Vol 584, 13 August 2020 Nature

Citation preview

The international journal of science / 13 August 2020

AIDS, malaria and tuberculosis are surging

nets, and a 75% drop in access to antimalarial drugs — would lead to 769,000 malaria deaths in sub-Saharan Africa this year, a mortality level not seen in 20 years. Right now, with the total number of COVID-19 infections approaching 20 million and deaths standing at more than 700,000, we can’t say how bad the pandemic will get. But we can say that, without interventions, TB, AIDS and malaria are likely to take more lives.

Efforts to defeat the coronavirus have fuelled a rise in other infectious diseases. Urgent action is needed to avert a catastrophe.

Four proposals

A

IDS, malaria and tuberculosis (TB), three of the deadliest infectious diseases, together kill 2.4 million people every year, with TB alone responsible for 1.5 million deaths. And deaths from these diseases could almost double over the next year, according to the Global Fund to Fight AIDS, Tuberculosis and Malaria, a consortium of donors that funds treatments. The reason: coronavirus. It’s a horrifying prospect, and calls for an urgent action plan. More than three months of lockdowns have prevented many people from accessing treatments for non-COVID infectious diseases; at the same time, new cases of these illnesses will have gone undetected. Although lockdowns are easing, it will take some time for health care to get back to normal, as authorities continue to prioritize COVID-19. Taken together, this is resulting in a surge of cases. That’s why there needs to be a step change in funding for AIDS, malaria and TB prevention, treatment and research, and greater public awareness of the rising threat posed by infectious diseases. And researchers — particularly epidemiologists — must continue to refine the models that are alerting the world to this approaching catastrophe. One model, developed by researchers at the London School of Hygiene and Tropical Medicine, projects that there will be around 200,000 extra deaths from TB across China, India and South Africa between 2020 and 2024 (C. F. McQuaid et al. Eur. Respir. J. http://doi.org/d6ck; 2020). The data for AIDS and malaria are just as troubling. In 2018, nearly half a million people in sub-Saharan Africa died from AIDS-related illnesses. If access to antiretroviral therapies is disrupted for just six months, this number is projected to double in the coming year, according to modelling by the WHO and the Joint United Nations Programme on HIV/AIDS (UNAIDS) published in May, a high not seen in more than a decade. A study by researchers at Nigeria’s National Malaria Elimination Programme in Abuja and Imperial College London, published in Nature Medicine, predicts that 779,000 people are at risk of dying from malaria in sub-Saharan Africa in 2020 — more than double the number for 2019 (E. SherrardSmith et al. Nature Med. http://doi.org/d6cn; 2020). Researchers at the WHO reached a similar estimate. They modelled the impact of COVID-19 on malaria in 41 countries under 9 scenarios. The worst case — a continuing suspension of campaigns to distribute insecticide-treated bed

COVID-19 has turned the clock back years, if not decades, in the fight against infectious diseases.”

Several things must now happen. First, hospitals and health authorities in affected cities and regions must recognize that AIDS, malaria and TB are surging again. In the case of TB, case detection — which has been affected by hospital testing facilities being diverted for COVID-19 — needs to be resumed quickly. It is possible for testing facilities to be shared for the two diseases. Some hospitals in the Asia–Pacific region are using the same equipment to run COVID-19 tests in the morning and TB tests in the afternoon — or vice versa. It is also possible to coordinate COVID-19 testing with rapid diagnostic testing for HIV and malaria. Second, researchers must keep refining their models using more real-world data. Third, there is a need for public-information campaigns. Public, private and non-governmental organizations must alert people to the risks of rising levels of infectious diseases. Such campaigns will also go some way towards reassuring existing patients, as well as those who become unwell, that they need to seek — or continue — treatment. Fourth, these campaigns cannot on their own keep surgeries and wards open, or equipment functioning. The resurgence of infectious diseases has created a greater demand for tests, treatments and research. All of these need more funding. In a June report (go.nature.com/3aez6jd), the Global Fund calculated that an extra US$28.5 billion is needed to ensure that HIV, TB and malaria programmes can continue to function, and that researchers can continue to develop common diagnostic tools — especially for TB and COVID-19. And this is just for the next 12 months. The Global Fund is confident it can access $6 billion of this sum, on top of its annual spending, but it cannot raise the rest alone. Some of the fund’s largest international donors, such as the United Kingdom, are cutting back on science-aid funding. If the usual channels for fundraising seem unlikely to deliver, alternative approaches should be tried, such as public events at which governments, companies and philanthropic organizations are invited to make funding pledges. The governments of richer countries do donate at such events, as seen in May, when a live COVID-19 donor conference organized by the European Union received €6.15 billion (US$7.2 billion) in pledges. COVID-19 has turned the clock back years, if not decades, in the fight against infectious diseases. It is, of course, imperative that all measures possible are taken to protect people from coronavirus and to treat those who have become sick. But saving people from one infectious disease only to have them die of another is the last thing anyone wants.

. d e v r e s e r s t h g i r l l A . d e t i m i L e r u t a N r e g n i r p S 0 2 0 2 ©

Nature | Vol 584 | 13 August 2020 | 169

A personal take on science and society

World view

By Nisreen A. Alwan

Pandemic policy must include defining and measuring what we mean by mild infection.

E

ight months into the global pandemic, we’re still measuring its effects only in deaths. Non-hospitalized cases are loosely termed ‘mild’ and are not followed up. Recovery is implied by discharge from hospital or testing negative for the virus. Ill health in those classed as ‘recovered’ is going largely unmeasured. And, worldwide, millions of those still alive who got ill without being tested or hospitalized are simply not being counted. Previously healthy people with persistent symptoms such as chest heaviness, breathlessness, muscle pains, palpitations and fatigue, which prevent them from resuming work or physical or caring activities, are still classed under the umbrella of ‘mild COVID’. Data from a UK smartphone app for tracking symptoms suggests that at least one in ten of those reporting are ill for more than three weeks. Symptoms lasting several weeks and impairing a person’s usual function should not be called mild. Defining and measuring recovery from COVID-19 should be more sophisticated than checking for hospital discharge, or testing negative for active infection or positive for antibodies. Once recovery is defined, we can differentiate COVID that quickly goes away from the prolonged form. I had COVID symptoms of fever, cough, gastrointestinal upset, chest and leg pains in late March. But at that time, non-hospitalized patients were not tested. Since then, I have had bad days with some symptoms, then OK days, then worse days of exhaustion, making me regret what I did on the OK days, such as taking a short walk. This is a difficult time for me as a public-health academic engaged in pandemic action while struggling with this strange pattern of illness. I don’t know what it means for my long-term health, which is concerning as a mother caring for young children. One consolation is knowing that I am not alone. There are many others who have not regained their previous health, even months after their initial symptoms. Among them, fluctuating symptoms like mine are common. Although clinicians and researchers have an idea of who is at increased risk of dying from COVID, we don’t know who is more likely to experience prolonged ill health following symptomatic, or even asymptomatic, infection. The idea of accepting certain levels of infection to run through society, while protecting the vulnerable, becomes meaningless without considering health and productivity as outcomes alongside death. Research that follows COVID patients after discharge from hospital is starting. But there is still a gap in

Nisreen A. Alwan is an associate professor of public health at the University of Southampton, UK. e-mail: n.a.alwan@ soton.ac.uk

. d e v r e s e r s t h g i r l l A . d e t i m i L e r u t a N r e g n i r p S 0 2 0 2 ©

170 | Nature | Vol 584 | 13 August 2020

Once recovery is defined, we can differentiate COVID that quickly goes away from the prolonged form.”

quantifying and characterizing COVID-related illness in those not hospitalized. The consequences of failing to do so are significant. Some people, especially the young and healthy, might not see a need to follow preventive measures, because they expect only a few days of flu-like symptoms at the worst. Sick people might not get the support they need, and the true human and economic costs of the pandemic will not be correctly estimated. As long as ‘long COVID’ is labelled as anecdotal, it will not be taken seriously, and public communication will neglect it. We need to quantify it properly and accurately. We must measure recovery in those not presenting with severe disease at the outset. Let us start simple. With other common viral illnesses, such as flu, we would expect recovery to mean going back to pre-infection levels of functionality and quality of life. This means we must follow up all patients with confirmed (by test) or highly probable (by symptoms) COVID and find out whether they have returned to their previous normal within a specified time from the onset of their symptoms. The ‘recovery’ definition must include duration, severity and fluctuation of symptoms, as well as functionality and quality of life. Everyone who is symptomatic would remain a ‘case’ until they fulfilled the recovery criteria or died. This is basic bread-and-butter epidemiology. We just need to apply it to this pandemic. To do so, we must also define who had the infection in the first place. When testing is absent or inaccurate, physicians must be provided with universal and simple criteria for what constitutes clinical COVID. A good starting point are the studies characterizing typical symptoms on a population level. Measuring recovery is not an easy ask with health and surveillance systems already struggling to cope. It makes sense to set up disease registers, akin to cancer registries, to track people over time and record their condition. This could be done through quick monthly, and subsequently annual, check-ups with health-care providers. If national registers are not quickly forthcoming, local ones could be started. For surveillance, public-health agencies must prioritize agreement on criteria for a definition of recovery, and on the structures in which these criteria could be implemented. We must overlay research on surveillance with studies of the characteristics of those experiencing prolonged ill health. We must learn to identify and protect the most vulnerable. The narrow narrative of death as the only bad outcome from COVID needs broadening to include people becoming less healthy, less capable, less productive and living with more pain. For that, we’ll need better surveillance. The essential first step is getting clear and universal definitions for recovery and COVID severity.

NISREEN A. ALWAN

A negative COVID-19 test does not mean recovery

The world this week

News in brief Beirut blast among largest involving ammonium nitrate An explosion in Beirut’s port has killed at least 220 people, injured more than 5,000 and left an estimated 300,000 people homeless. Lebanese authorities say that the explosion on 4 August was caused by 2,750 tonnes of ammonium nitrate, a chemical compound commonly used as an agricultural fertilizer, which had been stored for 6 years at a port warehouse. An investigation into what triggered the explosion is under way, and early reports suggest that it was probably a nearby fire. The blast is one of the largest accidental ammonium nitrate explosions ever recorded (see ‘Explosive chemical’) — so powerful that it was heard more than 200 kilometres away in Cyprus. Ammonium nitrate is cheap and usually safe to handle, but storing it can be a problem. Over time, it absorbs moisture and forms clumps. When a large quantity of compacted ammonium nitrate is exposed to intense heat, it can trigger an explosion. The disaster comes as Lebanon is struggling to cope with the coronavirus pandemic and an economic crisis. Efforts to treat injured people have been hampered by damage to hospitals, and the explosion destroyed grain silos and much of Beirut’s port.

EXPLOSIVE CHEMICAL SOURCE: NATURE NEWS TEAM ANALYSIS; IMAGE: TOMOHIRO OHSUMI/GETTY

The Beirut blast, which killed at least 220 people and injured more than 5,000, is one of the largest industrial disasters ever linked to ammonium nitrate (NH4NO3). Brest, France 1947

3,000 tonnes NH4NO3

Texas City 1947

2,960

Beirut 2020

2,750

Tianjin, China 2015 Oppau, Germany 1921 Neyshabur, Iran 2004 Toulouse, France 2001 West, Texas 2013

800 450 400 300 240

JAPAN CONSIDERS NEW RULES ON RESEARCH INTERFERENCE AMID US–CHINA TENSIONS The Japanese government is considering tougher rules to address the risk of foreign interference in scientific research, such as morethorough vetting of visa applications from international students and researchers, and requiring institutions to declare foreign sources of income. Last month, Japan’s cabinet approved an innovation strategy for 2020. This asks government agencies, research institutes and companies to strengthen codes of conduct around research integrity and conflicts of interest, and to prevent the outflow of sensitive research and technologies linked to national security, such as artificial intelligence and semiconductor manufacturing. It also proposes that government agencies consider withholding funding from institutions that fail to declare foreign income. No countries are named by the strategy, but researchers say the government is concerned mainly with the activities of Chinese institutions, including those with ties to the military. The government is now considering drawing up guidelines on these issues, says Takahiro Ueyama, an executive member of Japan’s Council

for Science, Technology and Innovation (CSTI), which is chaired by Prime Minister Shinzō Abe. “This is a very sensitive issue,” Ueyama says. The development follows crackdowns by US science agencies on researchers who do not disclose foreign ties, mainly with China. In the past two months, four ethnic Chinese researchers working in the United States have been charged with visa fraud for failing to declare links to China’s military. The Japanese government feels under pressure to strengthen its research-integrity guidelines and safeguard its scientific relationship with the United States, says Atsushi Sunami, a science-policy analyst at the National Graduate Institute for Policy Studies in Tokyo. “When the US and other Western countries started talking about these issues, it was natural that Japan would also address them more clearly.” The strategy’s language mirrors that of a 2019 report by the science group JASON, which advises the US government. The report was commissioned by the US National Science Foundation over concerns about foreign governments acquiring US science and technology.

. d e v r e s e r s t h g i r l l A . d e t i m i L e r u t a N r e g n i r p S 0 2 0 2 ©

Nature | Vol 584 | 13 August 2020 | 173

The world this week

PATRICK LANDMANN/SPL

News in focus

Controlling deforestation (shown here, in a tropical rainforest in the Congo Basin) could decrease the risk of future pandemics, experts say.

WHY DEFORESTATION AND EXTINCTIONS MAKE PANDEMICS MORE LIKELY Researchers are redoubling efforts to understand links between biodiversity and emerging diseases — and to use that information to predict and stop future outbreaks. By Jeff Tollefson

A

s humans diminish biodiversity by cutting down forests and building more infrastructure, they’re increasing the risk of pandemics of diseases such as COVID-19. Many ecologists have long suspected this, but a study now helps to reveal why. When some species are going extinct, those that tend to survive and thrive — rats and bats, for instance — are more likely to host potentially dangerous pathogens that can make the jump to humans. The analysis of around 6,800 ecological communities on 6 continents adds to a

growing body of evidence that connects trends in human development and biodiversity loss to disease outbreaks — but stops short of projecting where new disease outbreaks might occur. “We’ve been warning about this for decades,” says Kate Jones, an ecological modeller at University College London and an author of the study, published on 5 August in Nature1. “Nobody paid any attention.” Jones is one of a cadre of researchers that has long been delving into relationships between biodiversity, land use and emerging infectious diseases. Their work has mostly flown below the radar, but now, as the world reels from the

COVID-19 pandemic, efforts to map risks in communities around the globe and to project where diseases are most likely to emerge are taking centre stage. Last week, the Intergovernmental Science-Policy Platform on Biodiversity and Ecosystem Services (IPBES) hosted an online workshop on the nexus between biodiversity loss and emerging diseases. The organization’s goal now is to produce an expert assessment of the science underlying that connection ahead of a United Nations summit that’s planned for September in New York City, where governments are expected to make new commitments to preserve biodiversity.

. d e v r e s e r s t h g i r l l A . d e t i m i L e r u t a N r e g n i r p S 0 2 0 2 ©

Nature | Vol 584 | 13 August 2020 | 175

News in focus

Concentrating risk Previous research has shown that outbreaks of diseases such as severe acute respiratory syndrome (SARS) and bird influenza that cross over from animals to humans have increased in the past few decades3,4. This phenomenon is likely to be the direct result of increased contact between humans, wildlife and livestock, as people move into undeveloped areas. These interactions happen more frequently on the frontier of human expansion because of changes to the natural landscape and increased encounters with animals. But a key question over the past decade has

been whether the decline in biodiversity that inevitably accompanies human expansion on the rural frontier increases the pool of pathogens that can make the jump from animals to humans. Work by Jones and others5 suggests that the answer in many cases is yes, because a loss of biodiversity usually results in a few species replacing many — and these species tend to be the ones hosting pathogens that can spread to humans. For their latest analysis, Jones and her team compiled more than 3.2 million records from several hundred ecological studies at sites around the world, ranging from native forests to cropland to cities. They found that the populations of species known to host diseases transmissible to humans —

“We’ve been warning about this for decades. Nobody paid any attention.” including 143 mammals such as bats, rodents and various primates — increased as the landscape changed from natural to urban, and as bio­diversity generally decreased. Some researchers urge caution when communicating that biodiversity hotspots are where outbreaks are likely to occur. “My worry, frankly, is that people are going to cut down the forests more if this is where they think the next pandemic is going to come from,” says Dan Nepstad, a tropical ecologist and founder of the Earth Innovation Institute based in San Francisco, California, a non-profit organization that campaigns for sustainable development. Efforts to preserve biodiversity will work, he

Wildlife markets such as this one in Bali, Indonesia, sustain the livelihoods of many people.

. d e v r e s e r s t h g i r l l A . d e t i m i L e r u t a N r e g n i r p S 0 2 0 2 ©

176 | Nature | Vol 584 | 13 August 2020

says, only if they address the economic and cultural factors that drive deforestation and the dependency of rural poor people on hunting and trading wild animals. Ibrahima Socé Fall, an epidemiologist and head of emergency operations at the World Health Organization in Geneva, Switzerland, agrees that understanding the ecology — as well as the social and economic trends — of the rural frontier will be essential to projecting the risk of future disease outbreaks. “Sustainable development is crucial,” he says. “If we continue to have this level of deforestation, disorganized mining and unplanned development, we are going to have more outbreaks.”

Coordinating efforts One message that the upcoming IPBES report is likely to deliver is that scientists and policy­makers need to treat the rural frontier more holistically, addressing issues of public health, the environment and sustainable development in tandem. In the wake of the COVID-19 pandemic, many scientists and conservationists have emphasized curbing the wildlife trade — an industry worth an estimated US$20 billion annually in China, where the first coronavirus infections appeared. China has temporarily suspended its trade. But Daszak says the industry is just one piece of a larger puzzle that involves hunting, livestock, land use and ecology. “Ecologists should be working with infectious-disease researchers, public-health workers and medics to track environmental change, assess the risk of pathogens crossing over and reduce risky human activities,” he says. Daszak was an author of last month’s essay in Science, which argued that governments could substantially reduce the risk of future pandemics such as that of COVID-19 by investing in efforts to curb deforestation and the wildlife trade, as well as in efforts to monitor, prevent and control new virus outbreaks from wildlife and livestock. The team estimated that the cost of these actions would ring in at $22 billion to $33 billion annually. The total investment would be two orders of magnitude less than the $5.6-trillion price tag estimated for the COVID-19 pandemic, the team estimates. Fall says the key is to align efforts by government and international agencies focused on public health, animal health, the environment and sustainable development. With the right collaboration between human-health, animal-health and environmental authorities, Fall says, “you have some mechanisms for early warnings”. 1. Gibb, R. et al. Nature https://doi.org/10.1038/s41586-0202562-8 (2020). 2. Dobson, A. P. et al. Science 369, 379–381 (2020). 3. Jones, K. E. et al. Nature 451, 990–993 (2008). 4. Smith, K. F. et al. J. R. Soc. Interface 11, 20140950 (2014). 5. Faust, C. L. et al. Ecol. Lett. 21, 471–483 (2018).

AMILIA ROSO/THE SYDNEY MORNING HERALD VIA GETTY

Others are calling for a more wide-ranging course of action. On 24 July, an interdisciplinary group of scientists, including virologists, economists and ecologists, published an essay in Science2 arguing that governments can help to reduce the risk of future pandemics by controlling deforestation and curbing the wildlife trade, which involves the sale and consumption of wild — and often rare — animals that can host dangerous pathogens. Most efforts to prevent the spread of new diseases tend to focus on vaccine development, early diagnosis and containment, but that’s like treating the symptoms without addressing the underlying cause, says Peter Daszak, a zoologist at the non-governmental organization EcoHealth Alliance in New York City, who chaired the IPBES workshop. He says COVID-19 has helped to clarify the need to investigate biodiversity’s role in pathogen transmission. The latest work by Jones’s team bolsters the case for action, Daszak says. “We’re looking for ways to shift behaviour that would directly benefit biodiversity and reduce health risks.”

YANG JIANZHENG/VCG VIA GETTY

Artificial intelligence is one of the areas that the US National Science Foundation will prioritize.

NSF’S COMMITMENT TO BASIC RESEARCH QUESTIONED The US National Science Foundation shifts the focus of its graduate fellowships to computer science. By Giuliana Viglione

T

he major US agency tasked with funding basic research, the National Science Foundation (NSF), raised alarm among scientists late last month when it updated the guidance for its prestigious graduate-student fellowships to emphasize research in three areas of applied computational science. Critics fear that the new focus on artificial intelligence, computationally intensive research and quantum information science — despite the 70-year-old agency’s historical mandate to promote and support all basic scientific research — will decrease money for fundamental science that can struggle to attract funding from other government or industry sources. They also fear that the changes could make it even harder for graduate students from under-represented groups — including white women and Black and Latinx scientists — to win these grants. The change to the guidance for the Graduate Research Fellowship Program (GRFP) — which hands out hundreds of millions of dollars of research funding each year — comes amid concern in the United States

about the growth of China as a research superpower. It is just one of several recent efforts to steer the NSF towards applied-technology research that have concerned researchers. In late May, bills introduced to both houses of Congress with bipartisan support proposed to increase the NSF’s budget by US$100 billion over 5 years. The bills, which have not yet been voted on, would rebrand the organization as the National Science and Technology Foundation, and would allocate the new money to technology development, rather than to basic science.

Fears for basic science “The value of basic science is not always apparent to non-scientists,” says Margaret Byron, a mechanical engineer at Pennsylvania State University in State College, who received the fellowship in 2012. “If there’s pressure from those outside of science to push towards directions that have more immediate applications, then we need to push back.” The NSF said in both a public statement and an e-mail to Nature that the move is part of “a coordinated federal strategy to secure America’s position as a global leader in

research and innovation”, but that the fellowship “will continue to encourage and accept applications in all eligible fields of science and engineering”. The agency awards around 2,000 graduate fellowships each year, and it requested just over $275 million for the programme in 2021. The fellowships, which fund master’s or PhD students for 3 years, are typically awarded in 11 major fields of study, including engineering, life sciences and chemistry, but had previously not given preference to any particular subfields within these broad categories. The new GRFP guidance says applications are “encouraged” in all disciplines supported by the NSF that incorporate the three new high-­priority research areas. The focus on three types of computer-based science is “kind of bonkers to me”, says Michael Hoffman, who received the fellowship in 2003 and is now a computational biologist at the University of Toronto and the Princess Margaret Cancer Centre in Toronto, Canada. “These are focus areas that are already, right now, very well funded.” The strength of the GRFP, he says, is that it trains scientists across a broad range of disciplines that are not typically funded by other agencies. That’s important because “you can never predict which areas are going to have the really important discoveries”, Hoffman says.

Twitter backlash The changes prompted a backlash almost immediately, with scores of scientists expressing their discontent on Twitter. “These changes are incredibly restrictive and will almost certainly hurt bringing in bright and diverse students in the sciences broadly,” wrote Alexandra Harmon-Threatt, a pollination ecologist at the University of Illinois at Urbana–Champaign. Amy Tarangelo, a cancer biologist at the University of Texas Southwestern Medical Center in Dallas, tweeted: “As a former fellow, I’m super disappointed in this decision from @NSFGRFP. GRFP should fund promising scientists in ALL fields without regard to the ‘hot topic’ of the moment. This will hurt students at unis without resources for high-level computing, etc.” Some welcome the move towards computer science, however. “I think the changes are important and reflect the country’s need for more talents and advances in the fields,” says Anh Nguyen, a machine-learning researcher at Auburn University in Alabama. These fields “have a strong potential to transform human lives” across a variety of disciplines, he says. Still, the concentration of funding in certain fields without an expansion of the programme means that other areas — such as basic science — will be cut back, says Kelsey Lucas, a marine and aquatic comparative biomechanist at the University of Michigan in Ann Arbor, whose graduate work was supported by the GRFP. “By

. d e v r e s e r s t h g i r l l A . d e t i m i L e r u t a N r e g n i r p S 0 2 0 2 ©

Nature | Vol 584 | 13 August 2020 | 177

News in focus focusing on certain areas, that means other areas are going to be getting less funding.” In an e-mail to Nature, a spokesperson for the NSF wrote, “These changes are not intended to exclude any areas of science supported by NSF,” and pointed to the advances in basic science that the NSF has funded over the past seven decades. “NSF is simply signifying that these are areas of national importance and we are encouraging students to apply.” The spokesperson also said that the areas of emphasis would not change the review or selection process.

Diversity at risk Nevertheless, scientists fear that the move could make it harder for certain people to win these prestigious grants — including Black and Latinx scientists. “Frankly, I was disappointed,” says Christian Cazares, a neuroscientist at the University of California, San Diego, and a current GRFP award recipient. He views the changes as “completely antithetical” to the NSF’s stated commitment to promoting diversity in science. The lack of diversity in the three new priority fields — only 18.8% of US computer-science bachelor’s degrees went to Black and Latinx students and 18.7% to women in 2016 — means that the move will perpetuate already-existing disparities, he says. (According to a 2014 review of the programme, between 1994 and 2004, about 80% of GRFP awards went to white applicants.) Research has shown funding priorities to be one of the main drivers of inequity in grants awarded to Black scientists, says Alexandra Clark, a neuropsychologist at the University of California, San Diego. In 2019, a team at the US National Institutes of Health found that topic choice had the second-largest effect on the gap in award rates between white and Black researchers among proposals selected for discussion by reviewers (T. A. Hoppe et al. Sci. Adv. 5, eaaw7238; 2019). “When we know there’s an opportunity gap that’s at play,” Clark says, “we can’t really just continue on like business as usual.” “Increasing diversity and inclusiveness is a top priority for the Director and NSF,” wrote a spokesperson for the NSF in an e-mail to Nature, adding that the agency’s director has established a task force to make recommendations for addressing barriers to inclusion. Byron is hopeful that the changes will not significantly alter the GRFP selection process. But she does worry that students from under-represented backgrounds or lower-resourced schools will be dissuaded from applying because of the programme’s new emphasis. “The last thing anybody wants to see is for a really stellar young scientist to look at this new solicitation and say, ‘that’s not me’.”

AUSTRALIA’S PLAN TO END FOREIGN INTERFERENCE IN SCIENCE: DID IT WORK? Pioneering guidelines aren’t enough to prevent overseas militaries co-opting research, say experts. By Dyani Lewis

A

lmost a year after Australia introduced a pioneering system for minimizing the risk of foreign interference in research — in particular, from overseas militaries — observers are divided about whether it is working. The guidelines, which were introduced last November and are widely assumed to be a response to concerns about the Chinese military’s ties to universities, encourage institutions to perform risk assessments on potential collaborators, communicate the risk of foreign interference to staff and bolster cybersecurity. They also urge universities to ensure that they comply with laws that restrict exports of certain technologies, such as those that have military uses. Although other countries, including the United States, the United Kingdom and Japan, are grappling with similar concerns, Australia is the first to set such a specific set of guidelines for its universities. But some specialists warn that Australia’s guidelines and export laws aren’t sufficient

. d e v r e s e r s t h g i r l l A . d e t i m i L e r u t a N r e g n i r p S 0 2 0 2 ©

178 | Nature | Vol 584 | 13 August 2020

There are growing concerns about ties between China‘s military and its universities.

to help universities identify collaborations in which research could lead to military applications. Although the guidelines outline ‘best practice’ steps that universities could take to mitigate risks, the measures are optional, says Alex Joske, a China analyst at the Australian Strategic Policy Institute, a think tank in Canberra. Others question whether universities are up to the task of scrutinizing international partners, particularly those in China. The guidelines are “basically saying, do due diligence on your partner in an opaque authoritarian political system”, says Jeffrey Wilson, a political scientist at the think tank Perth USAsia Centre in Crawley, Australia. “You’re asking people to do something no one can do, except maybe a spy agency,” he says. Others say the system is working well. James Laurenceson, director of the Australia– China Relations Institute at the University of Technology Sydney, says export controls on weapons and technologies that could have military uses, such as facial-recognition or cybersecurity software, reduce the risk of research being used by international armed

forces. He says this is the right approach — focusing on the research, rather than who the collaboration is with. “Fundamental questions about the science being conducted — I think they’re more important ones for us to be asking,” he says.

SOURCE: J. LAURENCESON & M. ZHOU THE AUSTRALIA-CHINA SCIENCE BOOM HTTPS://GO.NATURE.COM/3ADYFUL (2020)

KEVIN FRAYER/GETTY

Research partners On paper, China is Australia’s biggest research partner. In 2019, Australian researchers co-authored close to 14,000 papers with authors who had affiliations in China, according to an analysis of papers indexed in the Scopus database (go.nature.com/3adyful). That’s 16.2% of Australia’s research output, more than for any other international partner nation (see ‘Australia’s top collaborators’). Fierce competition for limited government funds in Australia is driving collaborations, particularly with China, says Wilson, adding that this exacerbates the risk of interference. “One of the challenges for Australian universities is that they have been heavily incentivized to seek foreign research income,” he says. In most cases, research collaboration benefits all parties, says Yun Jiang, a geoeconomist at the Australian National University in Canberra. Problematic research is only a small part of research collaborations, she says. But as China increases links between its civilian universities and the armed forces — a policy known as military–civil fusion — those links are becoming more difficult to identify, says Joske. Chinese universities, he says, “are more and more integrated with military assets, especially around research and development. They’re working on classified projects and a lot of their graduates are going into the military or the defence industry.”

Military ties In 2019, Joske and his colleagues developed the China Defence Universities Tracker, which places Chinese institutions on a risk scale according to how closely they are associated with the military and whether they have been accused of or engaged in espionage and intellectual-property theft. Ninety-two institutions — including 60 run by the People’s Liberation Army or security and intelligence agencies, and 20 civilian universities — are rated as very high risk. A further 23 civilian universities are graded as high risk. Joske is concerned about collaborations with universities in China that have military ties — and about how these should be policed. In his view, the Australian government should provide universities with information about which foreign institutions deserve more scrutiny, and should establish a national research-integrity office that can enforce existing national-security and

export-control laws. Some academics think that collaborations with military institutions should be ruled out entirely. “These universities are exceptionally important to the development of China’s military technology,” says Clive Hamilton, a public-policy researcher at Charles Sturt University in Canberra, who has investigated Chinese influence at Australian universities. “So why take the risk?” The tracker also lists 12 state-owned defence-industry conglomerates, 4 of which have links to overseas universities. One example of the kinds of relationship that are raising alarm is outlined by Joske in a 2019 report accompanying the tracker (see go.nature.com/3gjynsf). The report noted that one of those conglomerates, the Commercial Aircraft Corporation of

“We want to be transparent. But how do you deal with that when your research partner is fundamentally not?” China (COMAC), signed a memorandum of understanding with Monash University in Melbourne in May 2017 to design 3D-printed aircraft components. The collaboration will also fund a Aus$10-million (US$7-million) Aeronautical Research Centre that will be established at the university, a spokesperson from the university said. Joske says the collaboration is a cause for concern because, last year, the cybersecurity company CrowdStrike in Sunnyvale, California, accused COMAC of using technology stolen through cyberespionage from rival airlines to design its C919 commercial aircraft, which could be converted into a military surveillance aircraft.

AUSTRALIA’S TOP COLLABORATORS

Last year, researchers in Australia co-authored almost 14,000 papers with people at Chinese institutions — more than with any other country. 2018

2019

China

United States United Kingdom

Germany

Canada 0

4

8

12

16

20

Percentage of total Australian publications

COMAC did not respond to Nature’s questions about the allegations. But the university defended the partnership, saying it followed the Australian Code for the Responsible Conduct of Research and the university’s own policies and procedures. “The university regularly monitors advice provided by the government, intelligence agencies and the wider education sector, and acts in accordance with advice received,” the spokesperson said.

Group of Eight Several Australian institutions have made changes in response to the guidelines produced last year. Nature contacted the ‘Group of Eight’ leading research universities, and of the seven that responded, five said they had introduced new processes or were deciding whether to do so. But some experts think that more needs to be done. The guidelines recommend that foreign affiliations and funding be recorded. Hamilton says universities should go a step further, and make these registers public. “Transparency must be priority number one,” he says. “Australian universities are public institutions, and I can see no reason why the public should not be permitted to know the affiliations and financial links academics may have with other organizations, at home or abroad,” he says. And he says the universities should take a hard line with researchers who fail to disclose foreign ties, including firing them in some cases. The government should also be stricter with universities, says Hamilton: “Universities that fail to heed government rules and guidelines concerning research links that jeopardize national security should be excluded from receiving funding from the Australian Research Council.”

Limits to transparency In the United States, funding agencies have alerted institutions to grant recipients with potential undisclosed foreign ties. Investigations have led to dozens of researchers being sacked or forced to give back research funding. But transparency has its limits, says Wilson. “We want to be transparent. But how do you deal with that when your research partner is fundamentally not?” he asks. Although information on connections to the military is openly available in some instances, he says, it could easily be hidden. In June, the Australian government announced that it had created a new integrity unit to identify and analyse emerging threats to the quality of higher education, and to assist universities in addressing foreigninterference threats. Joske says it’s a positive move, but it will take time to prove that it is working.

. d e v r e s e r s t h g i r l l A . d e t i m i L e r u t a N r e g n i r p S 0 2 0 2 ©

Nature | Vol 584 | 13 August 2020 | 179

News in focus

THOMAS HENRIKSON

Q&A

Pioneering research director: ‘We’re in a cataclysmic time of change’

You’ve been a scientist since the 1950s. What’s changed, and what hasn’t, regarding sexism in science? When I was an undergraduate, I went to the department chair to ask for a fellowship to allow me to pursue a master’s degree. He told me, quite bluntly, that they didn’t waste fellowships on women. I don’t think any chair, dean or faculty member would say that today. It would probably be nuanced, to the effect of, “We don’t have any available.” That makes it less obvious, and I’m not so sure less hurtful — because it still deflects the career path.

What do you want early-career female scientists to take away from your book? That you’re not alone. Your experiences that you endure are shared. Like other women, I assumed the problems I have had were some sort of problem of mine. But in fact, it’s really the system. What should institutions do to improve the situation for women? Agencies, particularly funding agencies, can have a tremendous effect. The NSF has really been a leader, tracing back to when Mary Clutter, who was an assistant director, forcibly mandated [in 1989] that committees and scientific meetings that received funding had to have female representatives on the committee and female speakers at the meeting. Racism in science is also deeply entrenched. Why is it important to change that? There’s tremendous talent distributed throughout the population, and the country needs all of it. It’s really important to draw from 100% of the population — and not 50% or less. The challenges we’re facing right now are huge. We’re in one of the cataclysmic times of change.

In 2001, you helped to hunt down the source of the US anthrax attacks. What lessons from that could help scientists investigating COVID-19? The anthrax episode was a fantastic experience showing patriotic cooperation among agencies and among branches of Congress. Because I was director of the NSF, I was able to have a leadership role, and because I was a microbiologist, I was poised to recognize immediately that we had to sequence the bacterium that had been sent out as a bioweapon. The group met every week for an hour in a classified meeting. We had maybe 16 or 17 agencies represented. The lesson was collaboration and cooperation. No single agency could solve the problem. There was shared trust and shared commitment, and no politics in any of our decisions. If we had been able to have an inter-agency, intergovernmental interaction straight away in January or February, I think we would not have 150,000 COVID-19 deaths — and climbing — in the United States. How has your laboratory changed its emphasis in response to COVID-19? I have been working on cholera for my entire career. It is an aquatic bacterium distributed in the environment. Twenty years ago, we developed the use of satellite sensing to monitor environmental parameters that allowed us to track and predict cholera outbreaks. We’re modifying that model for COVID. It’s very interdisciplinary. That’s the power of understanding diversity — human, as well as environmental — and protecting both. What role should international scientific collaboration have in fighting the pandemic? Diseases don’t carry passports, and they don’t behave like tourists clearing the border. It’s almost anti-human not to understand that we are a global society, and that the benefit of science — for example, vaccines — needs to be shared. And it needs to be shared empathetically in a way that makes it affordable for the poorest citizens of the world.

Protests have erupted in the United States over racism and COVID-19 shutdowns.

. d e v r e s e r s t h g i r l l A . d e t i m i L e r u t a N r e g n i r p S 0 2 0 2 ©

180 | Nature | Vol 584 | 13 August 2020

Interview by Alexandra Witze This interview has been edited for length and clarity.

FREDERIC J. BROWN/AFP/GETTY

Rita Colwell is the former director of the US National Science Foundation (NSF) — the first woman to hold that post — and a leader in cholera research. In her new book, A Lab of One’s Own: One Woman’s Personal Journey Through Sexism in Science (written with Sharon Bertsch McGrayne), she opens up on one topic she hasn’t said much about publicly: her battle to improve the situation for women in science. Colwell, a microbiologist at the University of Maryland College Park, spoke to Nature about discrimination, and how her experiences as a researcher and agency leader during the 2001 anthrax bioterror attacks can inform the response to the coronavirus pandemic.

ILLUSTRATION BY KAROL BANACH

Feature

. d e v r e s e r s t h g i r l l A . d e t i m i L e r u t a N r e g n i r p S 0 2 0 2 ©

182 | Nature | Vol 584 | 13 August 2020

WHAT ANIMALS REALLY THINK

Neuroscientists are scrutinizing mountains of data to uncover how brains create emotions and other internal states, such as aggression or desire. By Alison Abbott

T

wo years ago, Jennifer Li and Drew Robson were trawling through terabytes of data from a zebrafish-brain experiment when they came across a handful of cells that seemed to be psychic. The two neuroscientists had planned to map brain activity while zebrafish larvae were hunting for food, and to see how the neural chatter changed. It was their first major test of a technological platform they had built at Harvard University in Cambridge, Massachusetts. The platform allowed them to view every cell in the larvae’s brains while the creatures — barely the size of an eyelash — swam freely in a 35-millimetre-diameter dish of water, snacking on their microscopic prey. Out of the scientists’ mountain of data emerged a handful of neurons that predicted when a larva was next going to catch and swallow a morsel. Some of these neurons even became activated many seconds before the larva fixed its eyes on the prey1.

Something else was strange. Looking in more detail at the data, the researchers realized that the ‘psychic’ cells were active for an unusually long time — not seconds, as is typical for most neurons, but many minutes. In fact, more or less the duration of the larvae’s hunting bouts. “It was spooky,” says Li. “None of it made sense.” Li and Robson turned to the literature and slowly realized that the cells must be setting an overall ‘brain state’ — a pattern of prolonged brain activity that primed the larvae to engage with the food in front of them. The pair learnt that, in the past few years, other scientists using various approaches and different species had also found internal brain states that alter how an animal behaves, even when nothing has changed in its external environment. Some, such as Li and Robson, had come to the discovery serendipitously while trudging through their own brain-wide data. Others have hypothesized that neurons coding for internal brain states must exist, and have actively sought them in discrete and well-researched

brain regions. For example, earlier this year2, neurobiologist David Anderson at the California Institute of Technology (Caltech) in Pasadena and his colleagues identified an internal brain state — represented by a small network of neurons — that prepares fruit flies to engage in courtship or fighting behaviours. Neuroscientists wanting to understand the brain’s coding language have conventionally studied how its networks of cells respond to sensory information and how they generate behaviour, such as movement or speech. But they couldn’t look in detail at the important bit in between — the vast quantities of neuronal activity that conceal patterns representing the animal’s mood or desires, and which help it to calibrate its behaviour. Even just a few years ago, measuring the activities of specific networks that underlie internal brain states was impossible. A slew of new techniques is starting to change that. These methods allow scientists to track electrical activity in the brain in unprecedented detail, to quantify an animal’s natural behaviour on millisecond timescales, and to find patterns in the mountains of data these experiments generate. These patterns could be signatures of the innumerable internal states that a brain can adopt. Now the challenge is to find out what these states mean. Some neuroscientists are daring to wield the technologies to probe one powerful group of internal brain states: emotions. Others are applying them to states such as motivation, or existential drives such as thirst. Researchers are even finding signatures of states in their data for which they have no vocabulary. The current trickle of research papers on internal brain states is gaining momentum. The work might even have potential clinical applications. “Mental illness is essentially disruption of internal states,” says Joshua Gordon, director of the US National Institute of Mental Health in Bethesda, Maryland. “They need to be understood.”

Frames of mind The brain of any animal is constantly bombarded with information about the creature’s environment from sensory organs such as the eyes, ears, nose or skin. All of this information is initially processed in the brain’s sensory cortex. Then come more mysterious processing steps, in which that information is filtered through multiple internal brain states representing the creature’s constantly changing moods and needs. That finally leads the motor cortex to generate movements that are appropriate to the circumstances — to flap away a tickling fly, for example, or to move towards a tasty treat. Internal states can also be generated entirely in the brain, without sensory input and without a behavioural output: think of daydreaming, or replaying the events of the day in your mind.

. d e v r e s e r s t h g i r l l A . d e t i m i L e r u t a N r e g n i r p S 0 2 0 2 ©

Nature | Vol 584 | 13 August 2020 | 183

Over the past few years, insights into the nature of internal states are changing how neuroscientists who study brain networks think about animal behaviour. “We used to think of animals as being kind of stimulus-response machines,” says neuroscientist Anne Churchland at Cold Spring Harbor Laboratory, New York. “Now we’re starting to realize that all kinds of really interesting stuff is being generated within their brains that changes the way that sensory inputs are processed — and so changes the animals’ behavioural output.” Sketching out how to study this intriguing middle ground has long been a preoccupation for Anderson. Six years ago, he decided to create a theoretical framework for research into internal brain states that represent emotion. He was irked by the view of some psychologists, who think that because animals can’t express their feelings in words, those feelings can’t be studied at all. Together with his Caltech colleague Ralph Adolphs, Anderson developed and published a hypothesis3 about the characteristics that neural circuits associated with internal brain states should have. Most importantly, they thought, an internal brain state should outlast the original stimulus that triggered it. So a key feature of a neural circuit underpinning such a state would be its persistence, he says. “If you are hiking in mountains and see a snake, then you might jump in fear,” says Anderson. “Ten minutes later, your brain’s internal state of fear is still active, so when you see a stick on your path you might jump again.” Other characteristics of internal states should include generalizability, meaning that different stimuli should be able to prompt the same state, and scalability, in which different stimuli can create states of different strength. The paper became influential. Li says that it “was inspirational” as she and Robson were trying to make sense of their psychic cells. Anderson and Adolphs published their paper in 2014, just as a raft of neurotechnologies was starting to make the necessary experiments feasible. It was already possible to record from large numbers of individual neurons at the same time, and since then the technologies have improved and expanded remarkably, allowing scientists to analyse previously inaccessible activity. Leading the pack is the Neuropixels probe, just 10 mm long, which can directly record activity in hundreds of neurons across different brain areas4. And special imaging techniques can indicate where as many as tens of thousands of individual neurons are active across the brain. In calcium imaging, for instance, animals are genetically engineered to express a molecule in their cells that detects calcium ions — when these pour into a neuron as it fires, the molecule fluoresces. New automatic behaviour monitors take video recordings of freely behaving animals

Distinct groups of neurons control whether zebrafish larvae explore (right) or stay put (left).

over many hours, and analyse every movement in millisecond elements. The elements can then be aligned with neural recordings, matching moment-to-moment brain activity with specific movements. Neuroscientists have capitalized on a surge in machine learning, artificial intelligence and new mathematical tools to make sense of the gigabytes or terabytes of data that any experiment with these technologies can generate, and to coax out the neural activation patterns that could represent internal brain states.

Ready to engage For his first study of an internal state, Anderson decided to build on his laboratory’s previous interest in aggression in the fruit fly, which has a tiny brain containing about 100,000 neurons. In many animal species, males start to fight each other in the presence of females — a well-established behaviour that Anderson calls the ‘Helen of Troy effect’, after the Greek myth about a woman whose competing suitors started a war. Fruit flies are no exception: indirect evidence suggests that exposure to females causes males to engage in both courtship songs and aggressive behaviour towards other males for many minutes. “That’s a long time in the short life of a fruit fly,” he says. He decided to search for neural activity that correlated with the persistent courtship and fighting behaviours that are initiated by neurons known as P1, found in a region that controls such social behaviours. These neurons fire so quickly that they alone couldn’t be responsible for maintaining an internal state. Using imaging techniques along with automated behavioural analysis, his group identified cells in other brain areas that become

. d e v r e s e r s t h g i r l l A . d e t i m i L e r u t a N r e g n i r p S 0 2 0 2 ©

184 | Nature | Vol 584 | 13 August 2020

J. C. MARQUES ET AL./NATURE (REF. 1)

Feature

active as a consequence of P1 activation. Most of these ‘follower cells’ switched quickly on and off, but a cluster called pCd neurons stayed active for many minutes. When the researchers inserted a light-sensitive protein into these cells and switched them off using a flash from a laser, the persistent effect of P1 activation on behaviour disappeared. When they activated them directly, bypassing P1, nothing happened: the pCd neurons needed P1 as a trigger and, once sparked into action, they stayed on for much longer than the initial prompt2. If Anderson had to give the state a name, he might call it the ‘ready-to-engage-inthese-social-behaviours’ state, he says. His team has conducted a similar experiment in mice5, which have more complex brains containing about 100 million neurons. The researchers found a particular group of neurons in the hypothalamus that, just like the pCd neurons, became persistently activated in association with an innate drive — this time, fear. When the scientists placed a rat close to experimental mice for just a few seconds, the mice responded defensively by hugging the wall for several minutes and the group of neurons remained active for all this time. When the team again used light to switch the neurons on and off, the wall-hugging behaviour came and went in tandem, even with no rat present. Neuroscientists are now discovering other groups of neurons with persistent activity in different brain areas. Using calcium imaging in mice, Andreas Lüthi at the Friedrich Miescher Institute for Biomedical Research in Basel, Switzerland, and Jan Gründemann at the University of Basel searched in the amygdala, which is central to the regulation of a range of emotions

and behaviours. The team found two different populations of neurons that displayed sustained but opposing activation when the mice switched between two distinct behaviours6 — exploring the environment and performing defensive behaviours such as freezing. Gründemann acknowledges that the amygdala cells are unlikely to be working in isolation, and that cells across the whole brain are involved in maintaining the explorative or defensive states. “I’m sure it is just one node in larger, brain-wide networks,” he says.

The whole picture Whereas many researchers have searched particular brain areas for neurons that have enduring activity, Li and Robson, who moved to Germany last September to jointly run a lab at the Max Planck Institute for Biological Cybernetics in Tübingen, came across their persistently active neurons almost by chance. Their zebrafish larvae are less complex than fruit flies, having only 80,000 or so brain cells. Because these baby fish are transparent, the activity of nearly all of their neurons can be monitored simultaneously using calcium imaging. The pair has developed a method of concurrently following both the movements and the neural activity as fish larvae swim freely around a dish. They deploy a fluorescent-microscope tracking system that moves on its imaging platform to keep the fish in constant view, and captures every flash of each neuron as the larvae move. The system also films them — typically for 90 minutes, generating 4.5 terabytes of data — allowing the experimenters to align movement with neuronal activity second by second. Fish larvae might not seem to have the rich internal life enjoyed by mice, or even flies, but they have at least one robust behavioural choice to make in their lives — whether to hunt locally, or to swim to unfamiliar waters to search for new food sources. When Li and Robson watched larvae making this choice, they found three groups of neurons: one that was persistently active during local hunting, another that stayed active during exploration and a third that flashed on briefly as the fish switched states1. Surprisingly, hunger didn’t seem to influence the states, which switched automatically every few minutes — “just like our own sleep–wake states switch automatically, but on a much shorter timescale”, Robson says. Neuroscientists working with more complex organisms can’t monitor the whole brain at once, but they have been able to find hints of internal brain states with networks that are widely distributed in the brain. In technically challenging experiments in mice, they have recorded the activity of thousands of neurons throughout the brain using calcium imaging, and of hundreds of neurons using a single Neuropixels electrode, several of which can

be inserted at once. In a study published last year7, neuro­scientist Karl Deisseroth at Stanford University in California and his team used Neuropixels probes to record the activity of 24,000 neurons across 34 cortical and subcortical brain regions in thirsty mice that were licking water from a spout. The scientists were able to tease out signals related to the brain state of thirst from signals related to licking behaviour. They found that these state-signalling neurons were activated throughout the brain — not just in the hypothalamus, where dedicated thirst neurons are located. Using these extensive recording techniques,

MENTAL ILLNESS IS ESSENTIALLY DISRUPTION OF INTERNAL STATES. THEY NEED TO BE UNDERSTOOD.” neuroscientists are finding that there is a lot going on beneath the surface when an animal performs a task — and not all of it seems relevant at first glance. In landmark papers last year, groups led by Kenneth Harris at University College London and by Churchland showed that when a mouse is engaged in a task, neurons activate throughout the brain, but that a large proportion of the activation is not correlated with the task at all8,9. Some activity correlated instead with the animals’ fidgety movements. But around two-thirds of the off-task activation didn’t tally with any movement or action. “Part of this may be related to internal brain states,” says Harris.

Busy brain Many neuroscientists say that the sheer volume of data pouring out of whole-brain experiments is also the field’s biggest bottleneck. But they have been making progress in developing techniques to sift through the flood of measurements. One popular approach is to use a mathematical method called the hidden Markov model (HMM) to predict the probability that a system will switch between different states at a particular time. Mala Murthy at Princeton University, New Jersey, and her colleagues used the HMM to discover rhythms in the brains of male fruit flies10 that influenced their choice of song pattern when courting females. Whether male flies choose on a moment-to-moment

basis to sing in staccato pulses or longer hums depends in large part — but not totally — on how the females respond to them. Murthy’s group found that three different internal brain states also affected the male’s song choice. They dubbed the fly dispositions Close, Chasing and Whatever. No matter the complexity of the model organism that individual researchers have adopted — worm, fish, fly or mouse — the question of how the entire brain coordinates internal states “is what we are all starting to think about”, says Steve Flavell at the Massachusetts Institute of Technology in Cambridge. In 2013, Flavell and his colleagues discovered that even the brain of the Caenorhabditis elegans worm, which has only 302 neurons, displays properties of internal brain states that drive particular behaviours, including two sets of persistently active neurons controlling whether the animal lingers locally or moves with purpose11. His group has since identified the full circuitry involved in the two states and switching between them12. Aside from their questions about the basic biology, researchers have an eye on the clinical benefit of understanding how a particular state manifests in the brain. Those studying pain in rodent models, for instance, rely on standard tests such as observing when a rat lifts its paw from a hot plate. “That movement reflects protective aspects of pain, but not the actual perception of pain,” says neurologist Clifford Woolf at the Boston Children’s Hospital in Massachusetts. That makes it a poor model for pain, he argues, because it is one step removed from the actual sensation. He has launched a research programme to try to directly read brain signals that indicate the internal state of pain perception — potentially a more timely and specific readout than waiting for the animal’s response. “I’m extremely optimistic that we’re in one of those rare stages in science where this is going to be a transformation of the way we do things,” he says. In this new field, even the basics are up for grabs, says Li. “At this stage, we are still trying to understand what the questions are.” Alison Abbott is a writer based in Munich, Germany. 1. Marques, J. C., Li, M., Schaak, D., Robson, D. N. & Li, J. M. Nature 577, 239–243 (2020). 2. Jung, Y. et al. Neuron 105, 322–333 (2020). 3. Anderson, D. J. & Adolphs, R. Cell 157, 187–200 (2014). 4. Jun, J. J. et al. Nature 551, 232–236 (2017). 5. Kennedy, A., Kunwar, P. S., Li, L., Wagenaar, D. & Anderson, D. J. Preprint at bioRxiv https://doi. org/10.1101/805317 (2020). 6. Gründemann, J. et al. Science 364, eaav8736 (2019). 7. Allen, W. E. et al. Science 364, eaav3932 (2019). 8. Stringer, C. et al. Science 364, eaav7893 (2019). 9. Musall, S., Kaufman, M. T., Juavinett, A. L., Gluf, S. & Churchland, A. K. Nature Neurosci. 22, 1677–1686 (2019). 10. Calhoun, A. J., Pillow, J. W. & Murthy, M. Nature Neurosci. 22, 2040–2049 (2019). 11. Flavell, S. W. et al. Cell 154, 1023–1035 (2013). 12. Cermak, N. et al. eLife 9, e57093 (2020).

. d e v r e s e r s t h g i r l l A . d e t i m i L e r u t a N r e g n i r p S 0 2 0 2 ©

Nature | Vol 584 | 13 August 2020 | 185

Science in culture

Books & arts Crunch, rip, freeze or decay — how will it all end? Astrophysicist Katie Mack’s book explores all the ways the Universe could destroy itself. By Ramin Skibba

NASA/ESA

S

cientists know how the world will end. The Sun will run out of fuel and enter its red-giant phase. Its final burst of glory will expand and engulf the closest planets, leaving Earth a charred, lifeless rock. Our planet has around five billion years left. With this grim image, theoretical astrophysicist Katie Mack begins her book on the end of the Universe — a much more uncertain prospect. Cosmologists generally look backwards, because all the evidence they can examine with telescopes is far away and concerns things that happened long ago. Using the motions of distant stars and galaxies to predict possible futures involves more speculation. In Mack’s hands, this speculation makes for a fascinating story. Humans are, she writes, “a species poised between an awareness of our ultimate insignificance and an ability to reach far beyond our mundane lives, into the void, to solve the most fundamental mysteries of the cosmos”. She is a talented communicator of complex physics, and the passion and curiosity about astronomy that have made her a popular speaker and Twitter presence are evident here. (As are some nerdy jokes and a less compelling coda about new physics research tangential to the central theme.) Mack begins at the beginning, with the Big Bang. What followed was inflation — a period of rapid expansion. Then, structures of dark matter formed and the building blocks of stars, planets, life and galaxies assembled. Currently, dark energy, thought to pervade the Universe, The End of Everything (Astrophysically Speaking) Katie Mack Scribner (2020) 

The Universe is expanding — for now.

somehow counteracts the forces of gravity to keep driving expansion. The Universe’s fate depends on whether that expansion will continue, accelerate or reverse. Astrophysicists long considered the most likely denouement to be a reversal of the Big Bang — the Big Crunch. Outside our cosmic neighbourhood, every galaxy is zooming away from us; a clear sign of expansion. If the Universe holds enough matter, including dark matter, the combined gravitational attraction of everything will gradually halt this expansion and precipitate the ultimate collapse. Over time, galaxies, then individual stars, will smash into each other more frequently, killing off any

life on nearby planets. In the final moments, as densities and temperatures soar in a contracting inferno, all that remains will extinguish in a single point. But dark energy might mean that a different end awaits. The early years of the Universe’s evolution were determined by the amount of matter it held; over the past few billion years, dark energy has begun to dominate, pushing the universe outwards. Current data from the European Space Agency’s Planck telescope and other sources are consistent with this expansion continuing forever. Called the Heat Death or Big Freeze, this apocalypse will be “slow and agonizing”, Mack writes. In thermodynamic terms, she explains, the Universe will approach a state of minimum temperature and maximum entropy. As everything gets farther and farther apart, the material of dead stars will disperse so that new stars can’t form, and the galaxies they’re part of will gradually stop growing. It’s like a suffocation of all astrophysical activity, as the fuel for growth and reproduction becomes so diffuse as to be unusable. It is an end “marked by increasing isolation, inexorable decay, and an eons-long fade into darkness”. The third demise that Mack discusses is the Big Rip. This is in store if dark energy accelerates expansion even more than is currently expected. As the Universe balloons, eventually, gravitational forces won’t be able to keep galactic clusters together. Stars will be stripped away from each other, and solar systems such as ours won’t have the strength to stay together. The remaining stars and planets will explode. Finally, the last atoms will be ripped apart. The latest measurements point to a Heat Death, but a Big Crunch or Big Rip are within their uncertainties. The final doomsday scenario that Mack describes is extremely unlikely: vacuum decay. A tiny bubble of ‘true vacuum’ could form, owing to instability in the field associated with the Higgs boson. That might happen if, say, a black hole evaporates in just the wrong way. Such a bubble would expand at the speed of light, destroying everything, until it cancels the universe. Vacuum decay might already have begun in some distant place. We won’t see it coming. Not to worry, though. As Mack counsels, whatever it looks like, the end probably won’t be nigh for at least 200 billion years. Ramin Skibba is an astrophysicist turned science writer based in San Diego, California. e-mail: [email protected]

. d e v r e s e r s t h g i r l l A . d e t i m i L e r u t a N r e g n i r p S 0 2 0 2 ©

Nature | Vol 584 | 13 August 2020 | 187

Readers respond

Correspondence COVID-19: time to re-imagine academic publishing

COVID-19: indexed data speed up solutions

Denmark recycling COVID-19: full peer plan will cut waste by review in hours two-thirds

High-profile retractions have highlighted how the conventional model of academic publishing has struggled to keep pace with the race to understand the new coronavirus, SARSCoV-2. The system is ripe for innovation. To that end, an open-access overlay journal known as Rapid Reviews: COVID-19 (RR:C19; see go.nature. com/3fufauw) uses the speed of technology to democratize the review process and strengthen the quality of research. RR:C19 was launched this year by the MIT Press and the University of California, Berkeley, with support from the Patrick J. McGovern Foundation. Scientists, publishers and philanthropic foundations work together to swiftly deploy new models for digitally enabled publishing. The journal promotes rapid and transparent peer review of promising or controversial preprints, as well as dynamic curation of content (see B. M. Stern and E. K. O’Shea PLoS Biol. 17, e3000116; 2019). Philanthropic foundations have been leaders in funding risky scientific ventures. In our experience, extending that support to advance the publishing process will boost the quality of research and accelerate its dissemination.

Since the COVID-19 pandemic began, cross-disciplinary data sets for the virus have been proliferating daily. However, these can be difficult for researchers to find, link to and reuse — for example, if they want to explore new hypotheses or to test existing ones. To this end, we have developed the COVID-19 Data Index (www.covid19dataindex.org). Originally funded by the US National Institutes of Health as part of the Big Data to Knowledge project, the index hosts a metadata catalogue of COVID-19 data sets. These range from, for example, clinical, sociodemographic, environmental, economic and mobility data to case statistics and genomic sequences. These data are collected from large repositories, research papers and individual online sources, among others. Users can filter search results by type (‘-omics’ data versus clinical data, for instance), repository or geographic location. The index then supplies links to the original data and the download page. We update the COVID-19 Data Index daily. Finding data on COVID-19 is no longer an obstacle that could delay discoveries.

As one of the European Union’s largest energy consumers and greenhouse-gas polluters (go.nature.com/33piuuv), Denmark will launch the EU’s most-ambitious recycling plan in July next year. It aims to cut the country’s annual amount of waste for incineration from 800 to 250 kilograms per capita, reducing carbon dioxide emissions to 0.7 million tonnes by 2030. Citizens will sort their waste into ten different types. The move is in part a response to the COVID-19 pandemic and to a new EU directive for environmental sustainability that promotes a circular economy, lower emissions and a reduction in the use of raw materials and hazardous substances ( J. B. Zimmerman et al. Science 367, 397–400; 2020). It is hoped that the plan will limit ecosystem damage and the health effects of toxic industrial chemicals. It will also discourage Denmark’s unacceptable export of waste to low-income countries. If other countries were to adopt similar practices, the world would align faster with the United Nations Sustainable Development Goals on sustainability and planetary health.

Vilas Dhar* Patrick J. McGovern Foundation, Boston, Massachusetts, USA. Amy Brand* The MIT Press, Cambridge, Massachusetts, USA. [email protected] *V.D. and A.B declare competing interests; see go.nature. com/2xajnzt

Lucila Ohno-Machado* University of California, San Diego, La Jolla, California, USA. [email protected] Hua Xu The University of Texas Health Science Center at Houston, Texas, USA. *L.O.-M. declares competing interests; see go.nature. com3gfcgv

Christian Sonne Aarhus University, Roskilde, Denmark. [email protected]

William J. Sutherland University of Cambridge, UK. [email protected]

Su Shiung Lam Universiti Malaysia Terengganu.

. d e v r e s e r s t h g i r l l A . d e t i m i L e r u t a N r e g n i r p S 0 2 0 2 ©

192 | Nature | Vol 584 | 13 August 2020

Aage K. O. Alstrup Aarhus University, Aarhus, Denmark.

The impetus to rapidly disseminate scientific results during a crisis ahead of peer review could cause governments and international organizations to act prematurely — or to be reluctant to act at all. Having struggled with such challenges in our COVID-19 work, we recommend our tested review system, which has an ultrashort submission-to-acceptance time. One of us (W.S.) runs a workshop every September to identify horizon-scanning issues in conservation, which we aim to report on in Trends in Ecology and Evolution the following January. The journal’s editor (formerly K.A.L.) agrees a submission date and selects referees. Authors send in a working draft a week before formal submission so that referees have time to prepare their comments. In the first year (2009), the time from formal submission to return of detailed comments was 90 minutes. The crucial features of this process are advance selection of referees and the provision of a draft manuscript. Agreeing a submission date makes planning easier for referees but is not essential. Our model could be used for ultrafast peer review of COVID19 papers (see M. A. Johansson and D. Saderi Nature 579, 29; 2020). Setting up a pool of referees dedicated to rapid review of key papers would help relieve pressure on overstretched individuals.

Katrina A. Lythgoe University of Oxford, UK.

Expert insight into current research

News & views Chemical biology

Protein degraders extend their reach Claire Whitworth & Alessio Ciulli

Molecules have previously been made that induce protein destruction inside cells. A new class of molecule now induces the degradation of membrane and extracellular proteins — opening up avenues for drug discovery. See p.291 Most drugs act by binding to a specific site in a target protein to block or modulate the protein’s function. The activity of many proteins, however, cannot be altered in this way. An emerging class of drug instead brings proteins into proximity with other molecules, which then alter protein function in unconventional ways1–3. One such approach uses drug molecules called protein degraders, which promote the tagging of proteins with ubiquitin, another small protein. Tagged proteins are then broken down into small peptide molecules by the cell’s proteasome machinery. But because the ubiquitin-mediated degradation pathway occurs inside the cell, protein degraders developed so far attack mainly intracellular targets. On page 291, Banik et al.4 now report a different mechanism that opens up extracellular and membrane-bound proteins for targeted degradation. The authors report protein degraders that they call lysosome-targeting chimaeras (LYTACs), which are bifunctional (they have two binding regions; Fig. 1). One end carries an oligoglycopeptide group that binds to a transmembrane receptor (the cation-independent mannose-6-phosphate receptor; CI-M6PR) at the cell surface. The other end carries either an antibody or a small molecule that binds to the protein targeted for destruction. These two regions are joined by a chemical linker. The formation of a trimeric CI-M6PR– LYTAC–target complex at the plasma membrane directs the complex for destruction by protease enzymes in membrane-enclosed organelles called lysosomes. LYTACs are conceptually related, but complementary, to proteolysis-targeting chimaeras5 (PROTACs) — another bifunctional class of protein degrader that mainly targets intracellular proteins by recruiting them to E3

ligases (the enzymes that tag proteins with ubiquitin). Banik et al. began by making LYTACs of varying size and linker composition, and which used a small molecule called biotin as the protein-binding component — biotin binds with exceptionally high affinity to avidin proteins. The authors observed that these LYTACs rapidly shuttled an extracellular fluorescent avidin protein to intracellular lysosomes in a way that required engagement

with CI-M6PR. When the authors replaced biotin with an antibody that recognizes apolipoprotein E4 (a protein implicated in neurodegenerative diseases), this protein was also internalized and degraded by lyso­somes. LYTACs can, therefore, repurpose antibodies from their normal immune function to direct extracellular proteins for lysosomal degradation. Next, Banik et al. investigated whether LYTACs could induce the degradation of membrane proteins that are targets for drug discovery. In several cancer cell lines, LYTACs did indeed induce the internalization and lyso­ somal degradation of the epidermal growth factor receptor (EGFR) — a membrane protein that drives cell proliferation by activating a signalling pathway. Depletion of EGFR levels by LYTACs in the cancer cell lines reduced signal activation downstream of EGFR, compared with the amount observed when EGFRs were blocked by antibodies alone. This result confirms a previously reported5 advantage of using target degradation in therapeutic applications, rather than target blocking. Similar outcomes were observed with LYTACs for other single-pass transmembrane proteins (proteins that span the cell membrane only once), including programmed

LYTAC Antibody Extracellular protein

Oligoglycopeptide

Cell membrane

CI-M6PR

Transmembrane protein

Transport vesicle

Lysosome

Degradation

Recycling

Figure 1 | Mechanism of action of lysosome-targeting chimaeras (LYTACs).  Banik et al.4 report LYTAC molecules, which consist of an oligoglycopeptide group (which binds to a cell-surface receptor, CI-M6PR) and an antibody that binds to a specific transmembrane or extracellular protein. The antibody can also be replaced by a small protein-binding molecule (not shown). On simultaneously binding to both CI-M6PR and the target protein, the resulting complex is engulfed by the cell membrane, which forms a transport vesicle. This carries the complex to a lysosome (an organelle that contains protein-degrading enzymes). The protein is degraded and the receptor is recycled; it remains to be seen whether the LYTAC is also degraded. LYTACs are potentially useful for therapeutic applications.

. d e v r e s e r s t h g i r l l A . d e t i m i L e r u t a N r e g n i r p S 0 2 0 2 ©

Nature | Vol 584 | 13 August 2020 | 193

News & views death ligand 1 (PD-L1), which helps cancer cells to evade the immune system. The next step will be to establish whether LYTACs can also induce the degradation of multi-pass proteins that span the membrane several times, such as the ubiquitous G-protein-coupled receptors and proteins that transport materials across membranes (ion channels and solute-carrier proteins, for example). If so, it will be interesting to compare the performance of LYTACs, which would bind to the extracellular domains of such proteins, with that of PROTACs, which can bind to the intracellular domains of these proteins (as was recently demonstrated6 for solute-carrier proteins). As with any new drug modality, there is scope for improvement. For example, Banik and colleagues’ first PD-L1-targeting LYTACs produced only partial degradation of the protein, which the authors attributed to low expression of CI-M6PR in the cell lines used. When the authors made a second type of LYTAC that incorporated a more potent PD-L1 antibody, degradation increased, albeit in cells that expressed greater levels of CI-M6PR than did the original cell lines. This shows that low abundance of the lysosome-shuttling receptor hijacked by the LYTAC (in this case, CI-M6PR) can reduce the effectiveness of these degraders. Similarly, the loss of core components of E3 ligases is a common mechanism by which cells become resistant to PROTACs7. Lysosome-shuttling receptors other than CI-M6PR could be used by LYTACs as alternatives, should resistance emerge. Degraders that target cell-type-specific receptors might also have improved safety profiles compared with conventional small-molecule therapeutics, which are not always cell-type selective. What sets PROTACs and LYTACs apart from conventional drugs is their mode of action. For example, after a PROTAC has brought about the destruction of a target protein, the PROTAC is released and can induce further cycles of ubiquitin tagging and degrada­ tion, thereby acting as a catalyst at low concentrations1,5. Mechanistic studies are now warranted to determine whether LYTACs also work catalytically. Another aspect of the mode of action of both PROTACs and LYTACs is that they bring two proteins together, to form a trimeric complex. A general feature of such processes is the hook effect, whereby trimer formation, and thereby the associated biological activity, decreases at high drug concentrations. This is because dimeric complexes generally form preferentially at high drug concentrations — an undesirable effect that can be alleviated by ensuring that all three components interact in such a way that trimer formation is more favourable than is dimer formation1. Kinetics also matters for protein degraders. For example, stable and long-lived trimeric

molecules that incorporate large groups, such as antibodies and oligoglycopeptides, during drug discovery, but this problem can be overcome by further engineering the structures of these groups11. Banik and colleagues’ new approach to degradation therefore warrants an all-hands-on deck approach. Scientists working in drug discovery will eagerly await the development of LYTACs and the emergence of other methods for the drug-induced degradation of proteins12. Is no protein beyond the reach of degraders? Claire Whitworth and Alessio Ciulli are in the Division of Biological Chemistry and Drug Discovery, School of Life Sciences, James Black Centre, University of Dundee, Dundee DD1 5EH, UK. e-mail: [email protected] 1. Maniaci, C. & Ciulli, A. Curr. Opin. Chem. Biol. 52, 145–156 (2019). 2. Deshaies, R. J. Nature 580, 329–338 (2020). 3. Gerry, C. J. & Schreiber, S. L. Nature Chem. Biol. 16, 369–378 (2020). 4. Banik, S. M. et al. Nature 584, 291–297 (2020). 5. Burslem, G. M. & Crews, C. M. Cell 181, 102–114 (2020). 6. Bensimon, A. et al. Cell Chem. Biol. 27, 728–739 (2020). 7. Zhang, L., Riley-Gillis, B., Vijay, P. & Shen, Y. Mol. Cancer Ther. 18, 1302–1311 (2019). 8. Roy, M. J. et al. ACS Chem. Biol. 14, 361–368 (2019). 9. Hanan, E. J. et al. J. Med. Chem. https://doi.org/10.1021/ acs.jmedchem.0c00093 (2020). 10. Mullard, A. Nature Rev. Drug Discov. 19, 435 (2020). 11. Chiu, M. L. & Gilliland, G. L. Curr. Opin. Struct. Biol. 38, 163–173 (2016). 12. Takahashi, D. et al. Mol. Cell 76, 797–810 (2019). This article was published online on 29 July 2020.

Ecology

Rethinking extinctions that arise from habitat loss Joaquín Hortal & Ana M. C. Santos

Does the loss of species through habitat decline follow the same pattern whether the area lost is part of a large or a small habitat? An analysis sheds light on this long-running debate, with its implications for conservation strategies. See p.238 Understanding how habitat size affects the abundance of all the species living in a community provides ecological insights and is valuable for developing strategies to boost biodiversity. On page 238, Chase et al.1 report results that might help to settle a long-running debate about the relationship between the area of a habitat and the diversity of species it can host. Land transformation by human activity is a major component of global change. The loss of natural habitats reduces the local diversity and abundance of species2, and has been

. d e v r e s e r s t h g i r l l A . d e t i m i L e r u t a N r e g n i r p S 0 2 0 2 ©

194 | Nature | Vol 584 | 13 August 2020

complexes that involve PROTACs accelerate target degradation, improving drug potency and selectivity 8. It will be crucial to understand how the complexes formed by LYTACs can be optimized to improve degradation activity. PROTACs and LYTACs are larger molecules than conventional drugs. As a result of their size, PROTACs often do not permeate well through biological membranes, which can make them less potent drugs than the biologically active groups they contain. Size should be less of a problem for LYTACs because they do not need to cross the cell membrane, although they would still need to pass through biological barriers to combat diseases of the central nervous system. The development of lysosomal degraders that are smaller and less polar than LYTACs — and therefore more able to pass through membranes — will be eagerly anticipated. Small ‘glue’ molecules that bind to E3 ligases can already do the same job as PROTACs9. Targeted protein degradation is a promising therapeutic strategy, and the first PROTACs are currently in clinical trials10. LYTACs will need to play catch-up, but they have earned their place as a tool poised to expand the range of proteins that can be degraded. Their development as therapies will require an understanding of their behaviour in the human body — their pharmacokinetics, toxicity, and how they are metabolized, distributed and excreted, for example. It can be challenging to optimize the biological behaviour of

implicated in more than one-third of animal extinctions worldwide between 1600 and 1992 (ref. 3). A report from the Intergovernmental Science-Policy Platform on Biodiversity and Ecosystem Services estimates that currently more than half a million species — about 9% of all terrestrial species — might lack the amount of habitat needed for their long-term survival4. Moreover, their disappearance would compromise many key ecosystem services, such as pollination or the control of pests or disease-causing agents. The effect of habitat loss on biodiversity has

a Large ecosystem

b Small ecosystem

Sampled area Passive sampling

Ecosystem decay (individuals)

Ecosystem decay (evenness)

Species

been conventionally estimated on the basis of the relationship between area and species richness, which was first described more than 150 years ago5. This seemingly universal relationship is simple: the larger a given habitat’s area, the more species it holds, although the number of species increases with area in a nonlinear way6. There is a limit to the number of individuals of ecologically similar species that can persist in an area, owing to the limited resources that it harbours7. When a habitat loses part of its area, therefore, for many species, it also loses its capacity to support populations that are large enough to be viable. These species become extinct as habitat area diminishes with land-use intensification8. Chase and colleagues propose an elegant and simple approach to account for the dynamics of communities occupying habitat patches of different size. Rather than considering only the overall number of species in each habitat fragment, the authors focused on the number and relative abundance of different species in samples obtained from such fragments. This allows the structure of ecological communities to be compared directly2, while avoiding problems that can arise when taking into account the differences in the effort needed to sample large and small areas9. The authors’ approach also allows a comparison of variations in the relative abundance of individuals of all species, a measure of community structure that is associated with ecosystem dynamics10. Thanks to this method, Chase et al. could distinguish between three patterns of change that might occur as an outcome of habitat loss (Fig. 1). In the pattern described by the ‘passive sampling’ model, the structure of the community remains the same in large and small fragments. Therefore, each sample provides similar species richness (the number of species), abundance (the number of individuals) and evenness (the allocation of individuals to the different species), regardless of the total habitat size. In this case, species decline will mirror the loss of habitat area under the classical species—area theory5, and the total number of species in the entire fragment would depend solely on its size. The other two patterns are described as types of ecosystem decay — a hypothesis proposing that a habitat that shrinks undergoes a disproportionately high loss of organisms compared with the loss of habitat area. One type of ecosystem decay is proposed to occur owing to excessive loss of individuals. Smaller habitat fragments will contain fewer individuals per sample than will larger ones, and all species are equally affected. This generates communities with fewer species in smaller fragments, but no changes in the relative abundance of species per sample between small and large fragments. The other type of ecosystem decay occurs

Abundance per sample

Abundance per sample

Figure 1 | Assessing how habitat size affects ecosystem dynamics. Understanding the relationship between a decline in habitat area and the effect on species is crucial for designing conservation strategies. a, b, Chase et al.1 analysed studies that sampled species in particular habitats. The authors compared the diversity of organisms, such as insects, in samples obtained from large ecosystems (a) with samples taken from the same sampling area in a smaller fragments of the same type of habitat (b). These graphs show hypothetical results for species abundance per sample, and different species are shown in different colours. This method enabled the authors to distinguish between three possible outcomes as habitats become smaller. In the passive-sampling model, species are equally distributed in habitat fragments of any size, so the richness, abundance and relative species prevalence (evenness) per sample is constant, regardless of the total habitat size. In the ecosystem-decay (individuals) model, samples from smaller fragments have fewer individuals and species per sample than do samples from larger fragments, and all species abundances decline in a similar way as habitat is lost. In the ecosystem-decay (evenness) model, species vary in their response to habitat loss, and there is a change in their relative abundances. Chase et al. find that ecosystem decay, usually following the evenness model, is the best match for the observed data.

owing to uneven changes in relative species abundances coupled to species loss. In this scenario, the species present have different responses to habitat loss, and therefore species become relatively more or less abundant in smaller fragments than in larger fragments. Their relative abundance becomes more uneven in samples from smaller fragments as some species increase their numerical dominance, impoverishing the community and causing it to become species poor. Using data from around 120 humantransformed landscapes worldwide, Chase et  al. show that, in general, samples from small fragments of natural habitat have fewer individuals, fewer species and a more uneven abundance of species than samples taken from larger fragments do. This outcome is consistent with a generalized pattern of ecosystem decay, mainly as a result of a decline in evenness (see Fig. 1), and this result holds, regardless of the type of habitat or organism studied. This implies that the alteration of natural habitats causes major functional changes in ecosystem dynamics that go beyond simply losing populations and species. Therefore, current estimates of extinctions associated with habitat loss made on the basis of the passive-sampling model might be underestimating not only the number of

species that are threatened or already gone, but also the consequences of their loss for ecological functioning and the provision of ecosystem services. Changes in biodiversity after habitat loss alter many ecological processes11, eventually causing catastrophic effects that accelerate the extinction process12. But local extinctions are often not immediate. Some species persist with reduced abundances and declining population dynamics — known as ‘extinction debt’ — that lasts until the final individuals perish13. This causes an uneven distribution of species abundance that is vividly demonstrated by Chase and colleagues’ method. Their analysis reveals a few ‘winning’ species that dominate the community in small habitats, and a very large number of rare species, many of which are probably heading towards extinction. Declining species can be replaced by others coming from the neighbouring human-altered landscape, particularly in habitat edges14, producing what are described as ‘edge effects’ that are comparatively more important in smaller fragments. Indeed, in the early stages of land transformation, communities in small fragments are more different from pristine communities than are those in large fragments, with communities in small fragments becoming more similar to those in large

. d e v r e s e r s t h g i r l l A . d e t i m i L e r u t a N r e g n i r p S 0 2 0 2 ©

Nature | Vol 584 | 13 August 2020 | 195

News & views fragments over time, as they recover from the effect of land transformation2. According to Chase and colleagues, the degree of decay in diversity and species abundance found between large and small fragments is smaller in the older or ‘softly’ transformed European landscapes than in the more recently and dramatically transformed North American ones. This indicates that, over time, species moving in from the edges of the human-altered habitats might compensate, at least in part, for the ecological functions carried out by native species in larger habitats, causing small fragments to reach a new — yet different — ecological balance. Although this work underscores the key role of habitat area in maintaining eco­system processes, there is little exploration of how these processes are altered by habitat loss. Species from higher trophic levels (the upper levels of the food chain), such as predators, require larger areas to maintain their populations compared with species from lower trophic levels, so the number of individuals supported by smaller habitat fragments might not suffice to maintain populations of top predators or consumers, and hence would produce shorter food chains and alter the ecosystem structure15. Differences in extinction rates between trophic levels can cause striking changes in ecosystem functioning at habitat edges16, jeopardizing the functioning and ecosystem-service provision as natural habitats diminish in size11. Chase and colleagues’ results call for a reconsideration of the debate over whether a single large area devoted to conservation would preserve more species than would several small ones that combine to make up the same total size17. Some current evidence suggests that one continuous habitat might host fewer species than do many small patches that total the same area18. However, the large ecological changes that these small fragments might undergo could end up resulting in massive reductions in ecosystem function and, ultimately, increased extinction rates of native species over the long term compared with the case for a single, large protected area. Chase and colleagues’ approach is good for providing a general overview of the extent of these effects, but to understand exactly how ecological processes are changing locally, a higher level of detail will be needed. This will require going beyond the studies of trophic chains14,16 to assess more-complex food webs15, and to gather information on changes in species’ functional responses and trait diversity in increasingly smaller habitats. Ultimately, this information will reveal which ecological processes are decaying, and what the consequences of such ecosystem decay are for the maintenance of fully functional biodiversity.

1. Chase, J. M., Blowes, S. A., Knight, T. M., Gerstner, K. & May, F. Nature 584, 238–243 (2020). 2. Newbold, T. et al. Nature 520, 45–50 (2015). 3. World Conservation Monitoring Centre. Global Biodiversity: Status of the Earth’s Living Resources 199 (Chapman & Hall, 1992). 4. Díaz, S. et al. (eds) The Global Assessment Report on Biodiversity and Ecosystem Services: Summary for Policymakers (IPBES, 2019). 5. Rosenzweig, M. L. Species Diversity in Space and Time

(Cambridge Univ. Press, 1995). 6. Arrhenius, O. J. Ecol. 9, 95–99 (1921). 7. Wright, D. H. Oikos 41, 496–506 (1983). 8. Di Marco, M., Venter, O., Possingham, H. P. & Watson, J. E. M. Nature Commun. 9, 4621 (2018). 9. Chase, J. M. et al. Front. Biogeogr. 11, e40844 (2019). 10. Simons, N. K. et al. Agric. Ecosyst. Environ. 237, 143–153 (2017). 11. Dobson, A. et al. Ecology 87, 1915–1924 (2006). 12. Fischer, J. & Lindenmayer, D. B. Global Ecol. Biogeogr. 16, 265–280 (2007). 13. Tilman, D., May, R. M., Lehman, C. L. & Nowak, M. A. Nature 371, 65–66 (1994). 14. Didham, R. K., Lawton, J. H., Hammond, P. M. & Eggleton, P. Phil. Trans. R. Soc. B 353, 437–451 (1998). 15. Holt, R. D. in The Theory of Island Biogeography Revisited (eds Losos, J. B. & Ricklefs, R. E.) 143–185 (Princeton Univ. Press, 2010). 16. Harrison, M. L. K. & Banks-Leite, C. Conserv. Biol. https://doi.org/10.1111/cobi.13438 (2019). 17. Simberloff, D. S. & Abele, L. G. Science 191, 285–286 (1976). 18. Fahrig, L. Glob. Ecol. Biogeogr. 29, 615–628 (2020). This article was published online on 29 July 2020.

Mechanobiology

Stretch exercises for stem cells expand the skin Matthias Rübsam & Carien M. Niessen

Stretching the skin of mice reveals that mechanical strain is communicated by a subpopulation of stem cells that proliferate and promote mechanical resistance, and so generate extra skin. See p.268 The cells of our bodies are exposed to a range of mechanical forces — including compression, shear and stretching — that they must resist to maintain tissue integrity and function. For example, skin responds to stretching forces by expanding. Physicians have exploited this particular response for more than 60 years1, implanting stretching devices in the skin to cause tissue expansion for plastic surgery or to repair birth defects2. But exactly how mechanical strain creates extra tissue in a living organism has not been known. On page 268, Aragona et al.3 now provide compelling insights (at the molecular, single-cell and cell-population level) into how stem cells in the skin of mice sense and communicate stretch to make new tissue. The surface of the skin — a multi-­layered tissue called the epidermis — protects organisms against dehydration and environ­ mental stresses, including mechanical challenges. To ensure lifelong protection, the epidermis is constantly renewed through the generation of new stem cells in its basal layer. This renewal is balanced with differentiation and the movement of stem cells to generate the upper, barrier-forming layers of the epidermis. Ultimately, the barrier-forming cells are shed

. d e v r e s e r s t h g i r l l A . d e t i m i L e r u t a N r e g n i r p S 0 2 0 2 ©

196 | Nature | Vol 584 | 13 August 2020

Joaquín Hortal is in the Department of Biogeography and Global Change, Museo Nacional de Ciencias Naturales, Spanish National Research Council, Madrid 28006, Spain. Ana M. C. Santos is in the Department of Ecology, Autonomous University of Madrid, Madrid 28049, Spain, and at the Centro de Investigación en Biodiversidad y Cambio Global, Autonomous University of Madrid. e-mails: [email protected]; [email protected]

from the surface, to be replaced by new cells. Aragona et al. set out to examine how the epidermis responds to strain. The group positioned a device used in human surgeries — a self-inflating gel — under the skin of mice. They then examined indicators of force perception, including changes in cell shape, the structure of a mechanosensitive protein called α-catenin, and a network of keratin proteins that provides cells with mechanical resilience. This analysis revealed that epidermal stem cells do indeed sense and respond to strain. The authors observed a temporary increase in stem-cell division, followed by thickening of the epidermis. Thus, increased stem-cell renewal fuels stem-cell differentiation. The two effects combine to maintain a functional barrier at the same time as extra skin is generated. The researchers next genetically engineered cells in the basal epidermal layer such that the stem cells and their descendants were fluorescently marked. Tracking of these cell lineages over time confirmed that stretching tips the renewal–differentiation balance in favour of making more stem cells. This explains why the epidermis expands in response to stretching.

https://doi.org/10.1038/d41586-020-02158-y

News & views Mechanobiology

Stretch exercises for stem cells expand the skin Matthias Rübsam & Carien M. Niessen

Stretching the skin of mice reveals that mechanical strain is communicated by a subpopulation of stem cells that proliferate and promote mechanical resistance, and so generate extra skin. The cells of our bodies are exposed to a range of mechanical forces — including compression, shear and stretching — that they must resist to maintain tissue integrity and function. For example, skin responds to stretching forces by expanding. Physicians have exploited this particular response for more than 60 years1, implanting stretching devices in the skin to cause tissue expansion for plastic surgery or to repair birth defects2. But exactly how mechanical strain creates extra tissue in a living organism has not been known. Writing in Nature, Aragona et al.3 now provide compelling insights (at the molecular, single-cell and cell-population level) into how stem cells in the skin of mice sense and communicate stretch to make new tissue. The surface of the skin — a multi-­layered tissue called the epidermis — protects organisms against dehydration and environ­ mental stresses, including mechanical challenges. To ensure lifelong protection, the epidermis is constantly renewed through the generation of new stem cells in its basal layer. This renewal is balanced with differentiation and the movement of stem cells to generate the upper, barrier-forming layers of the epidermis. Ultimately, the barrier-forming cells are shed from the surface, to be replaced by new cells. Aragona et al. set out to examine how the epidermis responds to strain. The group positioned a device used in human surgeries — a self-inflating gel — under the skin of mice. They then examined indicators of force perception, including changes in cell shape, the structure of a mechanosensitive protein called α-catenin, and a network of keratin proteins that provides cells with mechanical resilience. This analysis revealed that epidermal stem cells do indeed sense and respond to strain.

The authors observed a temporary increase in stem-cell division, followed by thickening of the epidermis. Thus, increased stem-cell renewal fuels stem-cell differentiation. The two effects combine to maintain a functional barrier at the same time as extra skin is generated. The researchers next genetically engineered cells in the basal epidermal layer such that the stem cells and their descendants were fluorescently marked. Tracking of these cell lineages over time confirmed that stretching tips the renewal–differentiation balance in favour of a Normal

Actomyosin cytoskeleton

making more stem cells. This explains why the epidermis expands in response to stretching. Aragona et al. demonstrated that force changes stem cells at the molecular level in several ways. First, stretching increased the expression of genes involved in cell–cell adhesion, which have been shown to communicate force in vitro4. Second, expression of components of the actomyosin cytoskeleton — a network of protein filaments that generates contractile forces in cells5 — was increased. Third, stretching promoted signalling through the EGF–Map kinase–ERK pathway (a cascade of proteins that promotes growth). The researchers also assessed changes in chromatin, the DNA–protein complex that parcels up the genome in cells; such changes can lead to altered gene expression. This analysis revealed that stretch induced the expression of a network of regulatory genes that links stem-cell proliferation to skin maintenance. The authors then examined how strain alters gene-expression profiles of single epidermal stem cells, by sequencing the cells’ RNA. This revealed that only a subpopulation of stem cells undergoes the molecular changes associated with a stretched state. Why might this be? Perhaps those that take on the stretched state experience greater force. Alternatively, maybe stem cells exist in varying biochemical states, and thus are more or less sensitive to b Stretched

YAP/TAZ/MAL

Stem-cell proliferation Differentiation Skin expansion

Stem-cell layer

Gel

Figure 1 | How skin stem cells respond to stretching. a, The surface of skin is a multi-layered tissue called the epidermis, which has stem cells in its basal layer. Like all cells, the stem cells have a contractile network of protein filaments called the actomyosin cytoskeleton, and express the transcription factors YAP, TAZ and MAL. b, Aragona et al.3 placed an expanding gel under the skin of mice. They report that, in a subset of epidermal stem cells, the actomyosin cytoskeleton is reorganized. This somehow triggers movement of YAP, TAZ and MAL to the nucleus. The proteins induce gene-expression changes that promote an increase in both stem-cell proliferation and differentiation into cells that move into the upper layers of the epidermis. This dual response leads to expansion of skin tissue without compromising the barrier function of the epidermis.

. d e v r e s e r s t h g i r l l A . d e t i m i L e r u t a N r e g n i r p S 0 2 0 2 ©

Nature  |   1

News & views force. Or local differences in stem-cell shape and mechanics could determine how each cell responds to stretch. Answering this question will require measurements of cellular forces and stiffness in vivo, which is still a major challenge. In addition, it remains unclear whether the stretched stem cells alone are driven to proliferate — or whether these cells then induce expansion of surrounding stem cells. Aragona et al. next genetically engineered mice to lack Diaph3 and Myh9, genes involved in regulation of the actomyosin cytoskeleton. Without these genes, stem-cell responses to stretch were absent, leading to a barrier defect in the animals. The group observed similar effects in animals engineered to lack the genes encoding YAP and TAZ, and/or in which MAL was inhibited — these three transcription factors normally move to the nucleus to regulate gene expression in response to mechanical signals6. Next, the authors examined YAP, TAZ and MAL in mice lacking Diaph3 and Myh9. The transcription factors did not move to the nucleus in response to stretch in these animals. Thus, in normal skin, stretch reorganizes the actomyosin cytoskeleton to promote entry of YAP, TAZ and MAL into the nucleus. These proteins then coordinate transcriptional programs that promote skin growth and barrier formation (Fig. 1). Finally, the researchers inhibited MAL or the EGF-pathway component ERK in their animals. Inhibition of either protein blocked stem-cell proliferation, but only MAL inhibition led to loss of the ‘stretched’ molecular state in a subset of stem cells. Thus, MAL regulates variable cell response to strain, whereas both ERK and MAL are necessary to promote the self-renewal of stem cells. Whether ERK is downstream of YAP, TAZ and MAL, or directly activated by the cytoskeleton7, and whether EGF–ERK signal­ling promotes adaptation to strain in the upper epidermal layers to maintain barrier function during skin expansion8, remain open questions. Overall, Aragona and colleagues’ data support a model in which stretch is initially

. d e v r e s e r s t h g i r l l A . d e t i m i L e r u t a N r e g n i r p S 0 2 0 2 ©

2 | Nature

sensed by a subset of stem cells. These cells, through cytoskeletal reorganization and changes in gene expression, coordinate stem-cell renewal and differentiation with adaptation to the mechanical force being experienced. This response guarantees that the skin can maintain its protective function while expanding. The research opens several avenues for future research. First, what is the contribution of other compartments of the skin (such as the upper, barrier-forming epidermal layers, or the thick dermal layer that underlies the epidermis) in sensing and communicating stretch? The authors’ analysis of mice lacking YAP or MAL suggests that stretch also induces a cytoskeletal response in differentiated cells of the upper epidermal layers. Stem-cell differentiation and upward movement can trigger renewal of neighbouring stem cells9, thus begging the question of whether the cells immediately above the basal layer are also required for stem-cell responses to strain. Second, in vitro experiments have demonstrated10 that chromatin regulation in the nucleus is key to maintaining stem-cell identity and genome integrity under mechanical stress. Aragona and co-workers’ skin-expander model will now allow us to explore these mechanisms in vivo. Third, current models of stem-cell renewal postulate that a single stem cell is equally capable of undergoing renewal or differentiation. However, Aragona and colleagues’ lineage-tracing experiments revealed that the number of cells derived from one stem cell (called basal-cell clones) tended to be even. This bias towards clones that have even numbers of cells became much more pronounced on stretching. How the stretched state promotes even-numbered clones is unclear. The authors propose that this bias can be explained by a model in which stem cells exist in two-progenitor units, in which one stem cell is committed to renewal and the other to differentiation. Communication within and

between units would balance the loss of cells through differentiation with renewal. The group performed a mathematical comparison, which indicated that the even-numbered-cell bias and clone dynamics they observed were more consistent with a two-progenitor than with a single-progenitor model. However, the jury on this is still out, because a recent study has provided fresh evidence for the one-progenitor model11. The current work provides a major step forward in our understanding of how force is interpreted at the single-cell level in living organisms. Furthermore, it should encourage others to explore the use of mechanical signals to generate extra skin — not only for reconstructive surgery, but also for diseases associated with impaired regeneration. Matthias Rübsam and Carien M. Niessen are in the Cologne Excellence Cluster on Stress Responses in Aging-associated Diseases (CECAD), University of Cologne, 50931 Cologne, Germany. M.R. is also at the Max Planck Institute for Biology of Aging, Cologne. C.M.N. is also in the Department Cell Biology of the Skin and the Center for Molecular Medicine Cologne, University of Cologne. e-mails: [email protected]; [email protected]

1. Neumann, C. G. Plast. Reconstruct. Surg. 19, 124–130 (1957). 2. Zöllner, A. M., Holland, M. A., Honda, K. S., Gosain, A. K. & Kuhl, E. J. Mech. Behav. Biomed. Mater. 28, 495–509 (2013). 3. Aragona, M. et al. Nature https://doi.org/10.1038/s41586020-2555-7 (2020). 4. Ladoux, B. & Mège, R.-M. Nature Rev. Mol. Cell Biol. 18, 743–757 (2017). 5. Chug, P. & Paluch, E. K. J. Cell Sci. 131, jcs186254 (2018). 6. Panciera, T., Azzolin, L., Cordenonsi, M. & Piccolo, S. Nature Rev. Mol. Cell Biol. 18, 758–770 (2017). 7. Hirata, H. et al. EMBO Rep. 16, 250–257 (2015). 8. Rübsam, M. et al. Nature Commun. 8, 1250 (2017). 9. Mesa, K. R. et al. Cell Stem Cell 23, 677–686 (2018). 10. Nava, M. M. et al. Cell 181, 800–817 (2020). 11. Piedrafita, G. et al. Nature Commun. 11, 1429 (2020).

News & views organisms. Furthermore, it should encourage others to explore the use of mechanical signals to generate extra skin — not only for reconstructive surgery, but also for diseases associated with impaired regeneration. Matthias Rübsam and Carien M. Niessen are in the Cologne Excellence Cluster on Stress Responses in Aging-associated Diseases (CECAD), University of Cologne, 50931 Cologne, Germany. M.R. is also at the Max Planck Institute for Biology of Aging, Cologne. C.M.N. is also in the Department Cell Biology of the Skin and the Center for Molecular Medicine Cologne, University of Cologne.

e-mails: [email protected]; [email protected] 1. Neumann, C. G. Plast. Reconstruct. Surg. 19, 124–130 (1957). 2. Zöllner, A. M., Holland, M. A., Honda, K. S., Gosain, A. K. & Kuhl, E. J. Mech. Behav. Biomed. Mater. 28, 495–509 (2013). 3. Aragona, M. et al. Nature 584, 268–273 (2020). 4. Ladoux, B. & Mège, R.-M. Nature Rev. Mol. Cell Biol. 18, 743–757 (2017). 5. Chug, P. & Paluch, E. K. J. Cell Sci. 131, jcs186254 (2018). 6. Panciera, T., Azzolin, L., Cordenonsi, M. & Piccolo, S. Nature Rev. Mol. Cell Biol. 18, 758–770 (2017). 7. Hirata, H. et al. EMBO Rep. 16, 250–257 (2015). 8. Rübsam, M. et al. Nature Commun. 8, 1250 (2017). 9. Mesa, K. R. et al. Cell Stem Cell 23, 677–686 (2018). 10. Nava, M. M. et al. Cell 181, 800–817 (2020). 11. Piedrafita, G. et al. Nature Commun. 11, 1429 (2020). This article was published online on 29 July 2020.

Biogeochemistry

Carbon loss from tropical soils increases on warming Eric A. Davidson

Plots of tropical forest soils were warmed by 4 °C for two years to observe the effects on soil carbon emissions. The increase in efflux of carbon dioxide was larger than expected — a result with worrying implications for climate change. See p.234

Figure 1 | Preparing to use a heating device in an experiment. Nottingham et al.3 buried heating equipment in plots in a topical forest in Panama and monitored the affect of warming on carbon dioxide release from the soil.

. d e v r e s e r s t h g i r l l A . d e t i m i L e r u t a N r e g n i r p S 0 2 0 2 ©

198 | Nature | Vol 584 | 13 August 2020

to which soil carbon can bind and thereby be protected from decomposition by microbial enzymes; water content, which affects the diffusion of carbon to microbial enzymes;

GEETHA IYER

The carbon stored in soil could have a big impact on climate change. The global flux of carbon in and out of soils is six to ten times greater than the emissions of carbon dioxide produced as a result of human activities1,2. The inputs of carbon to soils from plant detritus (dead wood, leaves and roots) roughly balance the losses to the atmosphere produced by the respiration of soil microorganisms that feed on that material. However, just a 1% imbalance of global soil-carbon effluxes over influxes would equal about 10% of global anthropogenic carbon emissions. Carbon in tropical soils was thought to be less vulnerable to loss under climate change than is soil carbon at higher latitudes, but experimental evidence for this was lacking. On page 234, Nottingham et al.3 report that tropical-forest soils might be more vulnerable to warming than was thought. Various soil management practices (such as changes of land use and of tillage methods) can influence the amount of soil carbon present4, but climate affects the respiration rate of the microbes that feed on soil carbon, and hence the CO2 efflux from soil5. If the net efflux of carbon from soils to the atmosphere increases in a warming world, a positive feedback will accelerate the warming. Soils are amazingly diverse and differ in several respects that affect microbial respiration6. These include: the amount of minerals

the amount, timing and quality of the plant detritus going into the soil; and genetic variation in soil microbial communities. The intrinsic temperature sensitivity of microbial respiration reactions indicated by theory and demonstrated in the laboratory (where other factors are not limiting) therefore often varies from the apparent temperature sensitivity measured in real-world settings5,6. Several in situ soil-warming experiments have yielded insights into the effects of temperature on CO2 efflux from soil in temperate and boreal regions6,7, but such research is logistically more challenging to implement in tropical forests. Nottingham and co-workers now present results of a soil-warming experiment (Fig. 1). The authors placed warming rods around the perimeter of undisturbed soil plots in a tropical forest on Barro Colorado Island, Panama, and increased the temperature of the whole soil profile (to a depth of 1.2 metres) by 4 °C for two years. They measured CO2 efflux using chambers periodically placed over the soil, and observed an unexpectedly large increase (55%) in soil CO2 emissions. By excluding roots from the soil under some of the chambers, the authors determined that most of the increased CO2 efflux was due to a greater-than-expected increase in the respiration of soil microbes. The dependence of reaction rates on temperature, including the rates of enzymatic reactions, is described by the Arrhenius equation5. According to this equation, the fractional increase in reaction rate is less for a temperature increase of one degree Celsius at higher temperatures than at lower temperatures. This suggests that the response of soil microbial respiration to temperature changes

a

b 6

4

Observed Q10

3 4 Activation energy 60 kJ mol–1

2

Q10

Microbial CO2 efflux (micromol per square metre per second)

should be less pronounced in tropical climates than it is in temperate or boreal climates. The Arrhenius relationship breaks down for enzymatic reactions at temperatures beyond those that are optimal for enzyme function, and therefore should not be extrapolated beyond temperatures that enzymes normally experience. But such extreme temperatures were not reached in Nottingham and colleagues’ experiment. Each enzyme-catalysed reaction has a characteristic activation energy (Ea, simplistically defined as the minimum energy needed for the reaction to occur), but the Ea values of the vast majority of respiratory enzymes fall within a fairly narrow range8, averaging about 60 kilojoules per mol. If this common Ea value is used in an Arrhenius function to calculate how the average microbial CO2 efflux from Nottingham and colleagues’ plots varies with temperature (fitting the resulting curve to the efflux recorded in the non-warmed plots), the observed efflux from the warmed plots is found to be much higher than the calculated value (Fig. 2a). The Ea value would have to be an unlikely 97 kJ mol−1 to make the Arrhenius function fit with the efflux observed from the warmed plots. A short-hand indicator of the temperature sensitivity of soil CO2 efflux is Q10, the multiplication factor of the increase observed when the temperature increases by 10 °C. The Q10 for microbial CO2 efflux in Nottingham and co-workers’ experiment was 3.6, which is well above expectations for most respiratory enzymatic processes on the basis of common Ea values and the Arrhenius equation (Fig. 2b). Temperature sensitivities as high as this indicate that the increased efflux in the warmed plots probably cannot be attributed only to the kinetic response of enzymes to warming — other confounding factors must also co-vary with temperature9. Indeed, in in vitro assays, Nottingham et al. report Q10 values of 2 or less for specific respiratory enzymes in soil samples from the studied plots. The authors rule out the possibility that significant confounding effects arose during their warming experiment as a result of the modest changes in soil-water content and microbial carbon-use efficiency. The cause of the unexpectedly high observed apparent temperature sensitivity therefore remains a mystery. These results suggest that caution is needed when using only an Arrhenius or Q10 function to describe the temperature responses of soil microbial respiration — which most numerical models of Earth’s systems and their carbon cycling responses to climate change unfortunately do10. The Arrhenius function (where Ea is about 60 kJ mol−1) predicts a modest decrease in Q10 from about 2.6 at temperatures typical of high-latitude ecosystems to 2.2 at temperatures typical of tropical forests (Fig. 2b), whereas the current study and others 2,5,6

1

97 kJ mol–1 0

0

10 20 Temperature (°C)

2

30

0

0

10 20 Temperature (°C)

30

Figure 2 | Theoretical and observed temperature sensitivity of CO2 efflux from tropical soils. a, Nottingham et al.3 observed an increase in CO2 efflux from tropical soils warmed in situ from an ambient temperature of 26 °C to 30 °C (dark blue circles). The pale blue line shows the theoretical dependence of CO2 efflux associated with soil microbial respiration, calculated using an Arrhenius function5 (which describes the kinetic response of enzymes to temperature) and assuming that the activation energy (Ea; simplistically, the minimum amount of energy needed for a reaction to occur) of the microbial respiratory reactions is 60 kilojoules per mol, which is typical for such reactions8. The observed increase in efflux on warming is larger than that predicted from this Arrhenius function. To fit the observations, Ea would have to be 97 kJ mol−1 (dark blue line), which is unlikely for respiratory enzymes. b, The Arrhenius function also predicts much lower Q10 values (red line; the multiplication factor of the increase in efflux observed when the temperature increases by 10 °C) than that observed by the authors at 30 °C (red circle). The authors’ results suggest that unknown factors co-vary with temperature to increase CO2 efflux, and that the kinetic responses of respiratory enzymes, alone, are insufficient to account for their observations. 

report a broader range of observed Q10 values. Nottingham and colleagues’ results suggest that other factors affecting soil CO2 efflux merit further study. Efflux responses to experimental warming often decline after a few years, perhaps because the readily decomposable components of soil carbon that act as substrates for respiration reactions become depleted, or because of shifts in microbial community structures6,7. The Panamanian forest soil studied in the present work is relatively fertile as a result of the

“Tropical forests are unlikely to continue indefinitely to be carbon sinks as the world warms.” historical deposition of volcanic ash; this soil fertility might have enabled high rates of inputs of plant detritus to the soil, preventing soil microbial respiration from being limited by the availability of substrates. Indeed, the authors report an increase in soil microbial biomass in the warmed plots compared with control plots, indicating that warming might have increased the total microbial enzymatic capacity — thereby increasing CO 2 efflux beyond the expected kinetic response of the pre-existing enzymatic capacity. Besides the clear take-home message that the responses of soil respiration processes under climate change should not be represented in Earth-system models only by

simple Q10 or Arrhenius functions, Nottingham and co-workers’ study adds to recently accumulating evidence that tropical forests are unlikely to continue indefinitely to be carbon sinks as the world warms11. Tropical soil carbon does not receive as much attention as do the large and vulnerable soil-carbon stocks at high latitudes, which pose major concerns as a potential source of positive feedback to climate change12. But tropical-forest soils also contain substantial carbon stores that might be more vulnerable to loss in a warming world than was previously recognized. Eric A. Davidson is in the Appalachian Laboratory, University of Maryland Center for Environmental Science, Frostburg, Maryland 21532, USA. e-mail: [email protected] 1. Friedlingstein, P. et al. Earth Syst. Sci. Data 11, 1783–1838 (2019). 2. Bond-Lamberty, B. & Thomson, A. Biogeosciences 7, 1915–1926 (2010). 3. Nottingham, A. T., Meir, P., Velasquez, E. & Turner, B. L. Nature 584, 234–237 (2020). 4. Lal, R. Carbon Mgmt 4, 439–462 (2014). 5. Davidson, E. A. & Janssens, I. A. Nature 440, 165–173 (2006). 6. Conant, R. T. et al. Glob. Change Biol. 17, 3392–3404 (2011). 7. Carey, J. C. et al. Proc. Natl Acad. Sci. USA 113, 13797–13802 (2016). 8. Gillooly, J. F., Brown, J. H., West, G. B., Savage, V. M. & Charnov, E. L. Science 293, 2248–2251 (2001). 9. Davidson, E. A., Janssens, I. A. & Luo, Y. Glob. Change Biol. 12, 154–164 (2006). 10. Wieder, W. R. et al. Glob. Biogeochem. Cycles 29, 1782–1800 (2015). 11. Sullivan, M. J. P. et al. Science 368, 869–874 (2020). 12. Turetsky, M. R. et al. Nature 569, 32–34 (2019).

. d e v r e s e r s t h g i r l l A . d e t i m i L e r u t a N r e g n i r p S 0 2 0 2 ©

Nature | Vol 584 | 13 August 2020 | 199

Article

A dynamically cold disk galaxy in the early Universe https://doi.org/10.1038/s41586-020-2572-6

F. Rizzo1 ✉, S. Vegetti1, D. Powell1, F. Fraternali2, J. P. McKean2,3, H. R. Stacey1,2,3 & S. D. M. White1

Received: 26 September 2019 Accepted: 4 June 2020 Published online: 12 August 2020 Check for updates

The extreme astrophysical processes and conditions that characterize the early Universe are expected to result in young galaxies that are dynamically different from those observed today1–5. This is because the strong effects associated with galaxy mergers and supernova explosions would lead to most young star-forming galaxies being dynamically hot, chaotic and strongly unstable1,2. Here we report the presence of a dynamically cold, but highly star-forming, rotating disk in a galaxy at redshift6 z = 4.2, when the Universe was just 1.4 billion years old. Galaxy SPT–S J041839–4751.9 is strongly gravitationally lensed by a foreground galaxy at z = 0.263, and it is a typical dusty starburst, with global star-forming7 and dust properties8 that are in agreement with current numerical simulations9 and observations10. Interferometric imaging at a spatial resolution of about 60 parsecs reveals a ratio of rotational to random motions of 9.7 ± 0.4, which is at least four times larger than that expected from any galaxy evolution model at this epoch1–5 but similar to the ratios of spiral galaxies in the local Universe11. We derive a rotation curve with the typical shape of nearby massive spiral galaxies, which demonstrates that at least some young galaxies are dynamically akin to those observed in the local Universe, and only weakly affected by extreme physical processes.

SPT–S J041839–4751.9 (hereafter SPT0418–47) was observed with the Atacama Large Millimeter/submillimetre Array (ALMA) to image the thermal continuum emission from the galaxy’s dust and the emission from the 158-μm fine-structure line of ionized carbon [C ii] (Fig. 1a). The [C ii] line is typically the brightest fine-structure line emitted in star-forming galaxies, and it is the main gas coolant, coming from multiple phases of the inter-stellar medium, including the warm ionized, the warm and cold neutral atomic, and the dense molecular medium12. We applied a three-dimensional (3D) lens–kinematic modelling technique13 to the interferometric data cube containing the [C ii] line to determine the gas surface brightness distribution in each spectral channel (Figs. 1, 2, Extended Data Figs. 1, 2, Extended Data Table 1; see Methods for full discussion), from which we derived the gas kinematics, and performed a robust dynamical analysis of the galaxy. In addition, we reconstructed the far-infrared surface brightness distribution of the heated dust emission from the interferometric continuum dataset. We find that the rotation curve of SPT0418–47 has the typical shape of a bulge-dominated spiral galaxy in the local Universe11; it has a bump at 0.2 kpc from the galaxy centre and then declines before flattening at radii greater than 1.5 kpc (Fig. 2a, c). The [C ii] velocity dispersion σ is well described by an exponential profile, with average velocities of ~45 km s−1 in the inner regions (≱1 kpc) and ~18 km s−1 in the outer disk (≳1 kpc) (Fig. 2b, d, Extended Data Table 1). From a decomposition of the rotation curve, we derive the relative contribution of the different mass components—the stellar component, the gaseous disk and the dark-matter halo—to the total galactic gravitational potential (Extended Data Table 5, Fig. 2c; see Methods).

SPT0418–47 has global physical properties (total stellar mass 10 +0.3 12 Mstar = 1.2+0.2 −0.1 × 10 M☉ , dark-matter mass MDM = 1.7−0.3 × 10 M☉ , gas / +0.06 fraction fgas = 0.53−0.08 and stellar effective radius Re = 0.22+0.04 −0.02 kpc; uncertainties are the 16th and 84th percentile ranges of the probability distribution function of each parameter; M☉ is the mass of the Sun; see Extended Data Table 5 for further parameters) that are in agreement with the predictions from the most recent theoretical models9, as well as observations of the population of starburst galaxies at this redshift10. From the delensed dust emission, we derive an intrinsic infrared luminosity of (2.4 ± 0.4) × 1012L☉ (L☉, luminosity of the Sun) a star formation rate (SFR) of (352 ± 65)M☉ yr−1 and a gas-depletion timescale of 38 ± 9 Myr, which are all typical of millimetre-selected starburst galaxies at this redshift10 (see Methods). Our high-resolution 3D kinematic analysis shows that SPT0418– 47 has a ratio of rotational velocity (V) to velocity dispersion σ of V/σ = 9.7 ± 0.4 (Extended Data Table 2), which is similar to that of spiral galaxies in the local Universe11, but considerably larger than what is predicted by most numerical1,3 and analytical2,4,5 models. For example, for the majority of star-forming galaxies at z ≈ 4 with stellar masses in the range 109M☉ –1010.5M☉, the most recent cosmological magnetohydrodynamical simulation TNG501 gives V/σ ≱3 (light-blue band in Fig. 3); even though such galaxies have rotationally supported gas-rich disks, they are dynamically hotter than their low-redshift counterparts. Complex astrophysical processes—such as stellar feedback by supernovae or radiation pressure, or active galactic nucleus (AGN) feedback from the central supermassive black-hole, galaxy mergers, and gas inflows and outflows—are expected to have a substantial effect on the

Max-Planck Institute for Astrophysics, Garching, Germany. 2Kapteyn Astronomical Institute, University of Groningen, Groningen, The Netherlands. 3ASTRON, Netherlands Institute for Radio Astronomy, Dwingeloo, The Netherlands. ✉e-mail: [email protected]

1

Nature | Vol 584 | 13 August 2020 | 201

Article

7.5 5.0 1 kpc

2.5

0

1 kpc

−200

100

0

0

−100

−100

−200

−200

−300

−300

Measured Stars

4

−4 60

Gas Total Dark matter

0 Offset (kpc)

2

4

d

V Q

50

8 7 6

V (km

s–1)

300 Vc (km s–1)

−2

200

40

5

30

4 3

20

2

100

10

0

1

0 0

1

2 R (kpc)

3

202 | Nature | Vol 584 | 13 August 2020

0 0

1

2 R (kpc)

3

ΔVLOS (km

200 s–1)

300

Toomre parameter, Q

ΔVLOS (km s–1)

b

100

c

Velocity dispersion (km s–1)

s

s

s

39 .6

39 .5

40 30

1 kpc

20

the black solid contours are at Vsys ± ΔV, where Vsys is the systemic velocity Vsys = 0 km s−1 and ΔV = 40 km s−1. We note that the scales of the second-moment maps (c and f) are different because c shows the observed values, whereas f shows the intrinsic ones (beam-smearing corrected). These six maps are intended only for visualization; the source reconstruction and its kinematic modelling are performed using the full 3D information of the data cube containing the [C ii] emission line (seeMethods and Extended Data Fig. 1 for further details). dec., declination; RA, right ascension.

200

400

50

10

a

2

60

−100

Fig. 1 | [C ii] emission from the lensed galaxy SPT0418-47 and source plane reconstruction. a, Emission of the 158-μm fine-structure line of ionized carbon [C ii] integrated across a velocity range of 721 km s−1 (zeroth-moment map). The beam size, shown as a white ellipse on the lower left corner, is 0.19″ × 0.17″ at a position angle of −85.22°. b, c, As in a, but the emission is colour-coded by the flux-weighted velocity (b) and velocity dispersion (c) (firstand second-moment maps). d–f, Zeroth-, first- and second moment-maps of the reconstructed source. In d the white contours are set at n = 0.05, 0.1, 0.2, 0.4, 0.6, 0.8 times the value of the maximum flux of the zeroth-moment map. In e

0 Offset (kpc)

39 .7

s m h

100

0.0

−2

f

200

Velocity dispersion (km s–1)

10.0

e

Velocity (km s–1)

12.5

RA

4

4

15.0

−4

Velocity (km s–1)

h

18

m

RA

Velocity-integrated surface brightness (mJy kpc−2 km s−1)

17.5

300

25

39 .8

s 39 .8

s

4

20.0

d

50

−200

39 .5

s 39 .6

s

RA

h

18

m

39 .7

39 .8

s

0.0

75

18

0.5

s

54″

−100

100

s

1.0

0

125

39 .5

53″

175 150

100

39 .7

dec.

1.5

c

200

s

2.0

52″

b

39 .6

2.5

Velocity-integrated surface brightness (Jy km s–1 per beam)

a

–47° 51′ 51″

Fig. 2 | Kinematic and dynamical properties of SPT0418-47. a, b, [C ii] emission in the position– velocity diagrams along the major (a) and minor (b) axis. These diagrams show slices—the equivalent of putting long slits along the two axes. The x axis shows the offset from the galaxy centre along the major and minor axis, and the y axis represents the line-of-sight velocity (ΔVLOS) centred at the systemic velocity of the galaxy. The dark-blue contours show the reconstructed source, and the red contours show the best kinematic model (see Methods). The contour levels are set at n = 0.1, 0.2, 0.4, 0.8 times the value of the maximum flux in the major-axis position–velocity diagram. The grey circles show the rotation velocities derived using our 3D lens– kinematic methodology. c, Rotation curve decomposition. The green solid line shows the circular velocity (Vc) profile. The black dotted line is the best dynamical model, obtained by fitting the different mass components contributing to the total gravitational potential, as shown in the key. d, Velocity dispersion profile (solid green line) and Toomre parameter profile (dotted blue line). The coloured bands in c and d represent uncertainties obtained by error propagation from the 1 s.d. uncertainties of the parameters that define the respective profiles (see Methods and Extended Data Table 1).

13.0 10.0 8.0

6.0 5.0

4.0

Age of the Universe (Gyr) 3.0 2.0

1.5

1.2

14 12

V/V

10 8 6 4 2 0

0

1 Lelli et al.11 Swinbank et al.30 Harrison et al.29 Di Teodoro et al.25 Wisnioski et al.26

2

Redshift

3

Swinbank et al.30 Wisnioski et al.26 Lelli et al.27 Turner et al.28 Illustris TNG501

4

5

Violent disk instabilities2,3 Analytic4 Analytic5 SPT0418, Vmax/Vm SPT0418, Vflat/Vext

Fig. 3 | Comparison between SPT0418-47 and samples of observed and simulated galaxies. Mean ratios of the rotational to random motion (V/σ) of the gas versus redshift for the comparison samples11,25–30 of observed star-forming galaxies indicated in the key and for SPT0418-47 (red square and empty red circle). The gas tracers are: H i (ref. 11), Hα (refs. 25,26,29,30), [O ii] (ref. 30), [O iii] (ref. 28) and [C i] (ref. 27). For SPT0418, V/σ was obtained from the [C ii] emission line. For the comparison samples, the shaded regions show the area between the 16th and 84th percentiles of the distributions, and the horizontal bars show the median values (where available). In Extended Data Table 3 we show the different extraction methods used to compute V/σ. For the empty markers, the V/σ values were calculated using for each galaxy a proxy for the maximum rotation velocity, Vmax. The V/σ values shown by the full markers were calculated using the flat or the outer part of the rotation curve, Vflat. The violet cross is a lower limit for a single galaxy27. The light-blue band shows V/σ for simulated galaxies from Illustris TNG501 in the mass range 109M ☉–1011 M ☉. For these simulated galaxies, V/σ is calculated as the ratio between the maximum rotation velocity, Vmax, and the mean velocity dispersion, σm. The grey area shows the expected V/σ for galaxies dominated by violent disk instabilities2,3. The black and green areas show a prediction and an assumption, respectively, from two different analytical models (Analytic 14 and Analytic 2 5). The red square is the Vmax/σm for SPT0418-47, and the empty red circle shows the position of the V/σ value for SPT0418-47, calculated as the ratio between Vflat and the velocity dispersion at outer radii, σext (see Extended Data Table 2). The error bars for the red square and empty red circle show the 1 s.d. uncertainties.

gas kinematics within galaxies at this early epoch, and are predicted to be responsible for a progressive increase of chaotic, random motions with redshift1,4,5. Our finding firmly rules out models in which a high star-formation feedback and a high gas fraction necessarily produce large turbulent motions1,4 (light-blue, green and black bands in Fig. 3) and violent disk instabilities2,3, resulting in dispersion dominated systems with V/σ ≱2 at these redshifts (grey band in Fig. 3; see Methods for further discussion about these comparisons). Our result requires galaxy evolution models to produce dynamically cold galaxies that are not characterized by large turbulent motions1,4 and violent instabilities2,3, even at early times. Our kinematic analysis also allows the level of axisymmetric disk instabilities to be measured within SPT0418–47 via the Toomre parameter14 Q: a value of Q > 1 ensures that no instabilities will develop because large-scale collapse is prevented by differential rotation, whereas Q  ≱1 indicates that instabilities will be able to grow and lead to the creation of gas and star-formation clumps within the disk. For SPT0418–47, we find an average value of Q = 0.97 ± 0.06 at a distance of R > 1 kpc

(Fig. 2d), where the gas component is dominant (Fig. 2c), indicating a potentially unstable disk, prone to form clumpy star-forming regions. This result supports the hypothesis15 that the irregular morphologies found for dusty starburst galaxies in the optical/ultraviolet rest-frame wavelengths16 are poorly resolved clumpy star-forming regions, and not objects that are undergoing or have recently experienced a merging event. Using our rotation-curve decomposition (see Methods), we find that the stellar component of SPT0418–47 is well described by a Sérsic pro+0.2 10 file with a Sérsic index of 2.2+0.3 −0.2 and a stellar mass of 1.2−0.1 × 10 M☉ . Several observational studies of the structural properties17 of galaxies have confirmed that the Hubble sequence is already in place at z ≈ 2.5, with galaxies showing a large variety of morphologies. However, at z ≳ 3, the lack of spatially resolved data in the rest-frame optical emission for these galaxies has prevented the study of their structures and morphologies. The unprecedented spatial resolution of 60 pc of the dataset for SPT0418–47 allows one to study the morphological properties of a z ≈ 4 galaxy. The bump in the inner region of the rotation curve clearly indicates that a bulge is already in place at z ≈ 4, whereas the Sérsic index of ~2 is a signature of either a disky bulge18 or an embedded disk-like component. Dusty starburst galaxies are believed to be the progenitors of early-type galaxies (ETGs), which are the most massive galaxies observed today, dominated by old stellar populations. The most popular evolutionary track for this transformation2,3,19 predicts that the dusty starburst phase is followed by a quenching phase, during which AGN feedback leads to gas consumption and heating, with the consequent formation of a population of compact quiescent galaxies20 at z ≈ 2. In the final phase, minor dry mergers are expected to be responsible for a growth in galaxy size and the transformation into present-day ETGs. In Fig. 4a–d we compare the main physical quantities of SPT0418– 47, as inferred from our dynamical model, with the corresponding quantities for the sample of local ETGs from the ATLAS3D survey21. We consider only those local ETGs with estimated stellar ages of ≳12 Gyr (ref. 22), which is the look-back time corresponding to z ≈ 4. The orange thin diamond in each panel shows the values derived for SPT0418–47 as observed today, and the red diamond shows the corresponding baryonic values (gas plus stars), under the assumption that all the gas that we observe today will be converted into stars, preserving the disk configuration. Given the SFR estimated for SPT0418–47, this conversion will happen in ~38 Myr (see Methods). The comparison between the ETGs and the stellar/baryonic quantities for SPT0418–47 in the size–mass plane (Fig. 4a) indicates that this starburst galaxy should increase its stellar mass by a factor of 6 (3 for the red diamond) and its effective radius by a factor of 11 (3 for the red diamond), in order to evolve into an average ETG (yellow cross). This is in agreement with a simple toy model23 for mergers, in which a single major dry merger event would be responsible for an increase in both the size and the stellar mass of SPT0418–47 by a factor of 3. In Fig. 4a, the grey stars show the populations of compact galaxies20 at 2 2. Mon. Not. R. Astron. Soc. 457, 4406–4420 (2016). 64. Kennicutt, R. C. & Evans, N. J. Star formation in the Milky Way and nearby galaxies. Annu. Rev. Astron. Astrophys. 50, 531–608 (2012). 65. Tamburro, D. et al. What is driving the H i velocity dispersion? Astrophys. J. 137, 4424–4435 (2009). 66. Utomo, D., Blitz, L. & Falgarone, E. The origin of interstellar turbulence in M33. Astrophys. J. 871, 17 (2019). 67. Rafelski, M. et al. The star formation rate efficiency of neutral atomic-dominated hydrogen gas in the outskirts of star-forming galaxies from z~1 to z~3. Astrophys. J. 825, 87 (2016). 68. Leroy, A. K. et al. The multi-phase cold fountain in M82 revealed by a wide, sensitive map of the molecular interstellar medium. Astrophys. J. 814, 83 (2015). 69. Lelli, F., Verheijen, M. & Fraternali, F. Dynamics of starbursting dwarf galaxies. III. A H I study of 18 nearby objects. Astron. Astrophys. 566, A71 (2014). 70. Planck Collaboration. Planck 2015 results XIII. Cosmological parameters. Astron. Astrophys. 594, A13 (2016).

Acknowledgements This study used ALMA data 2016.1.01499.S. ALMA is a partnership of ESO (representing its member states), NSF (USA) and NINS (Japan) with NRC (Canada), NSC and ASIAA (Taiwan) and KASI (Republic of Korea), in cooperation with the Republic of Chile. The Joint ALMA Observatory is operated by ESO, AUI/NRAO and NAOJ. S.V. received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (grant agreement number 758853). F.R. thanks T. Naab and T. Costa for useful comments and discussions. Author contributions F.R., F.F. and S.V. analysed the data. D.P., F.R. and S.V. developed the software used for the lens–kinematic modelling. F.R. developed the software for the dynamical analysis. H.R.S. and F.R. reduced the data. F.R., J.P.M., S.V. and H.R.S. contributed to the writing of the manuscript. F.F. and S.D.M.W. helped with the interpretation of the scientific results. All authors discussed the results and provided comments on the paper. Competing interests The authors declare no competing interests. Additional information Correspondence and requests for materials should be addressed to F.R. Reprints and permissions information is available at http://www.nature.com/reprints.

Extended Data Fig. 1 | Reconstruction of the [C ii] emission and kinematic model. The rows show representative channel maps at the velocity shown on the upper left corner of column 4. Columns 1 and 2 show the ‘dirty’ image of the data and the model, respectively, colour-coded by the flux in units of mJy per beam. Column 3 shows the residuals (data minus model) normalized to the

noise. Columns 5 and 6 show the contours of the reconstructed source and of the kinematic model used to constrain the source reconstruction. The contour levels in the last columns are set at n = 0.1, 0.2, 0.4, 0.6, 0.8 times the value of the maximum flux of the kinematic model.

Article

Extended Data Fig. 2 | Corner plot showing the posterior distributions of the lens and kinematic parameters. The dark and light areas in the two-dimensional distributions show the 39% and 86% confidence levels, corresponding to 1 s.d. and 2 s.d., respectively, obtained with the fiducial

methodology described in Rizzo et al.13 (green) and with the direct forward modelling method (grey). From left to right, the first six panels show the lens parameters, and the other panels show the source kinematic parameters (see also Extended Data Table 1).

Extended Data Fig. 3 | Corner plot showing the posterior distributions of the dynamical parameters. The posterior distributions are the results of the decomposition of the circular velocity (Fig. 2c) into the physical components contributing to the total gravitational potential: the stars, the gas disks and the dark-matter halo. The fitted parameters are the stellar mass Mstar, the effective

radius, Re, the Sérsic index, n (Sérsic profile), the mass of an NFW dark-matter halo, MDM and the conversion factor between the [C ii] luminosity and the gas mass, α[Cii]. The dashed lines in the one-dimensional histograms show the 16th, 50th and 84th percentiles (see Extended Data Table 5).

Article Extended Data Table 1 | Lens and source kinematic parameters

The lens parameters describe a projected mass density profile with a cored elliptical power-law distribution (equation (1)) plus the contribution of an external shear component (Γsh, θsh). The kinematic parameters describe a rotating disk with a rotation curve defined by a multi-parameter function (equation (2)), a velocity dispersion profile defined by an exponential function (σ0, R0) and a geometry defined by the inclination (i) and the position angles (PA).

Extended Data Table 2 | Kinematic properties for SPT0418-47 derived under different assumptions

Parameters representing the rotation velocities and velocity dispersion profile, as well as the rotation support for this galaxy, calculated under different definitions. a Maximum rotation velocity. b Mean velocity dispersion. c Rotation-to-random-motion ratio, calculated from Vmax and σm. d Flat rotation velocity, calculated using the flat part of the rotation curve (R > 2 kpc; see Fig. 2c). e Velocity dispersion at outer radii (R > 1 kpc). f Rotation-to-random-motion ratio, calculated from Vflat and σext.

Article Extended Data Table 3 | Kinematic measurements for the comparison samples

See refs. 11,25–30. a We assume a value of σ = 10 km s−1, typical of H i in local spiral galaxies11. b VPV,max and VPV,min are the maximum and minimum velocities, respectively, along the position–velocity diagram.

Extended Data Table 4 | Assumptions for the dynamical fit

Article Extended Data Table 5 | Physical quantities for SPT0418-47 derived from the kinematic and dynamical modelling

Left: parameters inferred from a dynamical fit to the circular velocity. The stellar component is described by a Sérsic profile, the gas disk by an exponential profile and the dark matter by an NFW profile. Right: the quantities in square brackets are calculated considering the gas component, under the assumption that all the gas that we observe today will be converted into stars, preserving the disk configuration. a Total gas mass, computed as Mgas = α[cii]L[cii]. b Total baryonic mass, computed as Mbar = Mstar + Mgas. c Baryonic half-mass radius. d Dark-matter fraction within the half-mass radius. e Stellar-to-halo-mass ratio, Mstar/MDM = (Mstar + Mgas)/MDM. f Gas fraction, Mgas/(Mstar + Mgas). g Virial velocity of the dark-matter halo. h Virial radius of the dark-matter halo. i Maximum velocity for an NFW halo, computed as V200 0.216 c /[ln(1 + c) − c /(1 + c)] .

Article

Stabilization and operation of a Kerr-cat qubit https://doi.org/10.1038/s41586-020-2587-z Received: 28 July 2019

A. Grimm1,4,6 ✉, N. E. Frattini1,6, S. Puri2, S. O. Mundhada1, S. Touzard1, M. Mirrahimi3, S. M. Girvin2, S. Shankar1,5 & M. H. Devoret1 ✉

Accepted: 20 May 2020 Published online: 12 August 2020 Check for updates

Quantum superpositions of macroscopically distinct classical states—so-called Schrödinger cat states—are a resource for quantum metrology, quantum communication and quantum computation. In particular, the superpositions of two opposite-phase coherent states in an oscillator encode a qubit protected against phase-flip errors1,2. However, several challenges have to be overcome for this concept to become a practical way to encode and manipulate error-protected quantum information. The protection must be maintained by stabilizing these highly excited states and, at the same time, the system has to be compatible with fast gates on the encoded qubit and a quantum non-demolition readout of the encoded information. Here we experimentally demonstrate a method for the generation and stabilization of Schrödinger cat states based on the interplay between Kerr nonlinearity and single-mode squeezing1,3 in a superconducting microwave resonator4. We show an increase in the transverse relaxation time of the stabilized, error-protected qubit of more than one order of magnitude compared with the single-photon Fock-state encoding. We perform all single-qubit gate operations on timescales more than sixty times faster than the shortest coherence time and demonstrate single-shot readout of the protected qubit under stabilization. Our results showcase the combination of fast quantum control and robustness against errors, which is intrinsic to stabilized macroscopic states, as well as the potential of of these states as resources in quantum information processing5–8.

A quantum system that can be manipulated and measured tends to interact with uncontrolled degrees of freedom in its environment, leading to decoherence. This presents a challenge to the experimental investigation of quantum effects and in particular to the field of quantum computing, where quantum bits (qubits) must remain coherent while operations are performed. Most noisy environments are only locally correlated and thus cannot decohere quantum information encoded in a non-local manner. Therefore, quantum information can be protected through the use of spatial distance9–11 or entangled qubit states12,13. Crucially, this concept can be extended to non-local states in the phase space of a single oscillator2,14, with the additional benefit of involving fewer physical components, a property termed hardware efficiency. The latter is desirable because fully protecting a quantum system against all forms of decoherence is likely to involve several layers of encodings, and it is crucial to introduce efficient error protection into the physical layer while maintaining simplicity6,7,15. A natural choice for non-locally encoding a qubit into the phase space of an oscillator is superpositions of macroscopically distinct coherent states—the so-called Schrödinger cat states. Here we choose the states |C ±α⟩ = (| + α⟩± | − α⟩)/ 2 with average photon number n¯ = |α|2 and, respectively, even and odd photon number parity as the Z eigenstates of the encoded qubit (Fig. 1a). The coherent states | + α⟩ and | − α⟩ are

the approximate X eigenstates (Supplementary Information section I), and their distance in phase space ensures protection against any noise process that causes local displacements in this space. Crucially, this leads to a suppression of phase flips that is exponential in the average photon number n (refs. 1,2). In particular, photon loss, the usual noise process in an oscillator, cannot induce transitions between | + α⟩ and | − α⟩ because they are eigenstates of the annihilation operator aˆ. This is not the case for their superpositions, so a stochastic photon-loss event corresponds to a bit-flip error on the encoded qubit: a^|C ±α⟩ = α|C α∓⟩, which also affects the parity-less Y eigenstates |C ∓α i⟩ = (| + α⟩ ∓ i| − α⟩)/ 2 , where i = −1 . However, for a given single-photon-loss rate κa, the bitflip rate of approximately 2nκ ¯ a (ref. 16, Supplementary Information  section XI) increases only linearly with the photon number. A qubit with such ‘biased noise’ is an important resource in fault-tolerant quantum computation17,18. Additional layers of error correction can then focus strongly on the remaining bit-flip error6,7,18. This substantially reduces the hardware complexity compared with conventional approaches that use qubits without such error protection. Here we show that such a protected qubit can be stabilized autonomously in a simple and versatile implementation similar to a superconducting transmon under parametric driving. Unlike in other hardware-efficient encodings19–21, the most nonlinear mode of our

1 Department of Applied Physics, Yale University, New Haven, CT, USA. 2Department of Physics, Yale University, New Haven, CT, USA. 3QUANTIC team, Inria Paris, Paris, France. 4Present address: Photon Science Division, Paul Scherrer Institut, Villigen, Switzerland. 5Present address: Electrical and Computer Engineering, University of Texas, Austin, TX, USA. 6These authors contributed equally: A. Grimm, N. E. Frattini. ✉e-mail: [email protected]; [email protected]

Nature | Vol 584 | 13 August 2020 | 205

Article |D+〉

a

b

+Z

|D+i〉

|0〉 – i|1〉 2

|–D〉

|+D〉 +X

|D–〉

c

+Y

|0〉 + i|1〉 2

Wramp

0

1 2K

E/ƫK −50 −25 0 1

0

0

2

1

–1

Time

–2

0

Re(a)

−2

2

e

2Za

Za

2 cm

Im(a)

Im(a)

|0〉 – |1〉 2

|n = 1〉

E/ƫK −50 −25

–1

d

+Z

|0〉 + |1〉 2 +X

|D–i〉

+Y

|n = 0〉

0

f

G

Re(a)

2

4 μm

Fig. 1 | Qubit encoding, stabilization and implementation. a, Bloch sphere of the protected Kerr-cat qubit (KCQ) in the large-α limit (Supplementary Information section I). The states on all six cardinal points are labelled, indicated by coloured markers, and their Wigner function16 phase-space representations are sketched. Here, |± Z ⟩ = |C ±α⟩ = (| + α ⟩±| − α ⟩)/ 2 and |± Y ⟩ = |C ∓αi⟩ = (| + α ⟩ ∓ i| − α ⟩)/ 2 . The continuous X(θ) gate with the arbitrary rotation angle θ and the discrete Z(π/2) gate are shown by a purple circle and a blue arrow, respectively. b, Bloch sphere of the single-photon Fock qubit (FQ). c, Energy (E) dependence of equation (1) on classical phase-space coordinates Re(a) and Im(a) for squeezing drive amplitudes ϵ2/2π = 17.75 MHz (left) and ϵ2 = 0 (right) with a sketch showing the adiabatic ramp of the drive over a time τramp ≫ 1/2K. Black lines are constant energy contours. The quadrature expectation values of the |± X ⟩ states from a, b are indicated by their respective markers. d, Photograph of the nonlinear resonator (purple frame) inside the copper section of the readout cavity (Methods). Also represented are the X(θ)-rotation drive (ωa) and the squeezing-generation drive (2ωa). e, Schematic of the nonlinear resonator with pad offset δ to set the dispersive coupling to the readout cavity (Methods) and spiral symbol representing the nonlinear inductor (SNAIL element). f, Scanning electron micrograph of the SNAIL element consisting of four Josephson junctions in a loop threaded by an external magnetic flux Φ.

system encodes and stabilizes the qubit without requiring auxiliary nonlinear modes that could introduce additional uncorrectable errors. For this protected qubit to be practical, it is essential that operations can be performed faster than the shortest decoherence timescale, here the bit-flip time. We experimentally demonstrate such fast gate operations and quantum non-demolition single-shot readout in a system that maintains phase-flip protection via the simultaneous stabilization of two opposite-phase coherent states. Our approach is based on the application of a resonant single-mode squeezing drive to a Kerr-nonlinear resonator4. In the frame rotating at the resonator frequency ωa, the system is described by the Hamiltonian †2 2 †2 2 H^cat /ħ = − Ka^ a^ + ϵ 2(a^ + a^ ),

(1)

where K is the Kerr nonlinearity, ϵ 2 is the amplitude of the squeezing drive and ħ is the reduced Planck’s constant. Some intuition on this system can be gained from computing ⟨Hˆcat⟩ as a function of classically treated phase-space coordinates. There are two stable extrema at ± α = ± ϵ 2 /K , as indicated by the markers in Fig. 1c. They correspond to the lowest degenerate eigenstates of the quantum Hamiltonian4 (Supplementary Information) and thus do not decay to vacuum. 206 | Nature | Vol 584 | 13 August 2020

These eigenstates are separated from the rest of the spectrum by an energy gap Egap /ħ ≈ 4Kn¯ (Supplementary Information section III, Supplementary Fig. 1), which sets the speed limit for operations and readout. The energy barrier between these eigenstates prevents jumps along the X-axis of this ‘Kerr-cat qubit’ (KCQ). If no squeezing drive is applied, Hˆcat reduces to the Hamiltonian of an anharmonic oscillator. This resembles a superconducting transmon with anharmonicity 2K, commonly used to encode a ‘Fock qubit’ (FQ) into the first two photon number states 0⟩ and 1⟩ (Fig. 1b). Its classical energy displays one single extremum in phase space harbouring the quadrature expectation values of both X eigenstates without the protection of an energy barrier (Fig. 1c, right panel). By toggling the squeezing drive, a Kerr-nonlinear resonator can be tuned to implement either type of qubit. Our experimental implementation consists of a superconducting nonlinear resonator placed inside a three-dimensional (3D) microwave cavity (Fig. 1d). This is a standard setup in 3D transmon qubits with a few key modifications (Methods). The foremost modification, employing a superconducting nonlinear asymmetric inductive element (SNAIL)22 as a nonlinear inductor (Fig. 1f), allows us to create single-mode squeezing by applying a coherent microwave drive ωs at twice the resonator frequency ωa and makes the resonator flux tunable. Here we tune our device to a frequency ωa/2π = 6 GHz and Kerr nonlinearity K/2π = 6.7 MHz. At this frequency, the FQ has an amplitude damping time T1 = 15.5 μs and a transverse relaxation time T2 = 3.4 μs. The FQ is employed for initialization and measurement of the KCQ during most experiments described in this work. This is possible because the states |0⟩, |C +α⟩ (|1⟩, |C −α⟩) spanning the two Bloch spheres have the same even (odd) photon number parity, which is conserved by the system Hamiltonian (equation (1)). Consequently, ramping the squeezing drive on and off slowly with respect to 1/2K, as sketched in Fig. 1c, adiabatically maps between the FQ and KCQ (Methods). We now show that we indeed implement the Hamiltonian (1), and thus initialize and stabilize a KCQ, by demonstrating the unique features of Rabi oscillations around the X axis of its Bloch sphere. To this end, we apply an additional coherent drive ϵ xaˆ† + ϵ *xaˆ with amplitude ϵ x and frequency ωa = ωs/2 to the system. This lifts the degeneracy between the states | + α⟩ and | − α⟩, and therefore leads to oscillations with a Rabi frequency

Ωx = Re(4ϵ xα)

(2)

between their superposition states along the purple circle in Fig. 1a. This picture is valid for large enough α and for ϵ x ≪ Egap /ħ (ref. 4; see also Supplementary Information section IV). Note that equation (2) is different from the Rabi frequency of a FQ in two ways. First, it depends on the amplitude of the squeezing drive through α ∝ ϵ 2 . Second, it varies with the phase of the applied Rabi drive arg(ϵx) (for simplicity we chose ϵ2, α ∈ ℝ). We first focus on the effect of the squeezing drive on the Rabi frequency. We initialize in |C +α⟩ and apply a Rabi drive with constant |ϵx| and arg(ϵx) = 0 for a variable time Δt and a variable amplitude ϵ2 (Fig. 2a). Figure 2b shows the Rabi frequencies for each ϵ2, extracted from the oscillations in the measured FQ 0⟩-state population fraction at the end of the experiment. For ϵ2 = 0, we are simply driving FQ Rabi oscillations giving a direct calibration of |ϵx|/2π = 740 kHz. For large values of ϵ2, the Rabi frequency becomes a linear function of ϵ 2 , confirming the theoretical prediction of equation (2). The solid black line shows a one-parameter fit to a numerical simulation of our experiment (Methods). We now turn to another unique feature of these Rabi oscillations by setting ϵ2/2π = 15.5 MHz and varying Δt and arg(ϵx). As expected, the measured oscillations shown in Fig. 2c are π-periodic in arg(ϵx). Three cuts through this data (dashed lines) are shown in Fig. 2d. The top panel

a

c

Δt

0

Squeezing drive

3 MHz

4

1 0 –1 1 0 –1 1 0 –1 1 0 –1

0 0 1

–1.0

0

0.2

0.4 Time, Δt (μs)

0.6

Fig. 2 | Rabi oscillations of the protected KCQ. a, Pulse sequence to perform the following functions: (1) initialize the KCQ (|0⟩ → |C +α⟩), (2) drive Rabi oscillations for a varying time Δt, and (3) map onto the FQ and perform dispersive readout. ωs, ωs/2 and ωb are the frequencies of the respective drives. A black arrow indicates the endpoint of the numerical simulations performed for b, d, e (Methods). b, Dependence of the Rabi frequency Ω x on ϵ 2 . Experimental data are open grey circles. The solid black line is a one-parameter fit to the data used to calibrate ϵ2 (Methods). The red star indicates the condition ϵ2/2π = 15.5 MHz used for c, d, e. c, Dependence of the experimentally measured Rabi oscillations on time Δt and on the phase of the Rabi drive arg(ϵ x).

corresponds to a phase difference of π/2 between the coherent state amplitude and the Rabi drive, meaning that oscillations are suppressed. The middle and bottom panels at respective phase differences of π/4 and 0 display increasing Rabi frequencies, with the latter corresponding to the red star marker in Fig. 2b. In the bottom panel, the black line is the result of a numerical simulation scaled to match the contrast of the data. The black lines in the top and middle panels use the same scaling factor and are thus parameter-free predictions in good agreement with the measured data. Having benchmarked our simulation in this way, we use it to compute the full density matrix of the resonator, which we represent with Wigner functions (Fig. 2e). Apart from slight asymmetries due to the finite ramp time of the initial mapping pulse, they agree well with the expected | + Z ⟩, | + Y ⟩, | − Z ⟩ and | − Y ⟩ states of the KCQ. Next, we characterize the mapping operation and a set of single-qubit gates on the KCQ by performing process tomography (Methods). In all subsequent experiments, the average photon number of the cat states is set to n¯ ≈ 2.6 and frequency shifts induced by the squeezing ∼ ∼ drive are taken into account by setting ωs /2 = ωa , where ωa is the Stark-shifted resonator frequency (Supplementary Information section II). The pulse sequence for tomography of the mapping between the FQ and the KCQ is shown in Fig. 3a and the measured state vectors are plotted on a Bloch sphere in Fig. 3b. An estimate of the fidelity ℱmap ≈ 0.855± 0.002 (± one standard deviation here and for all subsequent values) is obtained by using the Pauli transfer matrix approach23 (Supplementary Information). This number reflects the fidelity of the tomography FQ pulses as well as of the mapping itself, because, apart from a normalization by the FQ Rabi contrast, the presented fidelities include state-preparation-and-measurement (SPAM) errors (Supplementary Information section V). We expect these errors to be dominated by decoherence during the comparatively slow adiabatic ramps. This could be remedied in future experiments by using optimal pulse shapes, which can reduce the duration of the mapping operation by a factor of more than 40 with respect to its present value4. We now turn to the pulse sequence shown in Fig. 3c, which additionally performs an X(π/2) gate on the KCQ. The process tomography data (Fig. 3d) shows the desired rotation around the X-axis with a fidelity of ℱX (π/2) ≈ 0.857± 0.001 . Comparing this value to ℱmap indicates that ℱX (π/2) is mostly limited by SPAM errors. From a complementary

0.8

0

0.2

0.4

0.6

0 0.8

−2.5

0 Re(a)

Time, Δt (μs)

Im(a)

ɽ2/2π

2

2/π W(a)

Im(a)

1

0

0 1

0.5

–0.5

0

e –2/π

1

Im(a)

Drive phase, arg(ɽx)/π

d

Im(a)

Ωx/2π (MHz)

1.0 P0

P0

5 4 3 2 1 0

0.8

P0

b

Zb

0.6

P0

Zs 2

Readout

0.4

1.0

Zs Rabi drive

0.2

2.5

The colour scale gives the ground state population of the FQ (P0) at the end of the experiment. d, Cuts of c for the three Rabi-drive phases indicated by dashed lines. Open grey circles are the experimental data and black lines are the simulation. Symbols in the bottom panel indicate the times for which the simulated oscillator state is shown in e. e, Simulated phase-space representation of the oscillator density matrix corresponding to the | + Z ⟩, | + Y ⟩, | − Z ⟩ and | − Y ⟩ states of the KCQ (from top to bottom). The colour scale gives the value of the Wigner function W(a) as a function of the real and imaginary part of the phase-space coordinate a.

b

a Squeezing drive Tomography pulses Readout

Za

±X, ±Y, ±Z 〈X〉, 〈Y〉, 〈Z〉

Zb

c

+Z

Za

±X, ±Y, ±Z

+X X

Zb

e

f

Squeezing drive

Readout

+Y

Zs 2

TX(π/2)

〈X〉, 〈Y〉, 〈Z〉

Tomography pulses

X(π/2)

Zs

KCQ Rabi pulse

Readout

+X

d

Squeezing drive

Tomography pulses

+Z

Zs

Z(π/2)

+Y +Z

Zs

TZ(π/2)

Za

±X, ±Y, ±Z 〈X〉, 〈Y〉, 〈Z〉

Zb

+X X

+Y

Fig. 3 | KCQ gate process tomography. a, c, e, Pulse sequences for process tomography of mapping between FQ and KCQ, which is ideally the identity operation 𝟙 (a), mapping and the X(π/2) gate (c), as well as mapping and the Z(π/2) gate (e). In each sequence, the FQ is initialized on the ±X, ±Y and ±Z cardinal points of the Bloch sphere, the respective operation is performed, and the expectation values ⟨X ⟩, ⟨Y ⟩ and ⟨Z ⟩ are measured by a combination of FQ pulses and dispersive readout (grey box). In c, TX(π/2) = 24 ns is the total duration of the Gaussian Rabi pulse applied to the KCQ. In e, TZ(π/2) = 38 ns is the duration for which the squeezing drive is switched off to perform the gate. b, d, f, Process tomography results presented on a Bloch sphere for the operations 𝟙 (b), X(π/2) (d) and Z(π/2) (f). The expectation-value vector after the operation for an initialization on the (+X, −X, +Y, −Y, +Z, −Z) cardinal point is represented by an (orange square, green diamond, gold right-facing triangle, purple left-facing triangle, red upward-facing triangle, blue downward-facing triangle). Error bars are smaller than the size of the markers. The fidelities are 0.855 ± 0.002, 0.857 ± 0.001 and 0.811 ± 0.001, respectively.

Nature | Vol 584 | 13 August 2020 | 207

Article Counts

0

b

Max

Y(±π/2)

–1

CR pulse

Q/V

1

c

|–D〉

0 –1

〈X〉

Probability

0.2 0

e

–1.0 –2

0 I/V

2

4

0

200

400 600 Time, Δt (μs)

–1.0

800

X(0, π)

0

5

10 15 Time, Δt (μs)

20

X(π/2)

Zcr

1.0

|D+〉

0

–0.5

|D+i〉

Zs Zs 2

CR pulse

0.5

0

Z(π/2)

KCQ pulses

|D–i〉

–0.5

|–D〉

Zcr

Δt

Squeezing

g

0.5

0

Zs Zs 2

1.0

|+D〉

–0.5

–4

X( π/2) CR pulse

Zcr

1.0 0.5

0.4

Za

f

Z(π/2)

KCQ pulses

〈Y〉

Q/V

FQ pulses

|+D〉

Δt

Squeezing

Zs

1 0

d

Δt

Squeezing

〈Z〉

a

–1.0

|D–〉 0

5

10 Time Δt (μs)

15

20

Fig. 4 | Cat-quadrature readout (CR) and coherence times. a, Top and middle: histogram of the cavity output field when performing CR (see b) after preparation of either | + α ⟩ (top) or | − α ⟩ (middle). Bottom: corresponding probability distribution along the I quadrature. Open orange (green) circles show measured data for | + α ⟩ (| − α ⟩) and solid lines are Gaussian fits of width σ used to scale the quadrature axes, I and Q. Setting a threshold at I/σ = 0 (dashed line) implements a direct single-shot readout of the KCQ along its X-axis. b, CR pulse sequence for the measurements presented in a, c. After state initialization in |± α ⟩ (Y(±π/2) gate on the FQ and mapping), a pulse at frequency ωcr = ωb − ωs/2 is applied for a time Tcr = 3.6 μs converting the quadrature amplitude of the KCQ to a drive on the readout cavity at ωb. The wait time Δt is set to zero to obtain the results shown in a

and varied in c. c, KCQ ⟨X ⟩-component coherence. Open blue circles are data and solid black lines are single-exponential fits with decay times τ+X = 105 μs ± 1 μs and τ−X = 106 μs ± 1 μs. d, Pulse sequence for e. After initialization in |C ∓αi⟩ (mapping |0⟩ → |C +α⟩ and X(∓π/2) gate) and variable wait time Δt, a Z(π/2) gate is performed followed by CR. e, KCQ ⟨Y ⟩-component coherence. Open blue circles are data and solid black lines are single-exponential fits with decay times τ+Y = 2.51 μs ± 0.06 μs and τ−Y = 2.60 μs ± 0.05 μs. f, Pulse sequence for g. After initialization in |C ±α⟩ (mapping |0⟩ → |C +α⟩ and either X(0) or X(π) gate) and variable wait time Δt, a X(π/2) gate and a Z(π/2) gate are performed followed by CR. g, KCQ ⟨Z ⟩-component coherence. Open blue circles are data and solid black lines are single-exponential fits with decay times τ+Z = 2.60 μs ± 0.07 μs and τ−Z = 2.56 μs ± 0.07 μs.

measurement (Supplementary Fig. 3), we estimate the infidelity due to over-rotation and decoherence during the gate operation to about 0.01. As this operation is compatible with an arbitrary angle of rotation, only a π/2 rotation around the Z-axis is needed to reach any point on the KCQ Bloch sphere. Nominally, such a gate is incompatible with the stabilization as it could be used to go between the states | + α⟩ and | − α⟩. However, for ϵ2 = 0, the free evolution of the Kerr Hamiltonian for a time π/2K ≈ 37.3 ns achieves the required operation4,24,25 (Methods and Fig. 3e). The tomography data shown in Fig.  3f gives a fidelity ℱZ (π/2) ≈ 0.811± 0.001 . We attribute the reduction of fidelity with respect to ℱmap to the difference between the actual gate time TZ(π/2) = 38 ns and π/2K, and to the finite rise time of the step function in ϵ2 of about 4 ns; both of which are not limitations of our device but of our room-temperature electronics (Supplementary Information section VI). So far, we have characterized the basic properties and gate operations of the KCQ by mapping back onto the FQ and using the well-understood dispersive readout method. This readout, however, destroys the state of the KCQ. We now demonstrate an entirely new way to perform a quantum non-demolition measurement on the X component of the stabilized KCQ, which we call the ‘cat-quadrature readout’ (CR). We apply an additional drive at frequency ωcr = ωb − ωs/2, where ωb is the frequency of the readout cavity. Through the three-wave mixing capability of our system, this generates a frequency-converting interaction between the nonlinear resonator and the readout cavity. ∼ In the frame rotating at both ωs /2 = ωa and ωb, this adds the following term to equation (1):

We characterize the fidelity of this readout by first initializing the KCQ along its X-axis and then applying a CR pulse for a time Tcr = 3.6 μs as shown in Fig. 4b. Two histograms of the measured cavity field are shown in Fig. 4a for initialization in | + α⟩ and | − α⟩, respectively. Their separation is large enough to implement a single-shot readout by setting a threshold at I/σ = 0 with total fidelity ℱ = 0.74 (Methods), where I/σ is a dimensionless quantity corresponding to the I-quadrature signal divided by the the standard deviation σ of the histograms. This fidelity is a lower bound including errors in state preparation caused by the thermal population of the FQ  1⟩ state (contributing an infidelity of about 2 × 4% = 8%) and imperfections during the initial FQ pulse and mapping. Finally, we characterize the quantum-non-demolition aspect of the CR as Q = 0.85 from two successive measurements (Methods). We now use this CR to investigate the phase-flip time of the KCQ. The decay of the ⟨X ⟩ component for either initial state along this axis is measured using the pulse sequence shown in Fig. 4b. We fit the data to a single-exponential decay with characteristic times τ+X = 105 μs ± 1 μs and τ−X = 106 μs ± 1 μs. Additional measurements with dispersive readout confirm this result (Supplementary Fig. 7). Similarly, the coherence times of both the ⟨Y ⟩ and ⟨Z ⟩ components are measured using CR, but employing only operations on the KCQ after the initial |0⟩ → |C +α⟩ mapping operation (Fig. 4d, f). The resulting decay curves are displayed in Fig. 4e, g. Single-exponential fits of the data yield the decay times τ+Y = 2.51 μs ± 0.06 μs, τ−Y = 2.60 μs ± 0.05 μs, τ+Z = 2.60 μs ± 0.07 μs and τ−Z = 2.56 μs ± 0.07 μs. These values are slightly smaller than the predicted bit-flip time due to photon loss τloss = T1 /2n¯ = 2.98 μs. We expect that photon-gain processes play a role in this reduction (Supplementary Information section XI, Supplementary Fig. 8). Our results demonstrate a 30-fold increase in the phase-flip time of the protected KCQ with respect to the FQ. Crucially, we perform a full set of single-qubit gates on the protected qubit on timescales that are much shorter than its bit-flip time. Although the measured gate fidelities are limited by SPAM errors, an upper bound of the error rate due to decoherence during the gate operations is given by: TX(π/2)/τ+Y ≈ TZ(π/2)/τ+Y  2γ4γ1/γ0 spontaneously opens in the RG spectrum. The double-headed arrow indicates the size of the bandgap. d, ρ xx as a function of total carrier density n = nt + nb measured at D = 0 for device 1 with rhombohedral (black curve) and hexagonal (blue curve) crystal structures (both are nine graphene layers thick). nt (nb) is the charge density induced by top (back) gate voltage. Left inset, optical micrograph of device 1; scale bar, 2 μm. e, ρxx as a function of the displacement field D; n = 0, T = 20 mK. Inset graphs in d and e show the corresponding σxx curves on a logarithmic scale.

elementary charge.) The RG device shows much stronger modulation of ρxx(n) as compared to the reference HG device, such that ρxx of the former spans 10 Ω to 23 kΩ and exhibits a profound high-resistivity region near n = 0. At the same time, the HG device displays metallic behaviour over the entire range of carrier densities (ρxx varies only from 8 Ω to 63 Ω), similarly to the reported dual-gated devices of tetralayer and hexalayer Bernal-stacked graphene20,21. Figure 1e shows three distinct high-resistance regions for RG device 1 (which are absent in the HG device) as a function of displacement field C V −C V D = t t2ε b b measured at n = 0 (ε0 is the vacuum permittivity). The 0 appearance of the high-resistivity region at the neutrality point (n = 0, D = 0) suggests that an energy gap opens in the surface states spectrum, whereas the ρxx value close to the resistance quantum h/e2 hints at edge-state transport caused possibly by topological character of the insulating state. The resistivity peak at the neutrality point was observed only in sufficiently thin ( Dc (see Fig. 2a–d), are in agreement with the theoretical prediction that a high enough D opens an energy gap in the surface state spectrum of RG22. The effect is similar to that observed in bilayer graphene under displacement but has much stronger nonlinear screening of charges at the two RG surfaces (Methods). Note that the observation of these displacement-induced gaps requires a gapped bulk of multilayer RG

–4 0 4 n (×1012 cm–2) 40

0.4 mS

100 mS 0.1

–4 0 4 n (×1012 cm–2) 3.0 nm

20 0 0.0

Vxx (S)

0.0

12 mS 0.012

–4 0 4 n (×1012 cm–2)

10–3

Dc (V nm–1)

–0.6 –0.3

0.15 mS

0.1

=0

8

100 mS

D

–4 0 4 n (×1012 cm–2)

HG

1 mS

0.025

f

e

RG

25 mS

–4 0 4 n (×1012 cm–2)

–0.6 0.0 0.6

8

RG

–1.0 10 μS

1 0.1

HG –8

16.5 nm

0.0

0.9

0

d 6.5 nm

–0.5

e

100

20

c 5.0 nm

0.5

J2

J1

b 3.0 nm

1.0

J4

J3

Uxx (kΩ)

a

c

Δ (meV)

a

0.4 0.8 D (V nm–1)

10–4

0.4

0.76 10

30 50 Nlayer

0.04

D = 0.57 V nm–1 0.08 0.12 1/T (K–1)

0.16

Fig. 2 | Thickness dependence of the transport characteristics of multilayer RG. a–d, Conductivity maps σxx (n, D) for RG devices 1 to 4, respectively, of different thicknesses as specified in the panels. e, Thickness dependence of the critical field Dc, above which the transport gap is opened (black diamonds); thickness is expressed as number of layers, Nlayer. Dc values were extracted from linear fits of σxx (D) at n = 0 near the insulating states in the maps shown in a–d. Error bars represent uncertainties of the fitting procedure. The red curve is the self-consistent model (Methods). f, Arrhenius plot of σxx for RG device 1 at D = 0.57 V nm−1 (red squares), 0.62 V nm−1 (blue squares), 0.66 V nm−1 (green squares) and 0.76 V nm−1 (yellow squares). The grey pentagons are the data for D = 0. Linear fits of high-T regions (solid lines) give the bandgap Δ presented in the inset. The red curve in the inset is the bandgap estimation using the self-consistent screening model (Methods). For comparison, the thermal activation energy of the high-resistance state at D = 0 V nm−1 is also plotted in the inset (blue star).

(Methods). The resistivity of our fully gapped RG film (both surface and bulk states are gapped) is of the order of 20 kΩ for the thin RG sample at 20 mK (device 1), and 2–3 kΩ for the 16.5-nm-thick sample (device 4) at 1.5 K. The effect of RG thickness on the displacement-field-induced bandgap is shown in Fig. 2a–d, which shows σxx(n, D) maps for four RG devices with thicknesses ranging from 3 nm to 16.5 nm (devices 1 to 4). Figure 2e summarizes how the critical displacement field Dc increases with sample thickness. The solid line is the self-consistent theory (see Methods). From the T dependence of σxx of 3-nm-thick RG device 1, we extracted the sizes of thermally activated gaps Δ at different D (Fig. 2f), which are plotted in Fig. 2f inset. The bandgap observed in RG at |D| > Dc is in stark contrast to the HG response to the displacement field where D causes the resistivity to decrease (Fig. 1e) due to the growing size of the Fermi surfaces of the top and bottom surface states22,23. The behaviour of σxx(D) presented in Fig. 2a–d is characteristic of the uniform ABCABC… stacking in our RG samples. We note that, if mixed ABC/ABA stacking were present in our devices, this should have prevented gap opening by the displacement field (Methods), which provides complementary evidence for complete ABC stacking in our RG devices (Extended Data Figs. 1 and 2). Next we explore electron transport in multilayer RG using magnetic field B applied perpendicular to the plane. Longitudinal (ρxx) Nature | Vol 584 | 13 August 2020 | 211

Article a

−6

Q = −18

8

12 101 100

8

10–1 10–2

4

10–3

0

–8

–4

b

0 n (×1012 cm–2)

4

c

12 20 28 36

B (T)

0.2

0 –30

0 E – EF (meV)

–0.2

12 20 28 36 Q (×nh/eB)

e ΔQSH ≠ 0

K, K′

Δ=0

16

Phase transition

4

Phase transition

8

Q = –6 Q=0

12

2

0

Gapped Free particle

Free particle

–0.0

5

0.01

–20 20 –16 16 –12 12 –8 8 –4 4 Q (×nh/eB)

8

E

14

12

B

↑ LL 8 16 ↓ 14 ↑ LL 12 ↓ 10

f ΔLAF ≠ 0 Δ = 0

30

dVxx/dQ (mS)

0.1

5 2

B (T)

8

11 1

8

Vxx (mS)

B (T)

11

d

Vxx (mS)

B (T)

12

K 17 K′ 16 15 14 13 12 11

↑ 16 ↓ 14 ↑ 12 ↓ 10

7

LL 8 LL 7

Fig. 3 | Quantum Hall effect in RG. a, Map of longitudinal conductivity σxx(n, B) for device 1; D = 0; T = 20 mK. Arrows and numbers indicate the filling factors of quantum Hall states. b, σxx (ν, B) for the hole side (same data as in a). c, Differential dσxx /dν(ν, B) map for the electron side (data from a). Red arrows mark the positions of LL crossings. d, Calculated spectra of LLs in multilayer RG, using gapped and gapless band structures at low (coloured curves) and high (black) energies, respectively. The free-particle and gapped regions are separated by phase transitions. Red and blue lines, depending on the order parameter, correspond to electronic states with opposite spins (quantum spin Hall) or of opposite valleys (layer antiferromagnetism). EF, Fermi energy. e, Effect of quantum spin Hall order parameter on the Landau spectrum. Numbers indicate the filling factors. LL 7 and LL 8 refer to the orbital indices of the zeroth LL. ↑ (red lines) and ↓ (blue lines) label spin-split LLs. ΔQSH is the QSH order parameter, and K and K′ are the opposite valleys. The dotted line demarcates the transition between gapped and free-particle regimes. f, Same as in e but for layer antiferromagnetism. Solid (dashed) coloured lines refer to K (K′) valleys.

and transverse (ρxy) resistivities of device 1 are shown in Extended Data Fig. 3a as a function of B at n = 2.3 × 1012 cm−2 and D = 0. Shubnikov–de Haas oscillations emerge at B ≈ 1 T, and well-quantized ρxy plateaux develop for B > 3 T manifesting the onset of the quantum Hall effect (QHE), such that under a strong magnetic field the electronic spectrum of 2D surface states collapses into a series of discrete Landau levels (LLs), leading to quantized ρxy. In high B (see, for example, Extended Data Fig. 3b for B = 10 T), ρxy(n) plateaux appear at even fractions of h/e2, indicating a two-fold (valley) degeneracy of the LLs. To examine the quantization, Fig. 3a shows the Landau fan diagram σxx(n, B) for device 1 at D = 0. We note strong electron–hole asymmetry in the diagram: the electron side exhibits straight LLs (blue lines follow the slopes B/n = h/eν, where ν is the filling factor) fanning out from the neutrality point, while the hole side shows a criss-cross pattern. We observed similar behaviour in RG devices of different thicknesses, which is consistent with the surface origin of the LLs (Extended Data Fig. 4). The QHE has also been reported in thin films of Bernal-stacked graphite23. In the latter case, the quantization was due to the formation of standing 212 | Nature | Vol 584 | 13 August 2020

waves in the bulk. In contrast, our multilayer RG has a gapped bulk but conducting surfaces, which should exhibit the QHE if their electronic quality is sufficiently high. In N-layer RG, a Berry phase ±Nπ around the opposite K points leads to N-degenerate zero-energy LLs for each spin and valley, which are localized on the opposite surfaces for opposite valleys24,25. A free-particle spectrum of LLs in RG calculated using the same method as in ref. 26 is plotted in Extended Data Fig. 4a. Typically for semimetals, conductionand valence-band LLs make numerous crossings at moderate B, which for multilayer RG happens on the hole side of the spectrum. In the absence of a displacement field, all LLs are valley degenerate. In addition, low-index valence-band LLs are triply degenerate, reflecting the triply degenerate maxima of the valence band (Fig. 1b). They are split at higher B where the three Fermi surfaces of the valence band merge into one. For device 1 (9 layers) in Fig. 3a, the total number of the zeroth LLs is 36 (2 spins, 2 valleys and 9 orbitals), and the lines corresponding to filling factors up to ±18 come from these zeroth LLs. Lines on the electron side, which trace ν = 8, 12, 16,… are well pronounced. These filling factors in multiples of 4 can be explained by relatively large Zeeman gaps between two-fold degenerate orbital levels of zeroth LLs. A pronounced gap at ν = −18 followed by a series of crossings below 8 T could be traced back to the crossings between the 2N-fold degenerate zeroth LL and the valence-band LLs. The robust ν = −18 state emerges when these crossings stop (Extended Data Fig. 4a). When only a single gate is used (Extended Data Fig. 5), the valley degeneracy becomes lifted, and a robust gap appears at ν = −7 (for 7-layer-thick device 7), ν = −9 (for 9-layer device 1) and at ν = −11 (for 11-layer device 6). This provides additional support for our assignment of the observed spectrum to the N-degenerate zeroth LL, which is in turn a manifestation of an Nπ Berry phase in N-layer RG. Although the free-particle theory adequately describes the experimental results at high doping (Fig. 3a, d above |n| > 2.5 × 1012 cm−2), it fails to explain the pronounced gaps observed at ν = 0 and ν = −6 (Fig. 3a). Apart from the crisscross pattern on the hole side (Fig. 3b), which agrees with the free-particle theory, there is a series of crossing points on the electron side at the symmetry broken states ν = 10, 14, 18, 22… The crossings are better seen if we replot the Landau fan as a function of ν (Fig. 3c; the crossing points are marked with red arrows). These crossings also cannot be explained using the free-particle picture. To account for the latter features, we suggest a spontaneous appearance of an interaction-induced spin and/or valley symmetry-breaking order parameter. It leads to a bandgap when the resulting band splitting exceeds the single-particle overlap between the conduction and valence bands, thus allowing the observation of quantum Hall states at ν = 0 and −6 (Fig. 3a, d). With increasing B, the gap switches from ν = 0 to ν = −6. This can be explained by large orbital magnetic moments that have opposite signs for opposite spins (or valleys) and lead to closure of the ν = 0 gap, as shown in Fig. 3d. The bandgap disappears with increasing doping, but the observed crossings on the electron side (Fig. 3c, and also Extended Data Fig. 4c, d for 3.3-nm-thick device 5) suggest that a small interaction-induced order parameter remains present up to n ≈ 2.5 × 1012 cm−2. The interplay between the orbital, valley and spin degrees of freedom leads to several possible states with spontaneously broken symmetry in multilayer RG. They include quantum valley Hall, quantum anomalous Hall, layer antiferromagnetic insulator and quantum spin Hall states3 (see Methods). The crossing points on the electron side (Fig. 3c, and Extended Data Fig. 4d for similar results in device 5) can be explained by the phase transition from spin-polarized states due to Zeeman splitting at higher electron concentration to quantum spin Hall states at electron concentration n >6!>pm6vklp>?6rlrlovp'67%"$0$"$" "2"3"*3$+"&'#6e"$"33+2"0"3?"$"*1&%$$(>>> !.6#006#>0!C*>%!7"05~*%!rlh%#*|3%$$(>>#06"3$#6!>€.#3.">Bg-yw>6!>pm6vklp>?6rlrlovp'9" $C0!3.$780"*~"30"$21h-$78"0y=#3."'6x"#"$"*"1 0#3$$*$"!>".6%".6..6#>"0%q>13"6%$.39"00j#ku9kmpl'9!"$"$&f:/tB: 2kl%$$(>>1116!0!6!>%#.">3"ykl6%$.39#*%$$(>>#!6!>0#*>"$">!062pm6!6"$6$#4$ !($"8&7746&$&( ((# p >#4$ !($"8&"/& '2!($" 44($"='4)& ((8"2&7$(/&"#&#`'(2"(82'7($!742!&$" '77#4$!($"8() (&($($4&7!&&2($"47'#$"%4"(&7("#"4/*:%:2&"+().&$4($2&( *:%:% $"488$4$"(+ p >8 >3G6&$&($"*:%:(&"#&##6$&($"+& 4$&(#($2&( 8'"4(&$"(/*:%:4"8$#"4$"(6&7+ 7)/!()$(($"%=()(((&($($4*:%:a=P=W+5$()4"8$#"4$"(6&7=884($E=#% 88#2&"#b6&7'"(# p Hq2&!!$"%:"$2&7&"#()%&"$2 l'2&"&4)!&($4$!&"( 7$"$4&7#&(&

p p

p

)sAHq m*>.4&2=2"47"&7~nmn=7(€k0xoxxu~Hv=x,onn #$7'($"8sA=x,xnnn#$7'($"8jf+=!vu*77$%"&7$"%=2"47"&7Gxt@xo=7(€xm=x,onn#$7'($"8sA=x,xnnn#$7'($"8 jf+=y$"4'7$"*1$%2&=2"47"&7yzxmx=7(€xowCtwuoy=x,onnn#$7'($"8jf+=lmo~>4*>.4&2t~oz=7(€k0ouxzuwHx= x,onn#$7'($"8sA=x,xnnn#$7'($"8jf+=>;;*@22((&7:=3&('onx~+=!&"4& *02 .%(&7:=C72(&.7$2onx~+=7$6*>2'(&7=3&(22onx~+ !vu*77$%"&7$"%Gxt@xo+5& 6&7$#&(#(595$()jf$"2'7($!7)'2&"&"#2'4777$" &"#)sA$"l7&477& #(&$7#"()2&"'8&4('g5.$(*)((!,DD555:477$%"&7:42D!#'4(D!$2&/H&"($.#$D"8H9.H!vuH#xtxoHB!H&..$(H 2&.Dwoto+: >;7("&($6C#$4$"onxz+ >;"($1C0;nvHwzx+5& 6&7$#&(#(59$")'2&"l7&4777$"8)sA& #(&$7#"()2&"'8&4('g 5.$(*)((!,DD555:2#2$77$!:42D…1D"D!#'4(D>"($H1C0;H>"($.#/=CC†3;*@22((&7:= 3&('onx~+ y$"4'7$"*>.4&2=otwmt+5& !(#(598jf$"2'7($!7)'2&"&"#2'4777$" & #(&$7#"()2&"'8&4('g 5.$(*)((!,DD555:$%2&&7#$4):42D4&(&7%D!#'4(D$%2&D6zxmx‡7&"%ˆ"?%$"ˆ…1+& 577& !6$'7/!'.7$)#8 jf$"2'($' 82'7&. '4)& f>;*@22((&7:=3&('onx~+=7$6&"#2./"$4(2477*>#7&"2$"$(&7:= A3>1onxz=G$$49B(&7:=onxz+

12345656762589 56 53 17425

>"($.#$

@'9&/($44777$"

A7$4/$"82&($"&.'(4777$" A$2&/2'$"."H2&5#$6#2&4!)&%=l@Hozm"$2&7&&"#…22$((8()…"$6$(/8 A""/76&"$&$"&44#&"45$()()%'$#7$" 8()3&($"&7s"($('( 8l&7():>77!4#' 5!82#&44#$"%( $"($('($"&7!&$"47& !(47: 3(()&(8'77$"82&($""()&!!6&78() ('#/!(472'(&7.!6$##$"()2&"'4$!(:



j$7#&"$2&7 ;4†$"!'(†-A1 lsA†q†lmo~>4†$"!'(†6) lsA†q†lG>m†$"!'(†-A1 lsA†q†lG>m†$"!'(†6) lsA†q†30x†$"!'(†6) lsA†q†30o†$"!'(†-A1 lsA†q†30o†$"!'(†6) lsA†q†!vu†$"!'(†-A1 lsA†q†!vu†$"!'(†6) )sA†q†30x†$"!'(†-A1 $"6$6†-A1x†lG>m†)sA†q $"6$6†-A1o†lG>m†)sA†q $"6$6†-A1m†lG>m†)sA†q $"6$6†6)$47x†lG>m†)sA†q

‹

C()#7%/

0!7$4&( 1q'"4$"%#!() >"($.#$

A&94&77$"%!&&2(

G&(&q'&7$(/ 18(5&

;).$7%$4&7!7$4&( 2&4!)&% *fCGC !$("&7+!7#82mHu2$4 @&4)7$.&/4"(&$" &(7&(on2$ n 77$"&# lG>m*>.4&2~nmn+=lmo~>4*>.4&2t~oz+=>;;"($1C0;nvHwzx+ A&94&77$"%!&&2(5 (&( m†)sA†q A7&#5"7&#B27 $"82.757$"9&"#7&# "s "ky )((!,DD#2:%"2!&4:%D#&(&2&"&%D8$7Dl2DA'.7$4D4&6&7/2&"~zzoD)sA†q†3%'/"†-&E&†onxz:B27

f5($o6$"o:o:v lrC@06t:xx .#(76o:ou ;f>;&"4$!($"8&4(.$"#$"%&"&7/$6x:n



Œ

Article

Lysosome-targeting chimaeras for degradation of extracellular proteins https://doi.org/10.1038/s41586-020-2545-9 Received: 1 May 2019

Steven M. Banik1, Kayvon Pedram1, Simon Wisnovsky1, Green Ahn1, Nicholas M. Riley1 & Carolyn R. Bertozzi1,2 ✉

Accepted: 22 June 2020 Published online: 29 July 2020 Check for updates

The majority of therapies that target individual proteins rely on specific activitymodulating interactions with the target protein—for example, enzyme inhibition or ligand blocking. However, several major classes of therapeutically relevant proteins have unknown or inaccessible activity profiles and so cannot be targeted by such strategies. Protein-degradation platforms such as proteolysis-targeting chimaeras (PROTACs)1,2 and others (for example, dTAGs3, Trim-Away4, chaperone-mediated autophagy targeting5 and SNIPERs6) have been developed for proteins that are typically difficult to target; however, these methods involve the manipulation of intracellular protein degradation machinery and are therefore fundamentally limited to proteins that contain cytosolic domains to which ligands can bind and recruit the requisite cellular components. Extracellular and membrane-associated proteins—the products of 40% of all protein-encoding genes7—are key agents in cancer, ageing-related diseases and autoimmune disorders8, and so a general strategy to selectively degrade these proteins has the potential to improve human health. Here we establish the targeted degradation of extracellular and membrane-associated proteins using conjugates that bind both a cell-surface lysosome-shuttling receptor and the extracellular domain of a target protein. These initial lysosome-targeting chimaeras, which we term LYTACs, consist of a small molecule or antibody fused to chemically synthesized glycopeptide ligands that are agonists of the cation-independent mannose-6-phosphate receptor (CI-M6PR). We use LYTACs to develop a CRISPR interference screen that reveals the biochemical pathway for CI-M6PR-mediated cargo internalization in cell lines, and uncover the exocyst complex as a previously unidentified—but essential—component of this pathway. We demonstrate the scope of this platform through the degradation of therapeutically relevant proteins, including apolipoprotein E4, epidermal growth factor receptor, CD71 and programmed death-ligand 1. Our results establish a modular strategy for directing secreted and membrane proteins for lysosomal degradation, with broad implications for biochemical research and for therapeutics.

Unlike the proteasomal pathway, the lysosomal pathway for protein degradation is not limited to proteins that have intracellular domains. Families of cell-surface lysosome-targeting receptors (LTRs) have been reported that facilitate the transport of proteins to lysosomes9. We proposed that chimeric molecules capable of binding both a cell-surface LTR and an extracellular protein might induce internalization and lysosomal degradation of the target, offering a means to accelerate the degradation of proteins by using binders that act in the extracellular space. The prototypical LTR is the cation-independent mannose-6-phosphate receptor (CI-M6PR, also called IGF2R), which endogenously transports proteins bearing N-glycans capped with mannose-6-phosphate (M6P) residues to lysosomes10. The receptor cycles constitutively between endosomes, the cell surface and the Golgi complex. CI-M6PR shuttles cargo efficiently to prelysosomal

compartments, where a lowered pH enables cargo to dissociate and progress to the lysosome while CI-M6PR is recycled. On the basis of its ability to efficiently deliver proteins to lysosomes and its expression in a range of different tissue types, CI-M6PR has been exploited to deliver therapeutic enzymes for the treatment of lysosomal storage disorders11. Here we present lysosome-targeting chimaeras—which we term LYTACs (Fig. 1a)—that enable the depletion of secreted and membrane-associated proteins via a mechanism of action that is orthogonal and complementary to that of existing technologies12, and which does not rely on an extracellular protease for efficacy13. We show that LYTACs can serve as biochemical probes for the study of receptor trafficking and protein degradation, and can mediate the degradation of both secreted and membrane-associated proteins of therapeutic interest.

Department of Chemistry, Stanford University, Stanford, CA, USA. 2Howard Hughes Medical Institute, Stanford, CA, USA. ✉e-mail: [email protected]

1

Nature | Vol 584 | 13 August 2020 | 291

Article b

LYTAC

Target

AcO AcO AcO

OAc O

AcO AcO

13 steps

OAc O

CI-M6PR

O

O

n

Merge

P = 0.0003 P < 0.0001 P = 0.0005

3 2 P = 0.0002

P < 0.0001

1

-2

3

M DL

H

1 36

EK

H

B-

29

a

0

-M

Hoechst

4

62

LysoTracker

5

DA

Poly(M6Pn) S: 9.7 KDa L: 37.1 KDa

NS

poly(GalNAc) poly(M6Pn)

P < 0.0001

6

eL

OH O

P < 0.0001

7

H

OH O P OH

NA-647

P < 0.0001

8

at

Poly(mannose) 18.9 KDa

g

f P < 0.0001 NS P = 0.0255 P = 0.0471

K5

OH O

9 8 7 6 5 4 3 2 1 0

1

e

Poly(M6Pn)-biotin

M

Poly(M6P) S: 8.2 KDa L: 24.7 KDa

R n

HO HO HO

HO HO

–––

N H

Me

rk

OH O P OH OH O O HO HO

O

H

O

Lysosome

P-

Poly(GalNAc) 27.1 KDa

N H

O

TH

OH OH O HO AcHN

K562

O

Fold change in MFI

Me

2. TMSBr, CH2Cl2 then K2CO3

O

H N

NA-647

Po l Po y(G ly( alN m an Ac) Po nos e ly( M ) Po 6P ly( ) Po M6 S ly( P) Po M6 L ly( Pn P M6 S) + oly( Pn) 10 M L m 6Pn M ) M S 6P

O

O

M6Pn-NCA

c

get Tar

H N

O

HN

CI-M6PR

d

O

Poly(M6Pn-co-Ala) M6Pn = 20–90

OH O

HO HO

O

HN

O

OAc

OH O P OH

Me

Ju

t ge Tar

Degradation

1. Ni(COD)2, bpy THF, 48 h, RT

OEt O P OEt

Fold change in MFI

a

Fig. 1 | LYTACs using CI-M6PR traffic proteins to lysosomes. a, The concept of LYTACs, in which a glycopolypeptide ligand for CI-M6PR is conjugated to an antibody to traffic secreted and membrane-associated proteins to lysosomes. b, Synthesis of M6Pn glycopolypeptide ligands for CI-M6PR. bpy, bipyridyl; COD, cyclooctadiene; RT, room temperature; THF, tetrahydrofuran; TMS, trimethylsilyl. c, Assay for the internalization of NA-647 by biotin-based LYTACs. d, Panel of synthetic M6P and M6Pn glycopolypeptides and controls. L, long; S, short. e, Fold changes in mean fluorescence intensity (MFI) for K562 cells incubated at 37 °C for 1 h with 500 nM NA-647 or 500 nM NA-647 and 2 μM

biotinylated glycopolypeptide in complete growth media. MFI was determined by live-cell flow cytometry. f, Live-cell confocal microscopy images of K562 cells treated as in e, then labelled with LysoTracker Green for 30 min. Scale bar, 10 μm. g, Panel of cell lines for NA-647 uptake experiments, performed as in e. For e, g, data are mean ± s.d. of three independent experiments. For f, images are representative of two independent experiments. P values were determined by unpaired two-tailed t-tests; fold changes are reported relative to incubation with protein targets alone. NS, not significant.

To develop ligands for CI-M6PR, we leveraged precedents aimed at enhancing lysosomal enzyme replacement therapies and drug delivery platforms. This previous work included biosynthetic engineering of M6P-bearing glycans14 as well as chemical synthesis of oligomeric M6P-containing scaffolds15. A recurring design parameter from these studies was the requirement for a multivalent ligand, in order to achieve optimal CI-M6PR agonism16–18. Previous work had also revealed that the 6-phosphoester could undergo hydrolysis in human serum19, leading to rapid endocytosis by macrophages bearing mannose receptors20. We reasoned that N-carboxyanhydride (NCA)-derived glycopolypeptides21 bearing multiple serine-O-mannose-6-phosphonate (M6Pn) residues22 would enable multivalent presentation on a biocompatible, phosphatase-inert23 and modular scaffold. We synthesized M6Pn glycopolypeptides starting with the conversion of mannose pentaacetate to M6Pn-NCA in 13 steps (Fig. 1b, Extended Data Fig. 1a). Subsequent copolymerization of M6Pn-NCA and alanine-NCA (1:1 ratio, for the purpose of spacing and polymerization kinetics) provided access to M6Pn glycopolypeptides (post-deprotection dispersity, Đ = 1.3–1.5) of various lengths, including short (20 M6Pn) and long (90 M6Pn) variants, to test in protein degradation assays. We also synthesized the corresponding M6P-containing copolymers bearing the natural phosphorylated glycan structure (Extended Data Fig. 1b), with similar molecular weights and dispersities to those obtained with M6Pn monomers. To demonstrate the feasibility of CI-M6PR-driven LYTACs, we designed an in cellulo assay to measure the uptake of NeutrAvidin-647 (NA-647), an Alexa Fluor-647 (AF647)-labelled protein to which biotin binds (Fig. 1c). We synthesized and biotinylated poly(M6P) and

poly(M6Pn) polypeptides, as well as poly(N-acetylgalactosamine) (poly(GalNAc)) and poly(mannose) as controls (Fig. 1d, Extended Data Figs. 1c, d, 2a). K562 cells were incubated with either NA-647 or NA-647 and biotinylated glycopolypeptide for one hour, then washed and analysed by flow cytometry. Co-incubation with M6P and M6Pn polypeptides increased cellular fluorescence by 5–6-fold compared with the background, with only minor differences in uptake efficiency observed as a result of glycopolypeptide length (short versus long, Fig. 1e). M6Pn polypeptides showed performance that was equivalent or superior to that of M6P polypeptides of similar length, whereas incubation with mannose- or GalNAc-containing glycopolypeptides showed no marked change in fluorescence compared with the background. NA-647 uptake mediated by poly(M6Pn) was attenuated by co-incubation with excess exogenous M6P. Furthermore, uptake remained continuous over time (Extended Data Fig. 2b), suggesting that surface-receptor recycling was the rate-limiting step24. AF647 is reported to be stable in endosomes and lysosomes25, with a steadily increasing fluorescent signal arising from intracellular fluorophore accumulation. Live-cell fluorescence microscopy experiments revealed that the AF647 signal co-localized with acidic endosomes and lysosomes after only 1 h (Fig. 1f, Extended Data Figs. 2c, d). Finally, biotinylated LYTACs mediated NA-647 uptake in various cell lines, demonstrating the breadth of CI-M6PR-targeting (Fig. 1g) as well as the ability to use small molecules as LYTAC ‘warheads’. We next performed a CRISPR interference (CRISPRi) pooled genetic screen with the aim of identifying genes for which knockdown ablates the delivery of NA-647 by biotinylated LYTACs26,27 (Fig. 2a). K562 cells expressing dCas9-KRAB were transduced with a genome-wide library

292 | Nature | Vol 584 | 13 August 2020

b

40 20 0 102 103 104 105 Cell surface CI-M6PR (AF-647)

8

g

A

N

1 C O

gR tro on C

F2

ls

sg 1 C

O

RN

RN

A

A

A

A N

RN

gR ls

R

F2

on

IG

RN

VINC

sg

VINC

A

HeLa

EXOC1

C

0

60

EXOC1

tro

0.2

Control sgRNA IGF2R sgRNA EXOC1 sgRNA EXOC2 sgRNA

80

CI-M6PR

NS

0.4

100

CI-M6PR

sg

0.6

K562

EX

0.8

f P < 0.0001

1.0

P < 0.0001

Relative surface CI-M6PR

1.2

P < 0.0001

e

2 4 6 log2(Enrichment)

Gene K562

EX

0 HeLa

0

d

Vacuolar membrane Lysosomal membrane Lytic vacuole membrane Coated vesicle Golgi transport complex Clathrin-coated vesicle TRAPP complex Tethering complex H+-transp. two-sector ATPase complex Exocyst complex CORVET complex H+-transp. V-type ATPase, V1 domain

5% FDR

sg

c

EXOC3 EXOC4

5

IG

Genes regulating M6Pn-mediated uptake

Amplification and sequencing

EXOC2 IGF2R

EXOC1

10

R

NA-647 uptake and FACS

15

Positive selection score

CRISPRi library infection

Normalized to mode

a

R

F2

IG

C

on

tro

ls

gR N A s EX gR O N C A 1 sg EX RN O C A 2 sg RN A

–0.2

Fig. 2 | CRISPRi screen identifies key cellular machinery for LYTACs. a, Schematic of a CRISPRi screen in K562 cells stably expressing dCas9-KRAB and a library of sgRNAs with genome-wide coverage. b, Selected gene hits for regulation of NA-647 internalization by LYTACs. c, Gene ontology (GO) annotation for significant hits (