D. A. Dombeck, D. W. Tank, Nature 461, 941–946 (2009). 41. S. Royer et al., Nat. Neurosci. 15, 769–775 (2012). 42. P. Ravassard et al., Science 340, 1342–1346 (2013). 43. I. Zutshi et al., Curr. Biol. 28, 1179–1188.e3 (2018).

2 February 2024

44. Y. Yamaguchi, Biol. Cybern. 89, 1–9 (2003). 45. M. Wang, D. J. Foster, B. E. Pfeiffer, Science 370, 247–250 (2020). 46. N. T. M. Robinson et al., Cell 183, 1586–1599.e10 (2020). 47. T. Geiller et al., Nature 601, 105–109 (2022). 48. D. Lee, B. J. Lin, A. K. Lee, Science 337, 849–853 (2012). 49. S. McKenzie et al., Neuron 109, 1040–1054.e7 (2021). 50. M. Valero, I. Zutshi, E. Yoon, G. Buzsáki, Science 375, 570–574 (2022). 51. Y. Yamaguchi, Y. Aota, B. L. McNaughton, P. Lipa, J. Neurophysiol. 87, 2629–2642 (2002). 52. M. Valero, A. Navas-Olive, L. M. de la Prida, G. Buzsáki, Cell Rep. 40, 111232 (2022). 53. A. P. Maurer, S. L. Cowen, S. N. Burke, C. A. Barnes, B. L. McNaughton, J. Neurosci. 26, 13485–13492 (2006). 54. V. Ego-Stengel, M. A. Wilson, Hippocampus 17, 161–174 (2007). 55. C. Grienberger, A. D. Milstein, K. C. Bittner, S. Romani, J. C. Magee, Nat. Neurosci. 20, 417–426 (2017). 56. A. Sirota et al., Neuron 60, 683–697 (2008). 57. M. Krumin, J. J. Lee, K. D. Harris, M. Carandini, eLife 7, e42583 (2018). 58. P. C. Petersen, G. Buzsáki, Neuron 107, 731–744.e3 (2020). 59. M. I. Schlesiger et al., Nat. Neurosci. 18, 1123–1132 (2015). 60. I. Zutshi, M. Valero, A. Fernández-Ruiz, G. Buzsáki, Neuron 110, 658–673.e5 (2022). 61. K. Mizuseki, G. Buzsáki, Cell Rep. 4, 1010–1021 (2013). 62. S. J. C. Venditto, B. Le, E. L. Newman, Hippocampus 29, 1075–1090 (2019). 63. Y. Wang, S. Romani, B. Lustig, A. Leonardo, E. Pastalkova, Nat. Neurosci. 18, 282–288 (2015). 64. H. E. Sloin, L. Spivak, A. Levi, R. Gattegno, S. Someck, E. Stark, Data for: Local activation of CA1 pyramidal cells induces theta-phase precession, Dryad (2024); https://doi.org/ 10.5061/dryad.1c59zw431. 65. H. E. Sloin, L. Spivak, A. Levi, R. Gattegno, S. Someck, E. Stark, Code for: Local activation of CA1 pyramidal cells induces theta-phase precession, Zenodo (2023); https://doi.org/10.5281/ zenodo.10405865. AC KNOWLED GME NTS

We thank G. Buzsáki, K. Diba, A. Marmelshtein, M. R. Mehta, S. Palgi, H. G. Rotstein, and M. Zugaro for constructive comments. Funding: This work was supported by the United States–Israel Binational Science Foundation (BSF grant 2015577); the European Research Council (grant 679253); the Israel Science Foundation (grant 638/16); the Canadian Institutes of Health Research (CIHR), International Development Research Centre (IDRC), Israel Science Foundation (ISF), and Azrieli Foundation (grant 2558/18); and the Rosetrees Trust (grant A1576). Author contributions: Conceptualization: E.S., H.E.S.; Funding acquisition: E.S.; Investigation: H.E.S., L.S., A.L., R.G., S.S., E.S.; Methodology: H.E.S., E.S.; Project administration: E.S.; Resources: E.S.; Supervision: E.S.; Visualization: H.E.S., E.S.; Writing – original draft: H.E.S., E.S.; Writing – review and editing: H.E.S., L.S., A.L., R.G., S.S., E.S. Competing interests: The authors declare no competing interests. Data and materials availability: All data are available in the manuscript or supplementary materials or have been deposited to Dryad (64). The code for temporal phase precession is available at Zenodo (65). 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.adk2456 Materials and Methods Figs. S1 to S15 Tables S1 to S5 References (66–89) MDAR Reproducibility Checklist Submitted 10 August 2023; accepted 21 December 2023 10.1126/science.adk2456

7 of 7

WORKING LIFE By Alexandra Hogan

Baby in tow

I

t was a career dream realized: An email arrived to say I had been appointed to a World Health Organization committee on vaccine planning. I was excited; I had worked hard as an infectious disease modeler during the COVID-19 pandemic and I wanted to contribute to global vaccine policies. But when I realized what the committee membership entailed, my stomach sank. I would need to fly 24 hours from Australia to attend a 3-day meeting in Switzerland. At the time, I was on maternity leave, having given birth 6 months earlier. My life revolved around breastfeeding and round-the-clock-parenting. How would I travel all the way to Switzerland?

The way I saw it, I had two options. Either I would not attend, or I would take my baby with me. To me, it didn’t feel right to leave him at home. I didn’t want to pump breastmilk or bottle feed, and I wasn’t interested in being apart from him overnight, let alone while I traveled to the other side of the world. I didn’t know anyone who had traveled internationally for a meeting with their baby. Parents I asked said they had never even considered it. But I would have felt gutted to turn the opportunity down, so I decided to try to make it work. The meeting was small—only 15 committee members—so it didn’t have on-site day care. I had to think of other ways to care for my baby while I worked. A colleague had once attended an event with a newborn baby and nanny in tow, allowing her to participate but have her baby cared for nearby so she could nurse when required. I wondered whether that might work for me. I learned that the hotel where the meeting was being held could assist me in booking a local nanny. It wasn’t a cheap option, but I got some funding from my university to help pay for it. And I was comforted to know that I was using a vetted and experienced professional. The 24 hours of travel would have been exhausting even without a child and all the additional luggage. At one point in the flight I managed to get my baby to nap, and I immediately fell asleep—so deeply that I didn’t hear him wake up. A fellow passenger had to tap me on the shoulder to tell me my child was screaming! Thankfully, I had built in several recovery days in Switzerland before the meeting started. That allowed us to settle in at the hotel and adjust to the time zone. Then, on the first morning of the meeting, the nanny arrived at our hotel room. I had been feeling nervous about leaving my son with

someone I had never met. But she was warm, friendly, and capable. She looked after my baby in our hotel room, occasionally taking him out for walks or to meet me in the lobby so I could breastfeed. He had a few meltdowns, as babies do. But overall it went well. In the middle of discussions on vaccine modeling, my phone would vibrate as the nanny texted me pictures of my baby on a swing at a local park or having a nap. I could relax and focus on work, as I knew my child was safe and happy. Going into the meeting, I had worried that colleagues would think me unprofessional if they saw me feeding or caring for my baby. But my fears turned out to be unfounded. If I hadn’t mentioned it, few people at the meeting would have known I was traveling with my infant. When I did tell colleagues, they were encouraging and supportive and seemed impressed at my commitment to attend. That’s not to say it was easy; I was completely exhausted by the end. Normally when traveling for work, you have the evenings to wind down and relax, and have a social drink and dinner with colleagues. But at the end of each day I had to switch back to parenting mode, with no evening support. Despite the challenges, I returned home to Australia having thoroughly enjoyed the adventure with my baby. I was also proud to have done my part to show that scientist parents, ideally with the help of their employers, can think creatively and flexibly about child care when attending work meetings. Local nanny services are one option, and you may find others. It isn’t easy. But it can be done. j

“I didn’t know anyone who had traveled internationally for a meeting with their baby.”

562

2 FEBRUARY 2024 • VOL 383 ISSUE 6682

Alexandra Hogan is a research fellow at UNSW Sydney. Do you have a career story to share? Send it to [email protected].

science.org SCIENCE

Produced by the Science/AAAS Custom Publishing Office

LIFE SCIENCE TECHNOLOGIES

new p rodu cts: dna/ rna an aly sis

Expanded Genomics Services Azenta Life Sciences has expanded services to the United Kingdom, opening a new lab in Oxford as its GENEWIZ Multiomics and Synthesis Solutions site for the region. The site will feature a new state-of-the-art genomics lab scheduled to open in March 2024. Drawing upon the experience of scientists at the former Oxford Genomics Centre (OGC), the new Oxford Genomics Lab will provide genomics services from GENEWIZ to support a wide range of project needs from its customers. The new lab will be headed by Dr. David Buck, the former head of the OGC, and equipped with the latest next-generation sequencing (NGS) technology. The NGS services provided by the Oxford Genomics Lab at opening will

   
   
 
 ""

extraction, library preparation, sequencing, and bioinformatics analysis backed by experienced Ph.D.-level study managers. The   
        "!#$

collection boxes, which simplify the sample submission process. In 

  
   
 
 



  

local collection boxes in Oxford and Cambridge. Azenta Life Sciences For info: +44-208-546-0869

    
In Situ Genomics Analysis 
 


 conduct in situ assays using Xenium platform. Designed by 10x   




 includes the Xenium, a versatile instrument with high-throughput  !
   
"  and protein signals, and performing secondary analysis. With  
  #!
$
  
 


 

 % 

 
 provides comprehensive analysis, including cell segmentation and clustering, allowing users to compare gene expression across tissue microenvironments, measure transcript density, and integrate histopathologyóall on the same section. BioChain can help with tissue preparation and conduct Xenium assays to answer research ! 

  BioChain For info: 510-894-9695 www.biochain.com Topical Treatment for Retinal Vascular Disease  !





 
! 


"# ""

 study. The study found that EXN407 met all required safety and pharmacokinetic parameters and showed promising initial signals of biological activity. Having regained full rights to its ophthalmology
 



$! 
#  topical treatment for retinal vascular diseases, including diabetic retinopathy and diabetic macular oedema. Current treatment for these diseases requires patients to undergo monthly injectionsó directly into the eyeóto prevent aberrant blood vessel growth. These treatments are invasive, distressing, and burdensome to endure in the long-term. Exonateís lead candidate for an alternative, non-

invasive treatment is EXN047, a small molecule inhibitor of splice       " " (#$((%* twice daily for 3 months (around 180 total applications per patient) with either EXN407 or a placebo. ìThe results from the EXN407 trial are very encouraging, with the data validating the hypothesis "+*% ,!& % , %*% %#  -) * Exonate CEO Dr. Catherine Beech in a release, adding that Exonate  " %+  #  %   )"   upcoming phase II clinical study. The company is optimistic that " "%%%%*+ '
    , severe diabetic macular oedema, and hopes to bring a treatment to market with enough ocular permeability to use in topical eye drops and eliminate the need for unpleasant intravitreal injections. Exonate Ltd. For info: +44 (0)1223-734-710 www.exonate.com Building Blocks for Flow Chemistry %!'%'*'*()(***)+*-/1%(%*/)2.*3)(1'*55* 13'*2*/*(((*)5)*5)*25**$)8*23%'5;*';'53)(/%%5;* and performance, the modules include the Cold Coil II Flow :25)*(*5*4)5)%*25)*5%*3)(1