Vol 585, 24 September 2020 Nature

Citation preview

The international journal of science / 24 September 2020

Biodiversity: what China’s researchers can show the world

economic development with controlling species and ecosystems loss. The world needs to hear these stories, in all their complexity.

Learn from China

As China prepares to take on a crucial role in the governance of global biodiversity, its researchers need to be at the table.

L

ast week, the United Nations confirmed that the world has failed, again, to achieve its goals to protect nature. This grim conclusion was delivered in the fifth edition of the United Nations Global Biodiversity Outlook report. The report from the UN Convention on Biological Diversity reviewed progress towards 20 biodiversity targets that the convention’s participating countries set for themselves in Aichi, Japan, a decade ago (www. cbd.int/gbo). None of the targets, which include making progress towards the sustainable harvesting of fish, controlling the spread of invasive species and preventing the extinction of threatened wildlife, will have been achieved by the deadline at the end of this year. This is no time for regret or apology, but for urgency to act. Last year, an analysis by the Intergovernmental Science-Policy Platform on Biodiversity and Ecosystem Services revealed that some one million plant and animal species are at risk of extinction. And the wildlife charity WWF’s latest Living Planet Index, published earlier this month (see go.nature.com/32wzvdz), was similarly sobering, stating that vertebrate populations monitored between 1970 and 2017 have declined by an average of 68%. All nations must do more, but some of the greatest responsibility now rests on the shoulders of China: the nation, along with the leaders of the UN biodiversity convention, will jointly host the next Conference of the Parties (COP) in Kunming next year. That summit, originally scheduled for this year, is where biodiversity targets for the next decade must be set. As we have written before, the previous targets were destined to fail, in part because their format made progress hard to measure, and because countries did not need to report on what they were doing. This must now change. The targets, furthermore, need to be more closely aligned with the UN System of Environmental Economic Accounting, which is becoming the global standard for environmental reporting. Without these changes, the next set of biodiversity targets will almost certainly fail again. At the same time, China’s biodiversity scientists and policy researchers should be at the table, too, as plans for Kunming start to take shape. The country has decades of experience of studying how to — and how not to — balance

All sides must put aside political differences to agree on ambitious targets.”

The Global Biodiversity Outlook report confirms that known species are on an accelerated path to extinction, with cycad and coral species among the groups most at risk. The report shows that, although deforestation has slowed in the past decade, forests are still being splintered by agriculture, tree-felling and urban growth. Such fragmentation will further harm biodiversity and increase carbon emissions. Demand for food and agricultural production continue to be the main drivers of biodiversity loss. And governments are not helping. On average, they invest some US$500 billion per year in initiatives that harm the environ­ ment — eclipsing financing for biodiversity projects by a factor of 6, the report says. China has a set of experiences that could help the world learn valuable lessons. Its rapid economic growth lifted a generation out of poverty; however, this created a cascade of environmental problems, not least elevated pollution in the air and on land. People in China rightly questioned their leaders for underestimating — if not downplaying — the environmental and social impacts of its industrialization. Partly in response, China’s authorities have been working with researchers from China and around the world to chart a greener way forward. For example, national and local administrations have been devising and experimenting with environmental targets, and creating mechanisms for monitoring and reporting progress towards them — albeit with mixed success. China’s national biodiversity strategy includes creating what it calls ‘redlines’ — areas where human activities are restricted to protect biodiversity — across the country. Then there’s China’s US$6-trillion Belt and Road Initiative — a massive programme to build roads, ports and infrastructure, which will run through natural habitats across Asia, Europe and Africa. Much of this investment did not initially come with safeguards to mitigate environmental risks — but these are now being actively studied. And last but not least, China has a large community of researchers working to quantify, in monetary terms, the value of natural capital and ecosystem services, so that people and policymakers can more clearly understand that nature’s services to people do not come for free. On 30 September, heads of governments will meet at the UN for a day of talks on biodiversity, ahead of next year’s Kunming COP. Nature spoke to a number of representatives of national delegations who plan to attend this meeting, including researchers and non-governmental observers. All want the Kunming COP to succeed in bringing nations together and reaching an agreement on targets that are measurable and meaningful. But they expressed concern over the limited public engagement from China’s government about its goals or strategy for Kunming — and the relatively limited involvement of its researchers in the process so far. Scientists in China have been central to their country’s

. 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 585 | 24 September 2020 | 481

Editorials

conservation and economic-development journey. Their collective experience on what works, and what doesn’t, can provide important learning opportunities for countries as they look to slow down and eventually reverse bio­diversity and ecosystem loss. These researchers are in the academy of sciences; in universities; in the academy of environmental planning; and in the community of Chinese and international non-governmental organizations. Many are also active in the China Council for Inter­ national Cooperation on Environment and Development, an organization located in both Canada and China, which last week concluded a two-day conference presenting its latest research outputs. This important but little-known advisory body, now nearly three decades old, has been instrumental in connecting China’s environmental-science and environmental-policy communities with international counterparts. Next year will be the first time that China has hosted an international environmental meeting — similar to the 2015 Paris climate accords — where the stakes are too high to fail. It must draw on its rich diversity of talent and experience. Other nations’ researchers must be equally forthcoming with their knowledge. All sides must put aside political differences to agree on ambitious targets, ways to achieve them and methods to measure that progress. The best way to preserve and revive biodiversity is to acknowledge where we’ve all failed it before, to learn from that and to try again, together.

The education revolution must be equalized The switch to online learning risks widening educational inequalities.

E

very day, hundreds of millions of students, teachers and support staff, are participating in a learning revolution: the COVID-19 pandemic has upended the centuries-old tradition that students travelled to a physical institution to learn. Now, in many places, school and university classrooms are on laptops and smartphone screens, and the Internet has replaced physical books. It’s been an extraordinary — and extraordinarily fast — transition, affecting everyone from the youngest children entering school right up to young adults in universities. Researchers are starting to study its full impact and its implications — for students, for staff and for the organizations that create and supply educational-technology platforms. Tertiary education has been venturing into online education for some time. Long before the pandemic, universities

. 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 ©

482 | Nature | Vol 585 | 24 September 2020

A staggering 850 million children and young adults are not in education or training because of COVID-19.”

around the world were offering massive open online courses (MOOCs) as a supplement to face-to-face teaching and learning. Now, as online courses become more central to university teaching, it will be important to rigorously assess the impact of this change. We already knew that this educational revolution presents significant risks. Before the pandemic, countries were making good progress towards ensuring that by 2030 children would at least complete a primary-school education — one of the few United Nations Sustainable Development Goals potentially within reach. That might no longer be the case — a prospect that should worry us all. As of this week, a staggering 850 million children and young adults — half of those enrolled in schools, colleges and universities worldwide — are not in education or training because of COVID-19, according to the UN science and education organization UNESCO. The agency is also tracking closures of schools up to secondary level daily and, although schools are reopening in many places, they remain closed in 52 countries. The majority affected are in the southern half of the globe, encompassing many low- and middle-income countries. That means that students there are much less likely to be taking part in the online revolution. Internet penetration in this hemisphere is low — and some 360 million young people do not have access, according to the International Telecommunications Union. Many countries are using terrestrial television and radio to broadcast lessons as a lower-cost alternative to broadband. While the pandemic continues, reopening educational institutions in poorer parts of the world — including deprived areas in high-income countries — is often not possible. Overcrowding prevents social distancing, and funding isn’t available to make schools COVID-19 secure. All this means that students from the poorest families, without Internet access, are more likely to be denied education — widening already deep educational inequalities. Because education is strongly linked to later jobs, income and health, setbacks now will last a lifetime. In universities, the transition to online education is enabling institutions to reach out to students from underserved areas and under-represented communities. But paradoxically, if children from these communities are unable to access earlier schooling, fewer will be able to proceed to higher education. The pandemic will force a large number of institutions will remain closed, and online learning will substitute for the real thing. But if broadband and laptops are the equivalent of the teacher, the library and the laboratory, it cannot be acceptable that these are available to only a fraction of students. If online education is to become more inclusive, public educational institutions — and those that fund them — must do more to ensure that more learners can benefit from new technologies. That includes prioritizing access to broadband, smartphones and laptops — something that is increasingly affordable in many countries. It’s a small price to pay now for an educated and resilient population decades down the line.

Editorials

conservation and economic-development journey. Their collective experience on what works, and what doesn’t, can provide important learning opportunities for countries as they look to slow down and eventually reverse bio­diversity and ecosystem loss. These researchers are in the academy of sciences; in universities; in the academy of environmental planning; and in the community of Chinese and international non-governmental organizations. Many are also active in the China Council for Inter­ national Cooperation on Environment and Development, an organization located in both Canada and China, which last week concluded a two-day conference presenting its latest research outputs. This important but little-known advisory body, now nearly three decades old, has been instrumental in connecting China’s environmental-science and environmental-policy communities with international counterparts. Next year will be the first time that China has hosted an international environmental meeting — similar to the 2015 Paris climate accords — where the stakes are too high to fail. It must draw on its rich diversity of talent and experience. Other nations’ researchers must be equally forthcoming with their knowledge. All sides must put aside political differences to agree on ambitious targets, ways to achieve them and methods to measure that progress. The best way to preserve and revive biodiversity is to acknowledge where we’ve all failed it before, to learn from that and to try again, together.

The education revolution must be equalized The switch to online learning risks widening educational inequalities.

E

very day, hundreds of millions of students, teachers and support staff, are participating in a learning revolution: the COVID-19 pandemic has upended the centuries-old tradition that students travelled to a physical institution to learn. Now, in many places, school and university classrooms are on laptops and smartphone screens, and the Internet has replaced physical books. It’s been an extraordinary — and extraordinarily fast — transition, affecting everyone from the youngest children entering school right up to young adults in universities. Researchers are starting to study its full impact and its implications — for students, for staff and for the organizations that create and supply educational-technology platforms. Tertiary education has been venturing into online education for some time. Long before the pandemic, universities

. 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 ©

482 | Nature | Vol 585 | 24 September 2020

A staggering 850 million children and young adults are not in education or training because of COVID-19.”

around the world were offering massive open online courses (MOOCs) as a supplement to face-to-face teaching and learning. Now, as online courses become more central to university teaching, it will be important to rigorously assess the impact of this change. We already knew that this educational revolution presents significant risks. Before the pandemic, countries were making good progress towards ensuring that by 2030 children would at least complete a primary-school education — one of the few United Nations Sustainable Development Goals potentially within reach. That might no longer be the case — a prospect that should worry us all. As of this week, a staggering 850 million children and young adults — half of those enrolled in schools, colleges and universities worldwide — are not in education or training because of COVID-19, according to the UN science and education organization UNESCO. The agency is also tracking closures of schools up to secondary level daily and, although schools are reopening in many places, they remain closed in 52 countries. The majority affected are in the southern half of the globe, encompassing many low- and middle-income countries. That means that students there are much less likely to be taking part in the online revolution. Internet penetration in this hemisphere is low — and some 360 million young people do not have access, according to the International Telecommunications Union. Many countries are using terrestrial television and radio to broadcast lessons as a lower-cost alternative to broadband. While the pandemic continues, reopening educational institutions in poorer parts of the world — including deprived areas in high-income countries — is often not possible. Overcrowding prevents social distancing, and funding isn’t available to make schools COVID-19 secure. All this means that students from the poorest families, without Internet access, are more likely to be denied education — widening already deep educational inequalities. Because education is strongly linked to later jobs, income and health, setbacks now will last a lifetime. In universities, the transition to online education is enabling institutions to reach out to students from underserved areas and under-represented communities. But paradoxically, if children from these communities are unable to access earlier schooling, fewer will be able to proceed to higher education. The pandemic will force a large number of institutions will remain closed, and online learning will substitute for the real thing. But if broadband and laptops are the equivalent of the teacher, the library and the laboratory, it cannot be acceptable that these are available to only a fraction of students. If online education is to become more inclusive, public educational institutions — and those that fund them — must do more to ensure that more learners can benefit from new technologies. That includes prioritizing access to broadband, smartphones and laptops — something that is increasingly affordable in many countries. It’s a small price to pay now for an educated and resilient population decades down the line.

The world this week

News in brief GENOMES SHOW CORONAVIRUS SPREAD ON FLIGHT

PHOTO: BEN STANSALL/AFP VIA GETTY; DATA SOURCE: V. HAAG ET AL. IAWA J. HTTPS://DOI.ORG/D9N8 (2020)

A GUIDE TO MAKING ‘COCKTAILS’ TO TREAT COVID-19 A new method pinpoints every mutation that a crucial SARS‑CoV-2 protein could use to evade an attacking antibody. The results could inform the development of antibody treatments for COVID-19. The immune system produces molecules called antibodies to fend off invaders. Antibodies that bind to an important region of the SARS-CoV-2 spike protein can inactivate the viral particles, making such antibodies attractive as therapies. But over time, viruses can accumulate mutations — and some can interfere with antibody binding and allow viral particles to ‘escape’ immune forces. James Crowe at the Vanderbilt University Medical Center in Nashville, Tennessee, Jesse Bloom at the Fred Hutchinson Cancer Center in Seattle, Washington, and their colleagues created the most detailed map so far of the spikeprotein mutations that could prevent binding by ten human antibodies (A. J. Greaney et al. Preprint at bioRxiv https://doi. org/d8zm; 2020). The team then used that information to design three antibody cocktails, each consisting of two antibodies. In laboratory tests of the cocktails against SARS-CoV-2, the virus did not develop mutations that could escape antibody binding. The findings have not yet been peer reviewed.

Genetic evidence strongly suggests that at least one member of a married couple flying from the United States to Hong Kong infected two flight attendants during the trip. Researchers led by Leo Poon at the University of Hong Kong and Deborah Watson-Jones at the London School of Hygiene & Tropical Medicine studied four people on the early-March flight (E. M. Choi et al. Emerg. Infect. Dis. https://doi.org/d9jn; 2020). Two were a husband and wife travelling in business class. The others were crew members: one in business class and one whose cabin assignment is unknown. The passengers had travelled in Canada and the United States before the flight and tested positive for the new coronavirus soon after arriving in Hong Kong. The flight attendants tested positive shortly thereafter. The team found that the viral genomes of all four were identical and that their virus was a close genetic relative of some North American SARS‑CoV-2 samples — but not of the SARS‑CoV-2 prevalent in Hong Kong. This suggests that one or both of the passengers transmitted the virus to the crew members during the flight, the authors say. The authors add that no previous reports of in-flight spread have been supported by genetic evidence.

Unsustainable charcoal Nearly half of charcoal bought in Europe contains wood from tropical and subtropical forests, with little of it certified as sustainable, raising fears that some is illegally logged. “This is just an overview but it’s absolutely enough to cause alarm,” says study leader Volker Haag, a wood anatomist at the Thünen Institute of Wood Research in Hamburg, Germany (V. Haag et al. IAWA J. https://doi. org/d9n8; 2020). Haag’s team used a microscopy technique that digitally reconstructs sections of charcoal from irregular lumps to create images from which the wood can be identified. They analysed 4,500 samples from 150 charcoal bags bought in 11 countries. Some 46% included wood from subtropical and tropical forests, which have high rates of deforestation. Of these, only one-quarter of bags bore the logos of sustainablecertification organizations. In addition, only onequarter of the bags specified the species or origin of the wood — and only half of these were correct. A wrongly labelled product is a strong indicator of illegality, says co-author Johannes Zahnen, a forestpolicy specialist at WWF Germany in Berlin.

CHARRED ORIGINS

Nearly half of charcoal bags bought in Europe contain wood that originates from tropical or subtropical forests and many are incorrectly labelled, raising fears that the material has been illegally logged. Subtropical or tropical wood

Certified sustainable

Incorrect or incomplete label

Belgium Spain Italy Poland Czech Republic Netherlands Denmark Germany Switzerland Norway 0

20

40 60 Percentage

80

100

. 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 585 | 24 September 2020 | 487

The world this week

SAUL LOEB/AFP/GETTY

News in focus

Israeli Prime Minister Benjamin Netanyahu, UAE foreign minister Abdullah bin Zayed Al Nahyan and his Bahraini counterpart Abdullatif Al Zayani.

ISRAEL–ARAB PEACE ACCORD FUELS HOPE FOR SURGE IN SCIENTIFIC COLLABORATION Space, water, food security and archaeology present opportunities for joint research as the United Arab Emirates and Bahrain end boycott of Israel. By Elizabeth Gibney

A

peace accord between Israel and the United Arab Emirates (UAE) is expected to lead to a surge in scientific collaboration between the countries — with the promise of joint research in space exploration, water and food security, along with exploration of the region’s shared archaeological heritage. For the first time since the UAE’s founding in 1971, Emiratis will be able to work and travel in Israel, and Israelis the same in the UAE. Previously, this was possible only in exceptional circumstances. Researchers, moreover, will be free to exchange materials, including biological samples and scientific equipment. The agreement to normalize diplomatic relations, called

the Abraham Accords — which also includes the Gulf state of Bahrain — was signed at the White House in Washington DC on 15 September. Experts told Nature that Emirati scientists could benefit from Israel’s well-established research base and collaborations with its technology firms, and Israeli scientists could gain from tapping into the UAE’s growing investment in research, diverse population and technological infrastructure. “What excites me, personally, is the UAE beginning to look at Israel as a potential friend, rather than a risk,” says Mohammed Baharoon, director-general of b’huth, a public-policy research centre in Dubai, UAE. But change will not happen overnight, he cautions. Shai-Lee Spigelman, director-general of the Israeli Ministry of Science and Technology, was

part of a US–Israeli delegation to the UAE on 31 August, which included a working group on space and science. “The meetings were really impressive and interesting and open. It really felt like both sides want to cooperate, want to find mutual ways to work together,” she says. Two universities have already signed an agreement to work together, the first of its kind between the countries. The Mohamed bin Zayed University of Artificial Intelligence in Abu Dhabi and the Weizmann Institute of Science in Rehovot, Israel, plan to create a joint virtual institute for artificial intelligence. Since Israel was founded in 1948, nations in the Arab League have been opposed to the Jewish state over the issue of Palestinian independence. Most have refused to deal with the country ever since: Bahrain and the UAE

. 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 585 | 24 September 2020 | 489

News in focus are only the third and fourth Arab countries to establish formal diplomatic relations with Israel, following Egypt in 1979 and Jordan in 1994. Until now, Israeli citizens had generally been barred from entering the UAE, and although Israel had no law banning UAE citizens, entry required explicit permission from the Ministry of Foreign Affairs. But a new generation of Gulf leaders, backed by the administration of US President Donald Trump, is challenging that narrative. Between the UAE and Israel, scientific cooperation is a high priority, says Spigelman. The 31 August meeting included early discussions about potential cooperation on satellites and experiments in low Earth orbit, as well as coordinating astronaut visits to the International Space Station, she says. “They didn’t sound like they were new in this neighbourhood, even though they are. So it was very impressive,” she adds. Israeli firm SpaceIL in Tel Aviv launched a government-backed mission to the Moon in 2019, although the lander crashed. The UAE has a human space-flight programme and was one of three nations to launch a Mars mission in July. Future collaborations are also likely to focus on artificial intelligence and quantum science, as well as agriculture, desert studies and water security, says Spigelman. Both countries are also carrying out extensive research in cybersecurity, energy and desalination technology. A UAE-based researcher who studies ancient civilization in the Middle East, and who asked not to be named because of the sensitivities surrounding the accords, says archaeology should also benefit. The UAE’s boycott of Israel meant that exchanging artefacts and samples had until now been a problem, she says. “There were some civilizations that lived in the Gulf region and also moved into the territories of Israel today, so I don’t really know how those civilizations are currently studied.”

Running start Collaboration will not start from scratch. Researchers from the UAE and Israel co-authored 248 papers between 2017 and 2019, according to the Scopus database (including co-authorship as part of mega collaborations, such as experiments at Europe’s particle-physics laboratory, CERN). This compares with 183 papers co-authored by scientists in Israel and Egypt during the same period, and 98 between Israel and Jordan. UAE universities awarded their first PhDs only in 2010, and many senior academics there come from other countries, which do have diplomatic ties to Israel. Moreover, technology businesses in the UAE — as well as Qatar and Saudi Arabia — already have informal relationships with counterparts in Israel to procure what are viewed as crucial technologies, such as those used in protecting oil and gas infrastructure, says Robert Mogielnicki, a researcher in political

Criticisms and concerns But sensitivities remain. Nature had difficulty finding Emirati scientists willing to speak about collaboration with Israel (people in the UAE can be jailed for speaking against government policy). And Palestinian academics are angry about the accords, says philosopher Sari Nusseibeh, former president of Al-Quds University in East Jerusalem. But Nusseibeh is confident that the agreement will boost Palestinian involvement in research collaboration. “Can the UAE use its new partner to help Palestinians? I am sure it can,” says Nusseibeh. “Given the Palestinian suffering under occupation, the sky is the limit as to what it can do. Let us hope it does.” At present, Palestinian scientists have restrictions on where they can travel, and on the materials they can import, says Mario

Martone, a particle physicist at the University of Texas at Austin and co-founder of the advocacy group Scientists for Palestine. Baharoon says that Emirati researchers are unlikely to let politics influence their business or life decisions, and that that attitude bodes well for future research collaborations. “From a number of people I spoke to, I think there is an admiration of Israel as the start-up nation, and one that has done a lot when it comes to science and technology,” he says. But Mogielnicki cautions that although governments are excited about the prospects for research and development, relationships between individual Israeli and Emirati academics will be key to success. “How will researchers in both countries navigate potentially awkward relations with colleagues, that are a bit more conservative and do not feel as optimistic about this normalization? That’s a big question that remains to be seen,” he says. Nonetheless, Baharoon hopes the accord will prove to be a ‘proof of concept’ for other Gulf countries. Bahrain publicized its intention to normalize relations with Israel just weeks after the UAE’s own announcement, and there is speculation that others will follow. Spigelman also hopes that the accord will inspire similar deals between Israel and other nations. “There are other very advanced countries in the Gulf with strong universities and resources in science and technology, and we would love to cooperate with them,” she says. Nature asked a representative of the UAE Ministry of Foreign Affairs and International Cooperation for comment, but the ministry did not respond by the time this article went to press.

STILLBIRTH RATE RISES DURING CORONAVIRUS PANDEMIC Emerging data link disrupted antenatal services to a rise in pregnancy complications in several countries. By Clare Watson

D

isruptions brought about by the COVID-19 pandemic have had a profound effect on health care worldwide, contributing to an increase in deaths from chronic conditions such as heart disease. Now, a slew of studies has reported a significant rise in the proportion of pregnancies ending in stillbirth, in which babies die in the womb. Researchers say that, in some countries, pregnant women have received less care

. 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 ©

490 | Nature | Vol 585 | 24 September 2020

economy at the Arab Gulf States Institute in Washington DC. Both countries already have extensive links with China, he adds. But now, researchers are looking forward to forming more and deeper connections. Not only can Israeli collaborators now visit, but UAE institutions can begin student exchanges, says Andrea Macciò, an Italian astrophysicist at New York University Abu Dhabi, who frequently collaborates with Israeli colleagues. Israel is “one of the closest countries in the area with a substantial research programme”, says Macciò, who hopes the accord will lead to institutional-level collaborations, as well as research calls for joint programmes and regional scientific summits. Spigelman says that the countries could indeed sign a bilateral scientific agreement under which they release joint funding calls.

than they need because of lockdown restrictions and disruptions to health care. As a result, complications that can lead to stillbirths were probably missed, they say. “What we’ve done is cause an unintended spike in stillbirth while trying to protect [pregnant women] from COVID-19,” says Jane Warland, a specialist in midwifery at the University of South Australia in Adelaide. The largest study to report a rise in the stillbirth rate, based on data from more than 20,000 women who gave birth in 9 hospitals

News in focus are only the third and fourth Arab countries to establish formal diplomatic relations with Israel, following Egypt in 1979 and Jordan in 1994. Until now, Israeli citizens had generally been barred from entering the UAE, and although Israel had no law banning UAE citizens, entry required explicit permission from the Ministry of Foreign Affairs. But a new generation of Gulf leaders, backed by the administration of US President Donald Trump, is challenging that narrative. Between the UAE and Israel, scientific cooperation is a high priority, says Spigelman. The 31 August meeting included early discussions about potential cooperation on satellites and experiments in low Earth orbit, as well as coordinating astronaut visits to the International Space Station, she says. “They didn’t sound like they were new in this neighbourhood, even though they are. So it was very impressive,” she adds. Israeli firm SpaceIL in Tel Aviv launched a government-backed mission to the Moon in 2019, although the lander crashed. The UAE has a human space-flight programme and was one of three nations to launch a Mars mission in July. Future collaborations are also likely to focus on artificial intelligence and quantum science, as well as agriculture, desert studies and water security, says Spigelman. Both countries are also carrying out extensive research in cybersecurity, energy and desalination technology. A UAE-based researcher who studies ancient civilization in the Middle East, and who asked not to be named because of the sensitivities surrounding the accords, says archaeology should also benefit. The UAE’s boycott of Israel meant that exchanging artefacts and samples had until now been a problem, she says. “There were some civilizations that lived in the Gulf region and also moved into the territories of Israel today, so I don’t really know how those civilizations are currently studied.”

Running start Collaboration will not start from scratch. Researchers from the UAE and Israel co-authored 248 papers between 2017 and 2019, according to the Scopus database (including co-authorship as part of mega collaborations, such as experiments at Europe’s particle-physics laboratory, CERN). This compares with 183 papers co-authored by scientists in Israel and Egypt during the same period, and 98 between Israel and Jordan. UAE universities awarded their first PhDs only in 2010, and many senior academics there come from other countries, which do have diplomatic ties to Israel. Moreover, technology businesses in the UAE — as well as Qatar and Saudi Arabia — already have informal relationships with counterparts in Israel to procure what are viewed as crucial technologies, such as those used in protecting oil and gas infrastructure, says Robert Mogielnicki, a researcher in political

Criticisms and concerns But sensitivities remain. Nature had difficulty finding Emirati scientists willing to speak about collaboration with Israel (people in the UAE can be jailed for speaking against government policy). And Palestinian academics are angry about the accords, says philosopher Sari Nusseibeh, former president of Al-Quds University in East Jerusalem. But Nusseibeh is confident that the agreement will boost Palestinian involvement in research collaboration. “Can the UAE use its new partner to help Palestinians? I am sure it can,” says Nusseibeh. “Given the Palestinian suffering under occupation, the sky is the limit as to what it can do. Let us hope it does.” At present, Palestinian scientists have restrictions on where they can travel, and on the materials they can import, says Mario

Martone, a particle physicist at the University of Texas at Austin and co-founder of the advocacy group Scientists for Palestine. Baharoon says that Emirati researchers are unlikely to let politics influence their business or life decisions, and that that attitude bodes well for future research collaborations. “From a number of people I spoke to, I think there is an admiration of Israel as the start-up nation, and one that has done a lot when it comes to science and technology,” he says. But Mogielnicki cautions that although governments are excited about the prospects for research and development, relationships between individual Israeli and Emirati academics will be key to success. “How will researchers in both countries navigate potentially awkward relations with colleagues, that are a bit more conservative and do not feel as optimistic about this normalization? That’s a big question that remains to be seen,” he says. Nonetheless, Baharoon hopes the accord will prove to be a ‘proof of concept’ for other Gulf countries. Bahrain publicized its intention to normalize relations with Israel just weeks after the UAE’s own announcement, and there is speculation that others will follow. Spigelman also hopes that the accord will inspire similar deals between Israel and other nations. “There are other very advanced countries in the Gulf with strong universities and resources in science and technology, and we would love to cooperate with them,” she says. Nature asked a representative of the UAE Ministry of Foreign Affairs and International Cooperation for comment, but the ministry did not respond by the time this article went to press.

STILLBIRTH RATE RISES DURING CORONAVIRUS PANDEMIC Emerging data link disrupted antenatal services to a rise in pregnancy complications in several countries. By Clare Watson

D

isruptions brought about by the COVID-19 pandemic have had a profound effect on health care worldwide, contributing to an increase in deaths from chronic conditions such as heart disease. Now, a slew of studies has reported a significant rise in the proportion of pregnancies ending in stillbirth, in which babies die in the womb. Researchers say that, in some countries, pregnant women have received less care

. 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 ©

490 | Nature | Vol 585 | 24 September 2020

economy at the Arab Gulf States Institute in Washington DC. Both countries already have extensive links with China, he adds. But now, researchers are looking forward to forming more and deeper connections. Not only can Israeli collaborators now visit, but UAE institutions can begin student exchanges, says Andrea Macciò, an Italian astrophysicist at New York University Abu Dhabi, who frequently collaborates with Israeli colleagues. Israel is “one of the closest countries in the area with a substantial research programme”, says Macciò, who hopes the accord will lead to institutional-level collaborations, as well as research calls for joint programmes and regional scientific summits. Spigelman says that the countries could indeed sign a bilateral scientific agreement under which they release joint funding calls.

than they need because of lockdown restrictions and disruptions to health care. As a result, complications that can lead to stillbirths were probably missed, they say. “What we’ve done is cause an unintended spike in stillbirth while trying to protect [pregnant women] from COVID-19,” says Jane Warland, a specialist in midwifery at the University of South Australia in Adelaide. The largest study to report a rise in the stillbirth rate, based on data from more than 20,000 women who gave birth in 9 hospitals

ANTHONY WALLACE/AFP/GETTY

Researchers are concerned that high-risk pregnancies are going undetected.

across Nepal, was published in The Lancet Global Health on 10 August1. It reported that stillbirths increased from 14 per 1,000 births before the country went into lockdown to stop the spread of the coronavirus in late March, to 21 per 1,000 births by the end of May — a rise of 50%. The sharpest rise was observed during the first four weeks of the lockdown, under which people were allowed to leave their homes only to buy food and receive essential care. The study, led by Ashish K.C., a perinatal epidemiologist at Uppsala University, Sweden, and his colleagues, found that although the rate of stillbirths jumped, the overall number was unchanged during the pandemic. This can be explained by the fact that hospital births halved, from an average of 1,261 births each week before lockdown to 651. And a higher proportion of hospital births during lockdown had complications. The researchers don’t know what happened to women who didn’t go to hospital, or to their babies, so they cannot say whether the rate of stillbirths increased across the population. The increase in the proportion of stillbirths among hospital births was not caused by COVID-19 infections, says K.C.. Rather, it is probably a result of how the pandemic has affected access to routine antenatal care, which might have otherwise picked up complications that can lead to stillbirth, he says. Pregnant women might have been unable to travel to health facilities for lack of public transport; in some cases, antenatal appointments were reportedly cancelled. Others might have avoided hospitals for fear of contracting SARS-CoV-2, the virus that causes COVID-19, or had consultations by phone or Internet. “Nepal has made significant progress in the

last 20 years in health outcomes for women and their babies, but the last few months have deaccelerated that progress,” says K.C.. Birth data from a large hospital in London showed a similar trend. In July, Asma Khalil, an obstetrician at St George’s, University of London, and her colleagues reported2 a nearly fourfold increase in the incidence of stillbirths at St George’s Hospital, from 2.38 per 1,000 births between October 2019 and the end of January this year, to 9.31 per 1,000 births between February and mid-June. Khalil calls this the collateral damage of the pandemic. She says that, during lockdown, pregnant women might have developed

“What we’ve done is cause an unintended spike in stillbirth while trying to protect pregnant women from COVID-19.” complications that were not diagnosed, and might have hesitated about coming to hospital and therefore been seen by doctors only when a complication was advanced, when less could be done. Four hospitals in India also reported3 a jump in the stillbirth rate during the country’s lockdown. And as in Nepal, fewer women had their babies in those hospitals. Referrals of women requiring emergency pregnancy care also dropped by two-thirds. This suggests that more births were happening unattended, at home or in small facilities, according to the authors. Scotland — one of a few countries that collates data on stillbirths and infant deaths monthly — has also detected an uptick in the

rate of stillbirths during the pandemic. In normal times, the World Health Organization recommends that women be seen by medical professionals at least eight times during pregnancy — even if the pregnancy is judged low-risk — to detect and manage problems that might harm the mother, the baby or both. Much of the risk of stillbirth can be averted if women sleep on their side from 28 weeks’ gestation, stop smoking and notify their midwife or doctor if their baby is moving less. The last trimester of pregnancy is particularly important for regular health checks, but women are typically monitored for risk factors such as restricted fetal growth and high blood pressure throughout pregnancy. When the pandemic hit, professional bodies for maternity-health providers recommended that some face-to-face consultations be substituted with remote appointments to protect women from the coronavirus. But health-care workers can’t take someone’s blood pressure, listen to their baby’s heartbeat or do an ultrasound remotely, says Warland. Because of this, high-risk pregnancies might have been missed, she says, particularly among first-time mothers who are less likely to know what an abnormality feels like. For instance, St George’s Hospital reported a drop in the number of pregnant women presenting with high blood pressure during the UK lockdown. This suggests that “women with hypertension aren’t being managed as they normally would, and undetected hypertension is a risk factor for stillbirth”, says Warland. The studies are a call to arms to support maternal and newborn health services, especially in low-to middle-income countries, says Caroline Homer, a midwifery researcher at the Burnet Institute in Melbourne, Australia. “This is not the moment to reduce” these services, she says. Homer says that, across the Asia-­ Pacific region, the maternal-health workforce has pivoted to working on the COVID-19 front line, and antenatal care services have reduced face-to-face contact with pregnant women. In some places, services have shut completely, she says. But Pat O’Brien, the vice-president of the Royal College of Obstetricians and Gynaecologists in London, says the reasons behind this rise in the rate of stillbirths need further exploration. “We are aware anecdotally of pregnant women presenting late with reduced fetal movements, which can be a sign their baby is unwell, and of women missing antenatal appointments. This may be due to confusion around whether these appointments count as essential travel, fear of attending a hospital or not wanting to burden the NHS,” says O’Brien. 1. K.C., A. et al. Lancet Glob. Health https://doi.org/10.1016/ S2214-109X(20)30345-4 (2020). 2. Khalil, A. et al. J. Am. Med. Assoc. 324, 705–706 (2020). 3. Kumari, V., Mehta, K. & Choudhary, R. Lancet Glob. Health 8, E1116–E1117 (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 585 | 24 September 2020 | 491

Nations are making plans for how to allocate coronavirus vaccines once they’re ready.

WHO GETS A COVID VACCINE FIRST? ACCESS PLANS ARE TAKING SHAPE Advisory groups around the world release guidance to prioritize health-care and front-line workers. By Nidhi Subbaraman

W

hether it takes weeks, as US Presid­ ent Donald Trump has hinted, or months, as most health-care experts expect, an approved vaccine against the coronavirus is coming, and it’s hotly anticipated. Still, it will initially be in short supply while manufactur­ ers scale up production. As the pandemic con­ tinues to put millions at risk daily, including health-care workers, older people and those with pre-existing diseases, who should get vaccinated first? This week, a strategic advisory group at the World Health Organization (WHO) weighed in with preliminary guidance for global vaccine allocation, identifying groups that should be prioritized. These recommendations join a draft plan from a panel assembled by the US National Academies of Sciences, Engineering, and Medicine (NASEM), released earlier this month. Experts praise both plans for addressing the historic scale and unique epidemiology of the coronavirus pandemic. And they commend the NASEM for including in their guidance minority racial and ethnic groups — which COVID-19 has hit hard — by addressing the

Head of the queue The WHO’s guidance at this point lists only which groups of people should have priority access to vaccines. The NASEM guidance goes a step further by ranking priority groups in order of who should get a vaccine first. After health-care workers, medically vulnerable groups should be among the first to receive a vaccine, according to the NASEM draft plan. These include older people living in crowded settings, and individuals with multiple existing conditions, such as serious heart disease or diabetes, that put them at risk of more serious COVID-19 infection. The plan also prioritizes workers in essential

. 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 ©

492 | Nature | Vol 585 | 24 September 2020

socio-economic factors that put them at risk. The WHO plan, on the other hand, will need more detail before its recommendations can become actionable, others say. “It’s important to have different groups thinking through the problem,” says Eric Toner, an emergency-medicine physician and pandemics expert who has done similar planning at Johns Hopkins Center for Health Security in Baltimore, Maryland. And although the plans differ somewhat, Toner says he sees a lot of agreement. “It’s great that there’s a con­ sensus of opinion on these issues.”

industries, such as public transport, because their jobs place them in contact with many people. Similarly, people who live in certain crowded settings — homeless shelters and pris­ ons, for example — are called out as deserving early access. Many nations already have general vac­ cine-allocation plans, but they are tailored for an influenza pandemic rather than the new coronavirus. They typically prioritize children and pregnant women; the COVID-19 plans do not, however, because most vaccine trials currently do not include pregnant women, and the coronavirus seems to be less deadly to children than is flu. Unlike the NASEM guidance, the WHO plan notes that government leaders should have early access, but cautions that people prioritized in this way should be “narrowly interpreted to include a very small number of individuals”. “We were very concerned about the possi­ bility that this group could serve as a loophole through which a truckload of people who iden­ tify as important could then push themselves to the front of the line,” says Ruth Faden, a bioethicist at the Johns Hopkins Berman Insti­ tute of Bioethics in Baltimore, who was part of the group that drafted the WHO guidance.

Hard-hit groups Access for disadvantaged groups is addressed in both the plans. Looking to past failures, the WHO guidance urges richer countries to ensure that poorer countries receive vaccines in the earliest days of allocation. During the 2009 H1N1 flu pandemic, “by the time the world had gotten around to figuring out how to get vaccines to some low- and middle-income countries, the pandemic was over”, says Faden. But the WHO proposal does not yet sug­ gest how nations might resolve the tension between allocating vaccines in a country and allocating them between countries, says Angus Dawson, a bioethicist at the Univer­ sity of Sydney in Australia, who published a review of national pandemic allocation ethics earlier this year ( J. H. Williams and A. Dawson BMC Med. Ethics 21, 40; 2020). In other words, should harder-hit nations receive a bigger allo­ cation of an early vaccine before other nations have a chance to dose their high-priority groups? The NASEM was asked to develop its alloca­ tion plan by both the US Centers for Disease Control and Prevention (CDC), which will set the US government’s COVID-19 vaccination plan, and the US National Institutes of Health, which is coordinating vaccine and treatment trials. When tapping the NASEM to create the proposal, leaders from both agencies requested that the report address how to give vaccine priority to “populations at high risk”, including “racial and ethnic groups” that have been affected by COVID-19 and have died in

NATALIA KOLESNIKOVA/AFP/GETTY

News in focus

disproportionately higher numbers than have other groups in the United States. The panel determined that these groups are vulnera­ ble chiefly for socio-economic reasons tied to systemic racism — for example, they have high-risk jobs and live in high-risk areas — and therefore addressed the request through this lens, without singling out the groups because of their identities. “We really are trying to make sure that peo­ ple of colour, who have been disproportion­ ately impacted, will also have priority — but for the factors that put them at risk, not highlight­ ing just their racial and ethnic make-up,” says Helene Gayle, president and chief executive of the Chicago Community Trust in Illinois and a co-chair of the NASEM committee that drafted the proposal. Faden says the recommendations acknowl­ edge the current focus on racial injustice in the United States. “I was reading to see: does this report speak to the cultural moment in the United States, does it speak to racism and other forms of structural inequality? And it does,” she says. The WHO’s strategic advisory group will continue to update its guidance, first to assign rankings to its priority groups, and then to include real data from vaccine trials, such as

how effective a given vaccine is in older people. In the United States, the NASEM committee is due to issue a final plan in October. Ultimately, the CDC will consider these recommenda­ tions, among others, while developing its own vaccine-allocation plan for the country, expected later this year. That will be the guidance that public-health departments, doctors and pharmacies throughout the United States should follow

“We really are trying to make sure that people of colour will have priority.” when handing out vaccines — assuming that one has been proved safe and people are will­ ing to take it. Trump has been rooting for a vaccine to be ready by November, in time for the US presi­ dential election — but a perception that the vaccine has been rushed could erode trust in it, says Sandra Crouse Quinn, a behavioural scientist at the Center for Health Equity at the University of Maryland in College Park. This could make vaccine-allocation plans less effective.

RESEARCHERS QUESTION RUSSIAN COVID VACCINE TRIAL RESULTS Scientists flag trial findings that seem to be duplicated and call for access to the underlying data. By Alison Abbott

A

group of researchers have expressed concern about repetitive patterns of data in a paper describing early-phase clinical trials of Russia’s coronavirus vaccine — the first jab worldwide to be approved for widespread use. In an open letter to the study authors, who published the trial results1 this month, the researchers highlight values that seem to be duplicated, and warn that the paper presents its results only as box plots, without providing a detailed breakdown of the data on which they are based. “While the research described in this study is potentially significant, the presentation of the data raises several concerns which require access to the original data to fully investigate”, the letter says. It has been signed by almost 40 scientists (see go.nature.com/3kqvsqv). The trials tested two slightly different

viral-vector vaccines — which use genetically engineered adenoviruses to produce corona­ virus proteins in the body — on 76 volunteers. The results indicated that the vaccine pro­ duced a strong immune response, and that side effects were limited to mild, short-term effects, such as irritation at injection sites or headaches, in a few people. In August, the Russian authorities approved the vaccine, called Sputnik V, for widespread use, and have said that it could be available to the general public within months. This fast-track approval caused consternation among researchers, who argued that the decision to roll out the vaccine before larger safety and efficacy trials had been completed was dangerously rushed.

Possible duplications The open letter was posted on a blog run by molecular biologist Enrico Bucci, who heads a science-integrity company called Resis

in Samone, Italy. Bucci says that he noticed irregularities in the paper soon after it was published (D. Y. Logunov et al. Lancet https:// doi.org/gg96hq; 2020). For example, in one figure, in which the authors report their meas­ urements of markers of a type of immune cell in the blood, many members of two groups of nine volunteers tested with different formu­ lations of the vaccine seem to have the same levels. “The odds of this arising by coincidence are extremely small,” Bucci says. “To see such similar data patterns between unrelated measurements is really not likely,” says Konstantin Andreev, who studies viral respiratory infections at Northwestern Uni­ versity at Evanston, Illinois. “These discrep­ ancies are not minor.” Andreev had been independently concerned about aspects of the clinical trial, and signed the open letter shortly after it was made public. “We are not alleging scientific misconduct, but asking for clarification about how these apparently similar data points came about,” says Bucci. “When we read reports that Russia had started to inject the vaccine into people outside clinical trials, we felt we had to speak out immediately.” Late-phase clinical trials of the vaccine, which will involve tens of thou­ sands of people, began on 26 August. The paper’s underlying data should be made available, says epidemiologist Michael Favorov, president of DiaPrep Systems, a diag­ nostics company in Atlanta, Georgia. “We have a lot of questionable data — in terms of its pres­ entation,” he says. “Maybe the data are good — we can’t judge.” He adds that the decision to publish the reports without the underlying data seems unusual. By contrast, when clin­ ical studies involving a coronavirus vaccine that was developed by the pharmaceutical company AstraZeneca and the University of Oxford, UK, were published in the same jour­ nal, they were accompanied by a large amount of supplementary data that other researchers were able to scrutinize (P. M. Folegatti et al. Lancet 396, 467–478; 2020). The Russian paper’s lead author, Denis Logunov at the Gamaleya National Research Centre for Epidemiology and Microbiology in Moscow, did not respond to requests for comment from Nature’s news team. But he told the Russian news outlet Meduza that he did not intend to respond to the open letter. He denied that there were errors in the publica­ tion, and stated that measured antibody levels were “exactly as they were presented” in the figures. He added that he was in contact with The Lancet and “was ready to clarify any issues”. The Lancet declined to comment on its policy for providing data in support of clini­ cal trials that it publishes, but said that it “has invited the authors of the Russian vaccine study to respond to the questions raised in the open letter by Enrico Bucci”, and that it would continue to follow the situation closely.

. 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 585 | 24 September 2020 | 493

disproportionately higher numbers than have other groups in the United States. The panel determined that these groups are vulnera­ ble chiefly for socio-economic reasons tied to systemic racism — for example, they have high-risk jobs and live in high-risk areas — and therefore addressed the request through this lens, without singling out the groups because of their identities. “We really are trying to make sure that peo­ ple of colour, who have been disproportion­ ately impacted, will also have priority — but for the factors that put them at risk, not highlight­ ing just their racial and ethnic make-up,” says Helene Gayle, president and chief executive of the Chicago Community Trust in Illinois and a co-chair of the NASEM committee that drafted the proposal. Faden says the recommendations acknowl­ edge the current focus on racial injustice in the United States. “I was reading to see: does this report speak to the cultural moment in the United States, does it speak to racism and other forms of structural inequality? And it does,” she says. The WHO’s strategic advisory group will continue to update its guidance, first to assign rankings to its priority groups, and then to include real data from vaccine trials, such as

how effective a given vaccine is in older people. In the United States, the NASEM committee is due to issue a final plan in October. Ultimately, the CDC will consider these recommenda­ tions, among others, while developing its own vaccine-allocation plan for the country, expected later this year. That will be the guidance that public-health departments, doctors and pharmacies throughout the United States should follow

“We really are trying to make sure that people of colour will have priority.” when handing out vaccines — assuming that one has been proved safe and people are will­ ing to take it. Trump has been rooting for a vaccine to be ready by November, in time for the US presi­ dential election — but a perception that the vaccine has been rushed could erode trust in it, says Sandra Crouse Quinn, a behavioural scientist at the Center for Health Equity at the University of Maryland in College Park. This could make vaccine-allocation plans less effective.

RESEARCHERS QUESTION RUSSIAN COVID VACCINE TRIAL RESULTS Scientists flag trial findings that seem to be duplicated and call for access to the underlying data. By Alison Abbott

A

group of researchers have expressed concern about repetitive patterns of data in a paper describing early-phase clinical trials of Russia’s coronavirus vaccine — the first jab worldwide to be approved for widespread use. In an open letter to the study authors, who published the trial results1 this month, the researchers highlight values that seem to be duplicated, and warn that the paper presents its results only as box plots, without providing a detailed breakdown of the data on which they are based. “While the research described in this study is potentially significant, the presentation of the data raises several concerns which require access to the original data to fully investigate”, the letter says. It has been signed by almost 40 scientists (see go.nature.com/3kqvsqv). The trials tested two slightly different

viral-vector vaccines — which use genetically engineered adenoviruses to produce corona­ virus proteins in the body — on 76 volunteers. The results indicated that the vaccine pro­ duced a strong immune response, and that side effects were limited to mild, short-term effects, such as irritation at injection sites or headaches, in a few people. In August, the Russian authorities approved the vaccine, called Sputnik V, for widespread use, and have said that it could be available to the general public within months. This fast-track approval caused consternation among researchers, who argued that the decision to roll out the vaccine before larger safety and efficacy trials had been completed was dangerously rushed.

Possible duplications The open letter was posted on a blog run by molecular biologist Enrico Bucci, who heads a science-integrity company called Resis

in Samone, Italy. Bucci says that he noticed irregularities in the paper soon after it was published (D. Y. Logunov et al. Lancet https:// doi.org/gg96hq; 2020). For example, in one figure, in which the authors report their meas­ urements of markers of a type of immune cell in the blood, many members of two groups of nine volunteers tested with different formu­ lations of the vaccine seem to have the same levels. “The odds of this arising by coincidence are extremely small,” Bucci says. “To see such similar data patterns between unrelated measurements is really not likely,” says Konstantin Andreev, who studies viral respiratory infections at Northwestern Uni­ versity at Evanston, Illinois. “These discrep­ ancies are not minor.” Andreev had been independently concerned about aspects of the clinical trial, and signed the open letter shortly after it was made public. “We are not alleging scientific misconduct, but asking for clarification about how these apparently similar data points came about,” says Bucci. “When we read reports that Russia had started to inject the vaccine into people outside clinical trials, we felt we had to speak out immediately.” Late-phase clinical trials of the vaccine, which will involve tens of thou­ sands of people, began on 26 August. The paper’s underlying data should be made available, says epidemiologist Michael Favorov, president of DiaPrep Systems, a diag­ nostics company in Atlanta, Georgia. “We have a lot of questionable data — in terms of its pres­ entation,” he says. “Maybe the data are good — we can’t judge.” He adds that the decision to publish the reports without the underlying data seems unusual. By contrast, when clin­ ical studies involving a coronavirus vaccine that was developed by the pharmaceutical company AstraZeneca and the University of Oxford, UK, were published in the same jour­ nal, they were accompanied by a large amount of supplementary data that other researchers were able to scrutinize (P. M. Folegatti et al. Lancet 396, 467–478; 2020). The Russian paper’s lead author, Denis Logunov at the Gamaleya National Research Centre for Epidemiology and Microbiology in Moscow, did not respond to requests for comment from Nature’s news team. But he told the Russian news outlet Meduza that he did not intend to respond to the open letter. He denied that there were errors in the publica­ tion, and stated that measured antibody levels were “exactly as they were presented” in the figures. He added that he was in contact with The Lancet and “was ready to clarify any issues”. The Lancet declined to comment on its policy for providing data in support of clini­ cal trials that it publishes, but said that it “has invited the authors of the Russian vaccine study to respond to the questions raised in the open letter by Enrico Bucci”, and that it would continue to follow the situation closely.

. 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 585 | 24 September 2020 | 493

CHINA’S COVID VACCINE SHOWS MILITARY’S ROLE IN MEDICAL RESEARCH

The People’s Liberation Army is investing in medical research as part of a modernization strategy. By Dyani Lewis

T

he largest armed force in the world, China’s People’s Liberation Army (PLA), is not known for its cutting-edge medical research. But since 2015, it has ramped up recruitment of scientists and investment in the field as part of its strategy to modernize its military. Now, the coronavirus pandemic is showcasing the PLA’s growing expertise in medical research, indicated, among other things, by its major role in developing the coronavirus vaccine that was the first in the world to be approved for restricted use. “China is definitely trying to leverage the crisis from a PR perspective,” says Abigail Coplin, who studies China’s biotechnology industry at Vassar College in Poughkeepsie, New York. Medical researcher Major General Chen Wei at the Beijing Institute of Biotechnology — part of the Academy of Military Medical Sciences — led the team that developed the vaccine, which included collaborators from government agencies, universities and the Tianjin-based pharmaceutical company CanSino Biologics. In July, the team became one of the first in the world to publish results in a peer-reviewed journal that showed a coronavirus vaccine to be safe and capable of eliciting an immune response (F.-C. Zhu et al. Lancet 396, 479–488; 2020). By then, the Chinese government had already approved the vaccine, called Ad5-nCoV, for limited use in military personnel, before large-scale testing to prove its efficacy. Chen and members of her team were among the first — of thousands in the military so far — to receive the vaccine. She and the Beijing Institute of Biotechnology did not respond to Nature’s request for comment about the vaccine work. Should the vaccine win approval for more widespread use before efforts backed by other countries, especially the United States, “it will be a pretty big propaganda victory” for Beijing, says Adam Ni, a China analyst at the Australian National University in Canberra. As well as its contributions to the development of a coronavirus vaccine, the PLA has also taken a high-profile role in controlling the pandemic in China, has sent assistance with pandemic response to a host of countries and has used the vaccine to forge new links abroad. Other militaries, including that of the United

Science a priority In 2015, Chinese President Xi Jinping announced reforms that made science and innovation key elements of modernizing its armed forces, says Elsa Kania, who analyses Chinese military strategy at the Center for a New American Security in Washington DC. The PLA established branches for electronic, cyber and space warfare alongside its more conventional army, navy and air force. And in

The People’s Liberation Army has been central to China’s ability to control the coronavirus.

. 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 ©

494 | Nature | Vol 585 | 24 September 2020

States, work on vaccines and conduct medical research. But the sheer size of the PLA and the speed at which reforms are taking place make its scientific transformation noteworthy and, for some, a cause for concern, especially given the growing political tensions between the United States and China, Ni says. In the past few months, US security officials have revealed that China has tried to spy on and steal information from US pharmaceutical companies and university research groups working on coronavirus vaccines. Scientists have also raised concerns about the ethics of approving military use of a vaccine that is still being trialled.

2016, a Science and Technology Commission, which decides what research is funded, became one of 15 newly formed military ‘sections’. “It’s gone from a fairly backward military in the 70s and 80s — large but certainly not professional and not technologically advanced — to a much more formidable military,” says Ni. The reforms also brought the Academy of Military Medical Sciences — which helped develop the Ad5-nCoV vaccine — under the umbrella of the Academy of Military Sciences, the PLA’s main military strategy body, which oversees nine other research institutions. Before the reforms, the PLA recruited scientists either internally or from military universities, partly because civilian scientists didn’t find research positions in the military attractive, says Ni. Working conditions were less flexible than in civilian institutions, he says. But since 2018, the PLA has been recruiting more civilian-trained scientists,making research positions more appealing — and has upped its recruitment of medical scientists. The Academy of Military Medical Sciences has recruited 213 civilians for scientific research positions since 2018, making it the second-highest recruiter of scientific talent among the Academy of Military Science’s 10 research institutions, according to an analysis posted online by Kai Lin Tay at the International Institute for Strategic Studies in London. The military is also increasing its ties to civilian universities in China, as part of a policy known as military–civil fusion, which the Chinese government also announced in 2015. The strategy highlights biology as a priority research area. The PLA has also been bolstering its scientific expertise by sending researchers abroad.

CHINE NOUVELLE/SIPA/SHUTTERSTOCK

News in focus

L. BRIAN STAUFFER/UNIV. ILLINOIS

And partnerships between the PLA and the medical-science companies have accelerated since the pandemic, according to Tay’s report. As well as collaborating with CanSino Biologics to develop the Ad5-nCoV vaccine, the PLA has worked with Beijing Chieftain, which makes medical equipment, since March. The pandemic has provided China with an opportunity to highlight its military’s scientific achievements on the domestic and international stage. PLA epidemiologists and medical workers have had an important role in treating the sick, monitoring the outbreak and overseeing distribution of medical supplies in Wuhan. The military has also assisted with the pandemic responses in Pakistan, Iran, Iraq, Lebanon, Vietnam, Laos, Myanmar, Cambodia and Italy, by deploying units and supplies. The PLA-developed vaccine could give China additional geopolitical influence, with favoured countries being given priority access to the vaccine, says Ian McCaslin, a China military analyst affiliated with Air University’s China Aerospace Studies Institute in Washington DC. CanSino Biologics already has agreements to conduct phase III trials in Russia, Mexico, Saudi Arabia and Pakistan. Because analysts who study the Chinese military have not previously focused on the medical research conducted by the PLA, its true impact remains unknown, says Kania. “We’re still trying to get a handle on how to understand the scope and scale of their activities, as well as the quality and competitiveness of their research,” she says. Some research is similar to that conducted by other military forces. The US military, for instance, is developing its own coronavirus vaccine and conducts basic research into trauma and infectious diseases. It’s important to bear in mind that the PLA’s scientific efforts represent only “a small minority of work being done in biotechnology, both within China and globally”, adds Coplin. Still, some China experts and foreign governments are concerned about the participation of PLA researchers in medical research. In July, the US Justice Department indicted two Chinese nationals for spying on three US-based entities involved in medical research to fight the coronavirus. “Tech transfer is clearly a policy and priority of the Chinese government at the highest levels and has involved fairly egregious instances of hacking, for purposes of data theft,” says Kania. Scientists are also concerned about China’s lack of safeguards to ensure that research on people is conducted ethically, says Kania. It is unclear whether military personnel were given a choice about whether to receive the PLA-backed coronavirus vaccine, she says. These are legitimate concerns, says Coplin, but she cautions the United States against using them as a reason to stymie otherwise productive collaborations with China.

Q&A

‘We didn’t model that people would go to a party if they tested positive’ As universities around the globe struggle with how to keep their doors open amid the COVID-19 pandemic, some have developed their own rapid coronavirus diagnostics to test students multiple times per week. The University of Illinois at Urbana–Champaign (UIUC) has a mass testing programme that has been touted as a model system. But the institute saw a spike in infections last month — on 31 August, it reported a 291% increase over the daily total a week earlier. Martin Burke (pictured), a chemist at UIUC who helped to develop the university’s RNAbased saliva test, spoke to Nature about the lessons learned. How did you get involved in developing this test? At the end of April, UIUC provost Andreas Cangellaris called me and asked if I would lead a team to build up and deploy scalable testing as part of our campus’s effort to reopen as safely as possible. We decided to take a comprehensive approach: we got a lot of data scientists to help us figure out how to model the epidemic on campus. We realized that we would need to test everybody on campus twice a week. Over the course of about six weeks, we discovered that we could use saliva samples and cut out almost all the bottlenecks associated with standard COVID-19 tests. We’re now doing more than 10,000 — sometimes 15,000 — tests per day.

cart down to the lab, where it gets straight into the water bath. The test itself takes 90 minutes; the results come back in less than 24 hours. The data go straight to an app on your phone. To get into any building, you have to scan your data to show that you are compliant with testing twice a week. UIUC reported a spike in campus infections last month. What happened? When we put the whole programme in place, we did a bunch of modelling to try to understand how student socialization was going to integrate with the fast, recurrent testing. We modelled that they were going to go to parties and that they probably weren’t going to wear masks, and it would lead to some level of transmission. What we didn’t model for is that people would choose to go to a party if they knew that they were positive. The overwhelming majority of our students have done a great job, but unfortunately, a small number of students chose to make very bad decisions. Does this call into question the idea that mass testing can keep campuses safe? The answer is definitely no. We caught this early, we made changes, and now we’re watching our numbers fall. [On 8 September, UIUC reported a total of 81 new COVID-19 infections in one day, a 65% decrease since the spike.]

What’s innovative about the test compared with standard RNA-based tests? Three things. We use saliva instead of a nasal swab. We skip RNA isolation, which saves a very expensive and slow step. And as soon as the sample tube comes into the lab, it gets heated in a water bath at 95 °C for 30 minutes, which inactivates the virus and protects the workers, but also breaks the virus open and exposes its RNA.

What protocol changes did UIUC make? People who made those bad choices have been suspended. We’ve started testing more frequently [in the fraternity houses and dormitories] where there were problems. Because some of the students were avoiding phone calls from public-health authorities, we built our own internal team, whose goal is to get everyone [who tests positive] safely isolated within 30 minutes.

Walk us through what happens when people take the test on campus. You swipe your ID card, and walk into a big open-air tent over to a square that’s 6 feet [2 metres] apart from all the other squares, and you just dribble into a tube. You put the tube in a rack, and once the rack is filled, [workers] seal it up and drive it on a golf

What lessons have you learnt? It’s not just a matter of getting the test done fast; it’s a matter of acting on the results as fast as possible. Interview by Giorgia Guglielmi This interview has been edited for length and clarity.

. 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 585 | 24 September 2020 | 495

L. BRIAN STAUFFER/UNIV. ILLINOIS

And partnerships between the PLA and the medical-science companies have accelerated since the pandemic, according to Tay’s report. As well as collaborating with CanSino Biologics to develop the Ad5-nCoV vaccine, the PLA has worked with Beijing Chieftain, which makes medical equipment, since March. The pandemic has provided China with an opportunity to highlight its military’s scientific achievements on the domestic and international stage. PLA epidemiologists and medical workers have had an important role in treating the sick, monitoring the outbreak and overseeing distribution of medical supplies in Wuhan. The military has also assisted with the pandemic responses in Pakistan, Iran, Iraq, Lebanon, Vietnam, Laos, Myanmar, Cambodia and Italy, by deploying units and supplies. The PLA-developed vaccine could give China additional geopolitical influence, with favoured countries being given priority access to the vaccine, says Ian McCaslin, a China military analyst affiliated with Air University’s China Aerospace Studies Institute in Washington DC. CanSino Biologics already has agreements to conduct phase III trials in Russia, Mexico, Saudi Arabia and Pakistan. Because analysts who study the Chinese military have not previously focused on the medical research conducted by the PLA, its true impact remains unknown, says Kania. “We’re still trying to get a handle on how to understand the scope and scale of their activities, as well as the quality and competitiveness of their research,” she says. Some research is similar to that conducted by other military forces. The US military, for instance, is developing its own coronavirus vaccine and conducts basic research into trauma and infectious diseases. It’s important to bear in mind that the PLA’s scientific efforts represent only “a small minority of work being done in biotechnology, both within China and globally”, adds Coplin. Still, some China experts and foreign governments are concerned about the participation of PLA researchers in medical research. In July, the US Justice Department indicted two Chinese nationals for spying on three US-based entities involved in medical research to fight the coronavirus. “Tech transfer is clearly a policy and priority of the Chinese government at the highest levels and has involved fairly egregious instances of hacking, for purposes of data theft,” says Kania. Scientists are also concerned about China’s lack of safeguards to ensure that research on people is conducted ethically, says Kania. It is unclear whether military personnel were given a choice about whether to receive the PLA-backed coronavirus vaccine, she says. These are legitimate concerns, says Coplin, but she cautions the United States against using them as a reason to stymie otherwise productive collaborations with China.

Q&A

‘We didn’t model that people would go to a party if they tested positive’ As universities around the globe struggle with how to keep their doors open amid the COVID-19 pandemic, some have developed their own rapid coronavirus diagnostics to test students multiple times per week. The University of Illinois at Urbana–Champaign (UIUC) has a mass testing programme that has been touted as a model system. But the institute saw a spike in infections last month — on 31 August, it reported a 291% increase over the daily total a week earlier. Martin Burke (pictured), a chemist at UIUC who helped to develop the university’s RNAbased saliva test, spoke to Nature about the lessons learned. How did you get involved in developing this test? At the end of April, UIUC provost Andreas Cangellaris called me and asked if I would lead a team to build up and deploy scalable testing as part of our campus’s effort to reopen as safely as possible. We decided to take a comprehensive approach: we got a lot of data scientists to help us figure out how to model the epidemic on campus. We realized that we would need to test everybody on campus twice a week. Over the course of about six weeks, we discovered that we could use saliva samples and cut out almost all the bottlenecks associated with standard COVID-19 tests. We’re now doing more than 10,000 — sometimes 15,000 — tests per day.

cart down to the lab, where it gets straight into the water bath. The test itself takes 90 minutes; the results come back in less than 24 hours. The data go straight to an app on your phone. To get into any building, you have to scan your data to show that you are compliant with testing twice a week. UIUC reported a spike in campus infections last month. What happened? When we put the whole programme in place, we did a bunch of modelling to try to understand how student socialization was going to integrate with the fast, recurrent testing. We modelled that they were going to go to parties and that they probably weren’t going to wear masks, and it would lead to some level of transmission. What we didn’t model for is that people would choose to go to a party if they knew that they were positive. The overwhelming majority of our students have done a great job, but unfortunately, a small number of students chose to make very bad decisions. Does this call into question the idea that mass testing can keep campuses safe? The answer is definitely no. We caught this early, we made changes, and now we’re watching our numbers fall. [On 8 September, UIUC reported a total of 81 new COVID-19 infections in one day, a 65% decrease since the spike.]

What’s innovative about the test compared with standard RNA-based tests? Three things. We use saliva instead of a nasal swab. We skip RNA isolation, which saves a very expensive and slow step. And as soon as the sample tube comes into the lab, it gets heated in a water bath at 95 °C for 30 minutes, which inactivates the virus and protects the workers, but also breaks the virus open and exposes its RNA.

What protocol changes did UIUC make? People who made those bad choices have been suspended. We’ve started testing more frequently [in the fraternity houses and dormitories] where there were problems. Because some of the students were avoiding phone calls from public-health authorities, we built our own internal team, whose goal is to get everyone [who tests positive] safely isolated within 30 minutes.

Walk us through what happens when people take the test on campus. You swipe your ID card, and walk into a big open-air tent over to a square that’s 6 feet [2 metres] apart from all the other squares, and you just dribble into a tube. You put the tube in a rack, and once the rack is filled, [workers] seal it up and drive it on a golf

What lessons have you learnt? It’s not just a matter of getting the test done fast; it’s a matter of acting on the results as fast as possible. Interview by Giorgia Guglielmi This interview has been edited for length and clarity.

. 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 585 | 24 September 2020 | 495

ANSHUMAN POYREKAR/HINDUSTAN TIMES VIA GETTY

Feature

Health-care workers test a resident of Mumbai, India, for coronavirus infection using a rapid antigen assay.

FAST CORONAVIRUS TESTS ARE COMING

Rapid antigen tests are designed to tell in a few minutes whether someone is infectious. Will they be game changers? By Giorgia Guglielmi

T

he United States leads the world in COVID-19 deaths but lags behind many countries — both large and small — in testing capacity. That could soon change. At the end of August, the US Food and Drug Administration (FDA) granted emergency-use approval to a new credit-card-sized testing device for the coronavirus that costs US$5, gives results in 15 minutes and doesn’t require a laboratory or a machine for processing. The United States is spending $760 million on 150 million of

. 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 ©

496 | Nature | Vol 585 | 24 September 2020

these tests from health-care company Abbott Laboratories, headquartered in Abbott Park, Illinois, which plans to ramp up production to 50 million per month in October. The tests detect specific proteins — known as antigens — on the surface of the virus, and can identify people who are at the peak of infection, when virus levels in the body are likely to be high. Proponents argue that this could be a game changer. Antigen tests could help to keep the pandemic at bay, because they can be rolled out in vast numbers and can spot those who are at greatest risk of spreading the disease. These

tests are also a key element in the testing strategies of other countries, such as India and Italy. Antigen assays are much faster and cheaper than the gold-standard tests that detect viral RNA using a technique called the polymerase chain reaction (PCR). But antigen tests aren’t as sensitive as the PCR versions, which can pick up minuscule amounts of the SARS-CoV-2 virus that causes COVID-19. This difference raises some concerns among specialists, who worry that antigen tests will miss infectious people and result in outbreaks in countries that have largely controlled coronavirus transmission. Others view the lower sensitivity as an attribute, because some people who receive positive PCR test results are infected, but are no longer able to spread the virus to others. So antigen tests could shift the focus to identifying the most infectious people. At present, antigen tests are administered by trained professionals, but some companies are developing versions that are simple enough to be used at home — similar to pregnancy tests. “Making the tests faster, cheaper, easier is definitely the goal — and I think the antigen test is the way to get there,” says Martin Burke, a chemist at the University of Illinois at Urbana-Champaign, who is co-developing rapid tests, including antigen-based assays.

SOURCE: REF. 2

What tests are there and how do they work? Tests for COVID-19 fall into two categories: diagnostic tests such as PCR and antigen assays, which detect parts of the SARS-CoV-2 virus, and antibody tests that sense molecules that people produce when they have been infected by the virus. Antibodies can take several days to develop after an infection and often stay in the blood for weeks after recovery, so antibody tests have limited use in diagnosis (see ‘Catching COVID-19’). The high-sensitivity PCR tests are almost 100% accurate in spotting infected people, when they are administered properly. But such tests generally require trained personnel, specific reagents and expensive machines that take hours to provide results. Countries such as South Korea and New Zealand have succeeded in boosting PCR-based testing, but scaling up these tests has proved difficult elsewhere. The United States, for example, has seen a slow and poorly coordinated response to outbreaks, faulty tests from the Centers for Disease Control and Prevention (CDC) and problems with the supply chain. All of this has hindered efforts to collect and process samples for PCR, pushing waiting times to days or even weeks. These delays, along with a lack of tests, have contributed to the rampant spread of COVID-19 across the country, which by 18 September had seen almost 200,000 deaths from the disease. A typical antigen test starts with a healthcare professional swabbing the back of a person’s nose or throat — although companies are developing kits that use saliva samples, which are easier and safer to collect than a swab. The sample is then mixed with a solution that breaks the virus open and frees specific viral proteins. The mix is added to a paper strip that contains an antibody tailored to bind to these proteins, if they’re present in the solution. A positive test result can be detected either as a fluorescent glow or as a dark band on the paper strip. Antigen tests give results in less than 30 minutes, don’t have to be processed in a lab and are cheap to produce. Yet that speed comes with a cost in sensitivity. Whereas a typical PCR test can detect a single molecule of RNA in a microlitre of solution, antigen tests need a sample to contain thousands — probably tens of thousands — of virus particles per microlitre to produce a positive result1. So, if a person has low amounts of virus in their body, the test might give a false-negative result. When used on people who were positive for SARS-CoV-2 in a standard PCR test, Abbott’s antigen assay correctly spotted the virus in 95–100% of cases if the samples were collected within a week of the onset of symptoms. But

that proportion dropped to 75% if samples were taken more than a week after people first showed symptoms. The sensitivity — or the rate of detecting infections correctly — of the other antigen tests used in the United States is between 84% and 98% if a person is tested in the week after showing symptoms. Companies and academic research labs are also rolling out other tests that are faster, cheaper and more user-friendly than standard PCR assays, although they are not being produced on the same scale as antigen tests. Some of these other tests use the gene-editing tool CRISPR to zero in on genetic snippets of the coronavirus. Others are quicker variants of the PCR test that use different reagents, meaning they’re not limited by the same supply-chain problems. Saliva-based PCR tests, for example, are being used as screening tools in universities and for professional basketball teams.

Which tests tell whether someone is infectious? Although the PCR method can test whether someone is infectious, it also detects people who have the virus but are not likely to spread it. Antigen-based testing, by contrast, could help to rapidly identify people who have high levels of virus — those who are most likely to be infectious to others — and isolate them from the community, says Marion Koopmans, a virologist at the Erasmus University Medical Centre in Rotterdam, the Netherlands. “The question is, what is the safe limit? Because the moment you get that wrong, the whole idea implodes,” she says. It’s still unclear what viral load is the

CATCHING COVID-19

Different types of COVID-19 test can detect the presence of the SARS-CoV-2 virus or the body’s response to infection. The probability of a positive result varies with each test before and after symptoms appear. PCR-based tests detect small amounts of viral genetic material, so a test can be positive long after a person stops being infectious. Rapid antigen tests detect the presence of viral proteins and can return positive results when a person is most infectious. Antibody tests detect the body’s immune response to the virus and are not effective at the earliest phase of infection. Exposure to virus

Symptom onset IgG antibody

Probability of detection

“This is by no means the perfect solution, it’s just the fastest thing we could get going now,” he says.

IgM antibody

–2

–1

0

1

2

3

4

5

Time from symptom onset (weeks)

6

threshold below which a person is no longer contagious, says Koopmans, who is working with the World Health Organization (WHO) to determine a standard to validate rapid tests. “It would be very worrying if everyone does that on their own, using different criteria,” she says. Viral load peaks early in SARS-CoV-2 infections and then gradually declines, with tiny amounts of virus RNA staying in someone’s nose or throat for weeks or possibly months2. And although there are not enough data to equate different viral levels with how infectious people are, there is evidence that individuals are unlikely to spread the virus about eight to ten days after showing symptoms3. “If you’re at risk of transmitting the virus to somebody else, you’re going to have plenty of viral particles — those would certainly show up in antigen tests,” says Michael Mina, an infectious-disease immunologist at the Harvard T. H. Chan School of Public Health in Boston, Massachusetts, who has been a vocal proponent of antigen tests. There are challenges at the start of the infection, when people have low levels of the virus. The answer, says Mina, is frequent testing — done multiple times per week. This could quickly identify infected people, even if the assays are less sensitive than a PCR-based test, because the amount of virus in their noses and throats rises within hours, he says. Mina and his colleagues have used statistical models to assess this strategy. In a preprint updated on 8 September, they suggest that testing people twice a week with a relatively insensitive test could be more effective at curbing the spread of SARS-CoV-2 than are more-accurate tests done once every two weeks1. Another study that modelled different scenarios for safely reopening university campuses reported similar findings4. To slow outbreaks, the focus should be on identifying those who are at risk of spreading SARS-CoV-2 to other people, rather than on spotting anyone who is infected with it, some experts say. When used as a screening tool to frequently assess as many people as possible, rapid antigen tests could be “a game changer”, says Rebecca Lee Smith, an epidemiologist at the University of Illinois.

How do countries plan to use antigen tests? At the beginning of April, as coronavirus outbreaks raged across the world, India had tested only about 150,000 people — one of the lowest testing rates per capita worldwide. On 21 August, the country conducted more than one million coronavirus tests in a single day. It reached that milestone after Indian authorities began using antigen assays to boost testing capacity. Delhi was the first Indian state to begin using rapid antigen tests, in June. By mid-July, the

. 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 585 | 24 September 2020 | 497

2 mm

3 mm 3 mm

Feature to a steep increase in the total number of cases, says Andrea Crisanti, a microbiologist at the University of Padua. Some researchers worry that there won’t be enough antigen tests available to greatly expand their use. “Rapid tests right now are for the happy few,” Koopmans says. “If we want to take these assays responsibly forward, we should talk about whether they can be produced to levels that would make them globally available.”

A technician in a mobile unit conducts rapid antigen tests for COVID-19 in New Delhi.

number of cases there had decreased and the daily death counts had plateaued, suggesting that the tests might have played some part in controlling the spread of the virus. Epidemiologist K. Srinath Reddy, president of the Public Health Foundation of India, a nonprofit organization in New Delhi, says that the Delhi example is interesting, but not clear-cut: he notes that the government started to lift lockdown restrictions in August, which led to a surge in infections. “Rapid antigen tests have picked up the increased number of cases, but whether they have been successful in limiting the spread of COVID, we’ll only know in the next couple of months,” Reddy says. So far, India has approved the use of three antigen tests for screening large numbers of people, whether or not they have symptoms. One of the kits was evaluated by the Indian Council of Medical Research (ICMR) and the All India Institute of Medical Sciences, which found that the test detected infections between 51% and 84% of the time. Guidance from the ICMR says that people who have a negative result from an antigen test should also get a PCR test if they show symptoms, to rule out the possibility that the rapid test missed an infection. The WHO and the US CDC have also advised getting a PCR test if people showing symptoms test negative with a rapid antigen test. The US FDA has so far granted emergency use authorization for four antigen tests, each of which has a higher sensitivity than those used in India. The 150 million tests bought from Abbott will be used in schools and “other special needs populations”, according to the Department of Health and Human Services. The FDA, however, has authorized antigen-based tests only for people who have had symptoms for 12 days or fewer. Tests must be prescribed by a physician and administered by a health-care professional.

“Testing should become a part of life: in the morning you take your cereals, your vitamins, and you quickly check your status.” considered to have a high risk of infection. Negative results do not have to be confirmed with a PCR test. The Italian health minister, Roberto Speranza, has announced plans to use antigen tests to screen passengers at all of the country’s airports, and a group of experts has urged the Italian government to use the rapid tests in schools and universities. But others don’t think rapid antigen tests are a good idea. When trying to contain small outbreaks, such as those happening in Italy, public-health authorities should use assays that are highly accurate, because missing even just one positive individual could lead

. 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 ©

498 | Nature | Vol 585 | 24 September 2020

Other countries are also considering the use of rapid antigen tests to meet targets. In July, the Philippine Society for Microbiology and Infectious Diseases issued temporary guidelines for clinicians and health-care workers, saying that antigen tests could be used as an alternative to PCR for diagnosing a coronavirus infection during the first week in people with symptoms. But it also recommend15 ss that all negative results should be confirmed with a PCR-based assay, says Edsel Salvaña, an infectious-diseases expert at the University of the Philippines Manila, who is advising Philippine officials on rapid testing. Antigen-based tests are being used in some of Italy’s major airports to screen people who arrive from four Mediterranean countries

Several experts have promoted the idea of developing an antigen test that is cheap and simple enough to use at home, without a health-care worker administering it. Burke says what’s needed is something as easy as a pregnancy test. “You just spit into a tube, put a piece of paper in it and you get the result within minutes,” Burke says. “Testing should become a part of life: in the morning you take your cereals, your vitamins, and you quickly check your status,” he says. A few companies are developing simple paper-strip antigen tests. But drug regulators have not yet approved them for emergency use. “We don’t have a lot of real-life experience with these tests, and a lot of the validations have only been done in the laboratory,” Salvaña says. Beyond concerns about costs and availability, researchers worry that, with an over-thecounter test, people who get positive results might not follow up with public-health authorities, so their contacts won’t be traced. Another risk would be people “gaming the system”, Smith says — for example, getting someone else to take their test — so they can be sure of a negative result and avoid quarantine. Without incentives such as freely available tests and a living salary for those who have to isolate, testing and self-isolation could become a luxury reserved for wealthier people, others have argued. Another concern is that people will get a false sense of security from tests that have only limited accuracy. “There’s a big risk that the moment these tests become widely available, people will just use them and say, ‘It’s negative, so I’m clear,’” Koopmans says. Even when testing negative, people should continue to wash their hands, wear masks and avoid gathering in big groups, she says. Testing, she adds, “cannot replace the basic control measures that need to be in place to keep this virus controlled”. Giorgia Guglielmi is a science journalist in Basel, Switzerland. 1. Larremore, D. B. et al. Preprint at medRxiv https://doi. org/10.1101/2020.06.22.20136309 (2020). 2. Sethuraman, N., Jeremiah, S. S. & Ryo, A. J. Am. Med. Assoc. 323, 2249–2251 (2020). 3. He, X. et al. Nature Med. 26, 672–675 (2020). 4. Paltiel, A. D., Zheng, A., Walensky, R. P. JAMA Netw. Open 3, e2016818 (2020).

MAYANK MAKHIJA/NURPHOTO VIA GETTY

Could antigen assays be used at home like pregnancy tests?

Science in culture

MARCOS DEL MAZO/LIGHTROCKET VIA GETTY

Books & arts

2020: Protesters elide vaccination, 5G mobile-phone technology and face masks in Spain, where COVID-19 rates are soaring.

Vaccines — lessons from three centuries of protest Immunization has always been a proxy for wider fears about social control, a history reminds us. By Julie Leask

T

he need to control outbreaks and pandemics has long created tensions between liberty and interdependence, similar to those playing out worldwide today. Anti-vaxxers is a book that reminds us of the historical precedents to the odd alliances — anti-vaccine, anti-mask, anti-5G, for instance — that are getting in the way of public health right now. Vaccination has always been a lightning rod for storms brewing over other problems, as physiologist and science writer Jonathan Berman shows. The people who protested against mandatory smallpox vaccination in

nineteenth-century England had previously led opposition to the 1834 Poor Law Amendment Act, which proposed that unemployed people must labour in workhouses for food, Anti-vaxxers: How to Challenge a Misinformed Movement Jonathan M. Berman MIT Press (2020)

often under conditions of exploitation, child labour and family separation. The protesters saw mandatory vaccination as a similar assault on poor people’s autonomy. After examining the rise of such opposition in England, Berman turns to the US experience in the twentieth and early twenty-first centuries. So where did vaccination — and opposition to it — all begin? Variolation, deliberate infection with matter from smallpox pustules or scabs to bring about natural immunity, had been described in Asia and Africa since at least the sixteenth century. Christian minister Cotton Mather championed the idea in Boston,

. 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 585 | 24 September 2020 | 499

TS/KEYSTONE USA/SHUTTERSTOCK

Books & arts

1918: A demonstration against mandatory smallpox vaccination in Toronto, Canada.

Massachusetts, in the early eighteenth century, having learnt it from an African man, Onesimus, enslaved in his household. Although the practice cut death rates, Mather was ridiculed. Vaccination, popularized by English physician Edward Jenner from the end of the eighteenth century, sought the same end. But instead of using matter from smallpox pustules, physicians inoculated people with cowpox, a cattle virus that causes milder disease in humans. The technique was successful, but opponents levelled that it was a “foreign assault on traditional order”.

Trust and suspicion The parallels with contemporary vaccine safety scares are clear. After the uptake of measles, mumps and rubella (MMR) vaccination, for example, declined in the United Kingdom, measles outbreaks rose, peaking in 2012, with 2,032 cases in England and Wales. Even with the world hungering for a vaccine against COVID-19, 26% of French adults reported in March that they would not use one if it became available (The COCONEL Group. Lancet Infect. Dis. 20, 769– 770; 2020). In the United States two months later, 14% of adults said the same (P. L. Reiter et al. Vaccine https://doi.org/d8wr; 2020). Berman’s case studies should satisfy those wanting to debunk anti-vaccine claims online

Nests of belief Anti-vaxxers joins a shelf of books published over the past decade that try to make sense of the modern anti-vaccination movement and connect it to the historical, social and political contexts in which it has found expression. Volumes include journalist Seth Mnookin’s 2011 The Panic Virus, and offerings by paediatricians David Isaacs (Defeating the Ministers of Death, 2019), Peter Hotez (the forthcoming Vaccines Did Not Cause Rachel’s Autism) and Paul Offit (Deadly Choices, 2010). Anthropologist Heidi Larson’s Stuck joined this collection earlier

. 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 ©

500 | Nature | Vol 585 | 24 September 2020

or at a family gathering. One is a supposed ‘scandal’ at the US Centers for Disease Control and Prevention (CDC) promoted in the 2016 film Vaxxed. In 2014, biologist Brian Hooker published a reanalysis of data from a 2004 CDC study, alleging that the agency had excluded a notional finding that African American boys given the MMR vaccine before age 36 months had increased chances of developing autism spectrum disorder. Hooker offered secret recordings of conversations with CDC scientist William Thompson, a co-author of the study, in support of his contention. But Berman shows there was no reliable link; the claim resulted from serious methodological failings in Hooker’s analysis, which was retracted.

this year, and a personal account by cultural analyst and essayist Eula Biss, On Immunity (2014), stands out. Such social studies acknowledge that an informed refutation of the latest argument against vaccination has never been enough to convince opponents. Berman’s historical approach also concludes that the root causes of vaccine rejection must be appreciated and addressed. Although the history of vaccination recounts advances in modern science, it is also part of the wider story of society wrestling with the promises and perils of technology. It is a story of parents coming to terms with the death or disability of a child (almost always unrelated to vaccinations), of the pressure to parent this way or that, and of belonging. And it’s a story of activist groups that don’t so much deny science as carefully select straws of information and misinformation to build their nests of belief. What are the solutions to this increasingly globalized phenomenon? (There have been instances of vaccine resistance from Nigeria to Pakistan, not just in Manhattan.) Many books urge scientists to communicate more effectively, or governments to fight back more actively against vaccination’s opponents. Anti-vaxxers refreshingly goes deeper, taking note of a growing body of social and behavioural research. In Perth,

Australia, a community-based project aligned pro-vaccination messages to shared values; and in the United States, trained community advocates in Washington state promote vaccination in their peer networks. Berman also draws together personal narratives from parents. Ingvar Ingvarsson, for example, is a father who chose not to vaccinate his children. Then his experience as a nurse, caring for older people dealing with the effects of measles and polio, triggered a re-evaluation. Eventually, his children received their vaccines.

Position of privilege The role of money and privilege deserves more attention. In the United States in 2018, just 73.2% of children aged 24 months from families without health insurance had received at least one of the recommended 2 doses of MMR vaccine. The figure was 93.7% in families that had private insurance (H. A. Hill et al. Morbid. Mortal. Wkly Rep. 68, 913–918; 2019). Berman distinguishes between two groups of parents whose children are not fully vaccinated: those who reject vaccination, and those who lack access to health care. There should be more emphasis on the have-nots, in my view. Instead, his focus is on the refusers, arguing that because some cannot access care, those who can should be vaccinated. This is a common blind spot in explanations of low take-up. Poverty, and lack of access to social resources and primary care, greatly affect uptake, as do housing insecurity, gender inequity and racism. The largest measles outbreaks in 2019 were in countries without sufficient primary care, such as Madagascar, or where conflict had displaced people and disrupted their access to vaccines, such as Yemen. Some of the most effective interventions include ensuring that supply chains are reliable, making services highly convenient and simply reminding people that they need to be vaccinated. The current pandemic reminds us that governments cannot ignore poverty and social exclusion if they are to prevent and manage this virus, others unvanquished and those yet to come. By taking the story of vaccine opposition back to its earliest examples, Anti-vaxxers cautions against simplistic solutions. In tracing the movement across three centuries, Berman underlines that is unlikely to be ended by keyboard warriors or the repetition of even the best scientific evidence. Julie Leask chairs the World Health Organization Measuring Behavioural and Social Drivers of Vaccination Working Group. She is a professor in the Susan Wakil School of Nursing and Midwifery at the University of Sydney in Australia, and visiting fellow at the National Centre for Immunisation Research and Surveillance in Sydney. e-mail: [email protected]

Books in brief Net Zero Dieter Helm William Collins (2020) Climate-change economist Dieter Helm was frustrated by a widely repeated claim from the UK Committee on Climate Change: “By reducing emissions produced in the UK to zero, we also end our contribution to rising global temperatures.” Not so, he objects: consumers also import goods and services from countries with high emissions, notably China. As Helm bluntly argues in international detail, reaching ‘net zero’ emissions will require unpopular unilateral changes in individual lifestyles and national infrastructures.

The Human Cosmos Jo Marchant Dutton (2020) Galileo composed horoscopes for his illegitimate daughters, notes science journalist Jo Marchant in her multifaceted meditation on humanity’s relationship with the cosmos. From possibly celestial Palaeolithic cave art at Lascaux in France to awestruck astronauts in space, she considers how patterns in the sky have governed lives on Earth, “shaping ideas about time and place; power and truth; life and death”. Although science is right to debunk astrology, she argues, the significance of the heavens has been eclipsed by modern astronomy.

The Inside Out of Flies Erica McAlister Natural History Museum (2020) “Flies are not filthy … they are always cleaning themselves,” notes entomologist Erica McAlister’s caption for a photo of a fly maintaining its antennae — one of many eye-popping images in her erudite, irresistible natural history of the insects. She agrees with naturalist Pliny, who wrote two millennia ago that insects display nature’s “exhaustless ingenuity”. Consider Ephydra hians, which “scuba-dives” in alkaline lakes — using hydrophobic hairs that trap an air bubble like an external lung — to lay its eggs on the lake bottom.

Outside the Box Marc Levinson Princeton Univ. Press (2020) This history of globalization evokes economist Marc Levinson’s 2006 book The Box, about container ships. These were key to the ‘third globalization’, starting in the 1980s: products were manufactured in places with low wages, then shipped to the advanced economies where they had been designed. In today’s ‘fourth globalization’, research, engineering and design are moving, and manufacturing can be done anywhere. Much of the process involves ideas, such as software, rather than ‘stuff’ in a box. But how can it be politically regulated?

The Polymath Peter Burke Yale Univ. Press (2020) From the mid-nineteenth century, science has abounded in specialists, yet polymaths such as Alan Turing and Linus Pauling have remained crucial. In a mind-stretching history, Peter Burke describes “500 western polymaths” from the half-millennium since Leonardo da Vinci. He discusses their curiosity, concentration, memory, speed, imagination, restlessness, hard work and horror of wasting time. But he overlooks specialists with polymathic tendencies, such as Albert Einstein, Florence Nightingale and Ronald Ross. Andrew Robinson

. 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 585 | 24 September 2020 | 501

Australia, a community-based project aligned pro-vaccination messages to shared values; and in the United States, trained community advocates in Washington state promote vaccination in their peer networks. Berman also draws together personal narratives from parents. Ingvar Ingvarsson, for example, is a father who chose not to vaccinate his children. Then his experience as a nurse, caring for older people dealing with the effects of measles and polio, triggered a re-evaluation. Eventually, his children received their vaccines.

Position of privilege The role of money and privilege deserves more attention. In the United States in 2018, just 73.2% of children aged 24 months from families without health insurance had received at least one of the recommended 2 doses of MMR vaccine. The figure was 93.7% in families that had private insurance (H. A. Hill et al. Morbid. Mortal. Wkly Rep. 68, 913–918; 2019). Berman distinguishes between two groups of parents whose children are not fully vaccinated: those who reject vaccination, and those who lack access to health care. There should be more emphasis on the have-nots, in my view. Instead, his focus is on the refusers, arguing that because some cannot access care, those who can should be vaccinated. This is a common blind spot in explanations of low take-up. Poverty, and lack of access to social resources and primary care, greatly affect uptake, as do housing insecurity, gender inequity and racism. The largest measles outbreaks in 2019 were in countries without sufficient primary care, such as Madagascar, or where conflict had displaced people and disrupted their access to vaccines, such as Yemen. Some of the most effective interventions include ensuring that supply chains are reliable, making services highly convenient and simply reminding people that they need to be vaccinated. The current pandemic reminds us that governments cannot ignore poverty and social exclusion if they are to prevent and manage this virus, others unvanquished and those yet to come. By taking the story of vaccine opposition back to its earliest examples, Anti-vaxxers cautions against simplistic solutions. In tracing the movement across three centuries, Berman underlines that is unlikely to be ended by keyboard warriors or the repetition of even the best scientific evidence. Julie Leask chairs the World Health Organization Measuring Behavioural and Social Drivers of Vaccination Working Group. She is a professor in the Susan Wakil School of Nursing and Midwifery at the University of Sydney in Australia, and visiting fellow at the National Centre for Immunisation Research and Surveillance in Sydney. e-mail: [email protected]

Books in brief Net Zero Dieter Helm William Collins (2020) Climate-change economist Dieter Helm was frustrated by a widely repeated claim from the UK Committee on Climate Change: “By reducing emissions produced in the UK to zero, we also end our contribution to rising global temperatures.” Not so, he objects: consumers also import goods and services from countries with high emissions, notably China. As Helm bluntly argues in international detail, reaching ‘net zero’ emissions will require unpopular unilateral changes in individual lifestyles and national infrastructures.

The Human Cosmos Jo Marchant Dutton (2020) Galileo composed horoscopes for his illegitimate daughters, notes science journalist Jo Marchant in her multifaceted meditation on humanity’s relationship with the cosmos. From possibly celestial Palaeolithic cave art at Lascaux in France to awestruck astronauts in space, she considers how patterns in the sky have governed lives on Earth, “shaping ideas about time and place; power and truth; life and death”. Although science is right to debunk astrology, she argues, the significance of the heavens has been eclipsed by modern astronomy.

The Inside Out of Flies Erica McAlister Natural History Museum (2020) “Flies are not filthy … they are always cleaning themselves,” notes entomologist Erica McAlister’s caption for a photo of a fly maintaining its antennae — one of many eye-popping images in her erudite, irresistible natural history of the insects. She agrees with naturalist Pliny, who wrote two millennia ago that insects display nature’s “exhaustless ingenuity”. Consider Ephydra hians, which “scuba-dives” in alkaline lakes — using hydrophobic hairs that trap an air bubble like an external lung — to lay its eggs on the lake bottom.

Outside the Box Marc Levinson Princeton Univ. Press (2020) This history of globalization evokes economist Marc Levinson’s 2006 book The Box, about container ships. These were key to the ‘third globalization’, starting in the 1980s: products were manufactured in places with low wages, then shipped to the advanced economies where they had been designed. In today’s ‘fourth globalization’, research, engineering and design are moving, and manufacturing can be done anywhere. Much of the process involves ideas, such as software, rather than ‘stuff’ in a box. But how can it be politically regulated?

The Polymath Peter Burke Yale Univ. Press (2020) From the mid-nineteenth century, science has abounded in specialists, yet polymaths such as Alan Turing and Linus Pauling have remained crucial. In a mind-stretching history, Peter Burke describes “500 western polymaths” from the half-millennium since Leonardo da Vinci. He discusses their curiosity, concentration, memory, speed, imagination, restlessness, hard work and horror of wasting time. But he overlooks specialists with polymathic tendencies, such as Albert Einstein, Florence Nightingale and Ronald Ross. Andrew Robinson

. 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 585 | 24 September 2020 | 501

Expert insight into current research

FLORIAN PLAUCHEUR/AFP/GETTY

News & views

Figure 1 | A bean field bordering a rainforest reserve near Sorriso, Brazil.

Conservation

A recipe to reverse the loss of nature Brett A. Bryan & Carla L. Archibald

How can the decline in global biodiversity be reversed, given the need to supply food? Computer modelling provides a way to assess the effectiveness of combining various conservation and food-system interventions to tackle this issue. See p.551 Nature is in trouble, and its plight will probably become even more precarious unless we do something about it1. On page 551, Leclère et al.2 quantify what might be needed to reverse this deeply worrying path while also feeding people’s increasingly voracious appetites. The authors’ answer is to team ambitious conservation measures with food-system transformation in the hope of reversing the trend of global terrestrial biodiversity loss. By nature, we mean the diversity of life that has evolved over billions of years to exist in dynamic balance with Earth’s biophysical environment and the ecosystems present. Nature contributes to human well-being in many ways, and the services it provides, such

as carbon sequestration by plants or pollination by insects, could impose a vast cost if lost3. Although the slow and long-term decline of Earth’s biodiversity4 is often overshadowed by climate change, and more recently by the COVID-19 pandemic, the loss of biodiversity is no less of a risk than those posed by the other challenges. Many would argue that the effect of biodiversity losses could surpass the combined impacts of climate change and COVID-19. More and more, the realization is growing that, as a planet, we are what we eat. Human demand for food is accelerating with the ever-increasing global population (projected to approach 10 billion by 2050), and each successive generation is wealthier and consumes

more resource-intensive diets than did the previous one5. Trying to balance this rapidly rising demand against the limited amount of land available for crops and pasture sets agriculture and nature (Fig. 1) on a collision course6. As Leclère and colleagues show, a bold and integrated strategy is required immediately to turn this around. Taking a long view out to the year 2100, Leclère et  al. present a global modelling study assessing the ability of ambitious conservation and food-system intervention scenarios to reverse the decline, or, as they call it, “bending the curve”, of biodiversity losses resulting from changes in agricultural land use and management. Projections of future land use and biodiversity are uncertain, and when these models are combined, this uncertainty is compounded. One of the great innovations of Leclère and colleagues’ work is in embracing this uncertainty by combining an ensemble of four global land-use models and eight global biodiversity models and measuring the performance of future land-use scenarios in terms of higher-level model-independent metrics such as the amount of biodiversity loss avoided. Importantly, the study also included a baseline (termed BASE) scenario — the world expected without interventions — and Leclère et al. used this to gauge the effectiveness of the intervention scenarios. Although it is not a focus of the paper, it’s worth pausing to ponder the sobering picture painted 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 585 | 24 September 2020 | 503

News & views this business-as-usual future largely bereft of birdsong and insect chirp. Choosing to act now can make a difference to nature’s plight. Most (61%) of the model combinations run by the authors indicated that implementing ambitious conservation actions led to a positive uptick in the bio­ diversity curve by 2050. Such conservation actions included: extending the global conservation network by establishing protected nature reserves; restoring degraded land; and basing future land-use decisions on comprehensive landscape-level conservation planning. This comprehensive conservation strategy avoids more than half (an average of 58%) of the biodiversity losses expected if nothing is done, but also leads to a hike in food prices. When conservation actions were teamed with a range of equally ambitious food-system interventions, the prognosis for global biodiversity in the model was improved further. Including both supply- and demand-side measures, these approaches included boosting agricultural yields, having an increasingly globalized food trade, reducing food waste by half, and the global adoption of healthy diets by halving meat consumption. These combined measures of conservation and food-systems actions avoided more than two-thirds of future biodiversity losses, with the integrated action portfolio (combining all actions) avoiding an average of 90% of future biodiversity losses. Almost all models predicted a biodiversity about-face by mid-century. These food-system measures also avoided adverse outcomes for food affordability. Leclère and colleagues’ work complements the current global climate-change scenario framework (tools for future planning by governments and others, including scenarios called shared socio-economic pathways, which integrate future socio-economic projections with greenhouse-gas emissions), and represents the most comprehensive incorporation of biodiversity into this scenario framing7 so far. However, a major limitation of the present study is that it does not consider the potential impact of climate change on biodiversity. This raises an internal inconsistency because, on the one hand, the baseline scenario considers land-use, social and economic changes under approximately 4 °C of global heating by 2100 (ref. 8), yet, on the other hand, it does not consider the profound effect of warming on plant and animal populations and the ecosystems they comprise9. Also absent from the models were other threats to biodiversity, including harvesting, hunting and invasive species 10. Although Leclère and colleagues recognized these limitations and assigned them a high priority for future research, unfortunately for us all, omitting these key threats probably means that the authors’ estimates of biodiversity’s

Brett A. Bryan and Carla L. Archibald are at the Centre for Integrative Ecology, Deakin University, Melbourne, Victoria 3125, Australia. e-mail: [email protected]

1. Díaz, S. et al. Science 366, eaax3100 (2019). 2. Leclère, D. et al. Nature 585, 551–556 (2020). 3. Costanza, R. et al. Glob. Environ. Change Hum. Policy Dimens. 26, 152–158 (2014). 4. Butchart, S. H. M. et al. Science 328, 1164–1168 (2010). 5. Springmann, M. et al. Nature 562, 519–525 (2018). 6. Montesino Pouzols, F. et al. Nature 516, 383–386 (2014). 7. Kok, M. T. J. et al. Biol. Conserv. 221, 137–150 (2018). 8. Leclère, D. et al. Towards Pathways Bending the Curve of Terrestrial Biodiversity Trends Within the 21st Century https://doi.org/10.22022/ESM/04-2018.15241 (Int. Inst. Appl. Syst. Analysis, 2018). 9. Warren, R., Price, J., Graham, E., Forstenhaeusler, N. & VanDerWal, J. Science 360, 791–795 (2018). 10. Driscoll, D. A. et al. Nature Ecol. Evol. 2, 775–781 (2018). This article was published online on 9 September 2020.

Biotechnology

Yeast learns a sorceress’s secret José Montaño López & José L. Avalos

Yeast has been engineered to convert simple sugars and amino acids into drugs that inhibit a neurotransmitter molecule. The work marks a step towards making the production of these drugs more reliable and sustainable. See p.614 In Homer’s Odyssey, the sorceress Circe slipped Odysseus’ companions a poison to induce amnesia and hallucinations. Scientists have speculated1 that Circe’s concoction contained the plant jimsonweed (Datura stramonium), which is rich in drugs called tropane alkaloids that are used to treat asthma, influenza symptoms and pain, and that can induce hallu­cino­ genic an­d other psychotropic effects. Tropane alkaloids, like most other plant natural products, are still typically extracted from natural sources, but this approach has many pitfalls. For instance, vulnera­bility to weather and market fluctua­tions can limit access for both patients and researchers, and extraction can be environ­mentally harmful2,3. In addition, plants typically contain very low levels of these active ingredients. On page 614, Srinivasan and Smolke4 report an alternative way to make tropane alkaloids that could relieve these limitations — using engineered strains of the baker’s yeast Saccharomyces cerevisiae. Plants produce a variety of specialized compounds that help them to adapt and survive. Biosynthesis of these natural products often involves lengthy metabolic pathways that have complex dynamics and regulation. One of the major achievements in the

. 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 ©

504 | Nature | Vol 585 | 24 September 2020

plight and the effectiveness of integrated global conservation and food-system action are overly optimistic. To truly bend the curve, Leclère and colleagues’ integrated portfolio will need to be substantially expanded to address the full range of threats to biodiversity. Although the models say that a better future is possible, is the combination of the multiple ambitious conservation and food-system interventions considered by Leclère et al. a realistic possibility? Achieving each one of the conservation and food-system actions would require a monumental coordinated effort from all nations. And even if the global community were to get its act together in prioritizing conservation and food-system transformation, would such efforts come in time and be enough to save our planet’s natural legacy? We certainly hope so.

field of metabolic engineering has been the development of microorganisms that can produce plant natural products5–7. However, the approach is far from routine because the enzymes involved in biosynthesis are often unknown, might be inactive in microbial hosts, and can be segregated across different plant subcellular compartments, cells or tissues. Srinivasan and Smolke have overcome these challenges to produce a strain of S. cerevisiae that converts simple sugars and amino acids into two tropane alkaloids, hyoscyamine and scopolamine. These tropane alkaloids block the action of the neurotransmitter molecule acetylcholine8. They are used to treat nausea, gastrointestinal problems, excessive bodily secretions and neuromuscular disorders, including Parkinson’s disease9,10. Srinivasan and Smolke genetically engin­ eered their yeast strain to overexpress 26 genes from different kingdoms of life. Together, these genes encode several metabolic enzymes and transporter proteins. Key to the authors’ achievement is the fact that they separated the enzymes and transporters into six subcellular locations — the cytosolic fluid, four organelles (the mitochondrion, peroxisome,vacuole and endoplasmic reticulum),

News & views this business-as-usual future largely bereft of birdsong and insect chirp. Choosing to act now can make a difference to nature’s plight. Most (61%) of the model combinations run by the authors indicated that implementing ambitious conservation actions led to a positive uptick in the bio­ diversity curve by 2050. Such conservation actions included: extending the global conservation network by establishing protected nature reserves; restoring degraded land; and basing future land-use decisions on comprehensive landscape-level conservation planning. This comprehensive conservation strategy avoids more than half (an average of 58%) of the biodiversity losses expected if nothing is done, but also leads to a hike in food prices. When conservation actions were teamed with a range of equally ambitious food-system interventions, the prognosis for global biodiversity in the model was improved further. Including both supply- and demand-side measures, these approaches included boosting agricultural yields, having an increasingly globalized food trade, reducing food waste by half, and the global adoption of healthy diets by halving meat consumption. These combined measures of conservation and food-systems actions avoided more than two-thirds of future biodiversity losses, with the integrated action portfolio (combining all actions) avoiding an average of 90% of future biodiversity losses. Almost all models predicted a biodiversity about-face by mid-century. These food-system measures also avoided adverse outcomes for food affordability. Leclère and colleagues’ work complements the current global climate-change scenario framework (tools for future planning by governments and others, including scenarios called shared socio-economic pathways, which integrate future socio-economic projections with greenhouse-gas emissions), and represents the most comprehensive incorporation of biodiversity into this scenario framing7 so far. However, a major limitation of the present study is that it does not consider the potential impact of climate change on biodiversity. This raises an internal inconsistency because, on the one hand, the baseline scenario considers land-use, social and economic changes under approximately 4 °C of global heating by 2100 (ref. 8), yet, on the other hand, it does not consider the profound effect of warming on plant and animal populations and the ecosystems they comprise9. Also absent from the models were other threats to biodiversity, including harvesting, hunting and invasive species 10. Although Leclère and colleagues recognized these limitations and assigned them a high priority for future research, unfortunately for us all, omitting these key threats probably means that the authors’ estimates of biodiversity’s

Brett A. Bryan and Carla L. Archibald are at the Centre for Integrative Ecology, Deakin University, Melbourne, Victoria 3125, Australia. e-mail: [email protected]

1. Díaz, S. et al. Science 366, eaax3100 (2019). 2. Leclère, D. et al. Nature 585, 551–556 (2020). 3. Costanza, R. et al. Glob. Environ. Change Hum. Policy Dimens. 26, 152–158 (2014). 4. Butchart, S. H. M. et al. Science 328, 1164–1168 (2010). 5. Springmann, M. et al. Nature 562, 519–525 (2018). 6. Montesino Pouzols, F. et al. Nature 516, 383–386 (2014). 7. Kok, M. T. J. et al. Biol. Conserv. 221, 137–150 (2018). 8. Leclère, D. et al. Towards Pathways Bending the Curve of Terrestrial Biodiversity Trends Within the 21st Century https://doi.org/10.22022/ESM/04-2018.15241 (Int. Inst. Appl. Syst. Analysis, 2018). 9. Warren, R., Price, J., Graham, E., Forstenhaeusler, N. & VanDerWal, J. Science 360, 791–795 (2018). 10. Driscoll, D. A. et al. Nature Ecol. Evol. 2, 775–781 (2018). This article was published online on 9 September 2020.

Biotechnology

Yeast learns a sorceress’s secret José Montaño López & José L. Avalos

Yeast has been engineered to convert simple sugars and amino acids into drugs that inhibit a neurotransmitter molecule. The work marks a step towards making the production of these drugs more reliable and sustainable. See p.614 In Homer’s Odyssey, the sorceress Circe slipped Odysseus’ companions a poison to induce amnesia and hallucinations. Scientists have speculated1 that Circe’s concoction contained the plant jimsonweed (Datura stramonium), which is rich in drugs called tropane alkaloids that are used to treat asthma, influenza symptoms and pain, and that can induce hallu­cino­ genic an­d other psychotropic effects. Tropane alkaloids, like most other plant natural products, are still typically extracted from natural sources, but this approach has many pitfalls. For instance, vulnera­bility to weather and market fluctua­tions can limit access for both patients and researchers, and extraction can be environ­mentally harmful2,3. In addition, plants typically contain very low levels of these active ingredients. On page 614, Srinivasan and Smolke4 report an alternative way to make tropane alkaloids that could relieve these limitations — using engineered strains of the baker’s yeast Saccharomyces cerevisiae. Plants produce a variety of specialized compounds that help them to adapt and survive. Biosynthesis of these natural products often involves lengthy metabolic pathways that have complex dynamics and regulation. One of the major achievements in the

. 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 ©

504 | Nature | Vol 585 | 24 September 2020

plight and the effectiveness of integrated global conservation and food-system action are overly optimistic. To truly bend the curve, Leclère and colleagues’ integrated portfolio will need to be substantially expanded to address the full range of threats to biodiversity. Although the models say that a better future is possible, is the combination of the multiple ambitious conservation and food-system interventions considered by Leclère et al. a realistic possibility? Achieving each one of the conservation and food-system actions would require a monumental coordinated effort from all nations. And even if the global community were to get its act together in prioritizing conservation and food-system transformation, would such efforts come in time and be enough to save our planet’s natural legacy? We certainly hope so.

field of metabolic engineering has been the development of microorganisms that can produce plant natural products5–7. However, the approach is far from routine because the enzymes involved in biosynthesis are often unknown, might be inactive in microbial hosts, and can be segregated across different plant subcellular compartments, cells or tissues. Srinivasan and Smolke have overcome these challenges to produce a strain of S. cerevisiae that converts simple sugars and amino acids into two tropane alkaloids, hyoscyamine and scopolamine. These tropane alkaloids block the action of the neurotransmitter molecule acetylcholine8. They are used to treat nausea, gastrointestinal problems, excessive bodily secretions and neuromuscular disorders, including Parkinson’s disease9,10. Srinivasan and Smolke genetically engin­ eered their yeast strain to overexpress 26 genes from different kingdoms of life. Together, these genes encode several metabolic enzymes and transporter proteins. Key to the authors’ achievement is the fact that they separated the enzymes and transporters into six subcellular locations — the cytosolic fluid, four organelles (the mitochondrion, peroxisome,vacuole and endoplasmic reticulum),

and the vacuolar membranes. Subcellular compartmentalization of enzymes can improve product biosynthesis by enabling proper enzymatic activity and isolating metabolic intermediates to reduce their toxicity and loss to competing pathways11. By restricting space, compartmentalization also increases local interactions between the enzymes and their targets. Such separation of enzymes is therefore akin to what happens in chemical factories, in which different synth­ esis steps are conducted in different reactors, and so each step can be separately optimized to maximize productivity. The authors divide the biosynthetic pathway for hyoscyamine and scopolamine into five modules (Fig. 1), the first two of which they described in work published last year12. In module  I, the glucose-derived amino acid glutamate is converted to another amino acid, arginine, in a series of reactions that occurs partly in the mitochondrion. Arginine is then converted to putrescine in the cytosol. In module II, putrescine is converted to tropine — the functional core that gives tropane alkaloids their name — through several cytosolic reactions, in addition to one catalysed in the peroxisome and another catalysed by an enzyme anchored to the membrane of the endoplasmic reticulum. Module III occurs in parallel with modules I and II in the cytosol, and converts glucose and the amino acid phenylalanine into the molecule phenyllactic acid glucoside (PLA glucoside). For this module, the authors engineered their strain to express an enzyme called PLA UDP-glucosyltransferase, which is found in the deadly nightshade plant Atropa belladonna and catalyses the production of PLA glucoside. The tropine produced in module II and the PLA glucoside from module III are imported into the vacuole. Next, in module V (which is counter-intuitively numbered last because all its elements constitute new discoveries), tropine and PLA glucoside are converted into the molecule littorine. Building module V involved two key steps. First, Srinivasan and Smolke engineered their strain to express a transporter protein from the tobacco plant Nicotiana tabacum that imports tropine into vacuoles. Second, they engineered the cells to express a variant of the A. belladonna enzyme littorine synthase (AbLS). When expressed in yeast, AbLS stalls in the trans-Golgi network (TGN; part of an organelle called the Golgi), and so cannot catalyse vacuolar littorine production. The authors therefore engineered AbLS to become a transmembrane protein — these proteins are transported from the TGN to the vacuole by default. This AbLS variant is able to catalyse littorine production in the vacuole. The final step of the pathway, module IV, partly occurs in the membrane of the endoplasmic reticulum. In this module,

Yeast cell

Module I

Module II

Module IV

CH3 N

OH O

Glucose

Putrescine Peroxisome

Tropine

ER

CH3

Mitochondrion Module III

O Hyoscyamine

Module V

N O

OH O O Scopolamine

Phenylalanine

PLA glucoside

Vacuole

Littorine

Figure 1 | Producing tropane-alkaloid molecules in yeast. Srinivasan and Smolke4 engineered the yeast Saccharomyces cerevisiae to make the drugs hyoscyamine and scopolamine from glucose and amino acids. Their biosynthetic pathway is divided into five modules, and several reactions are restricted to membranebound organelles. In module I, glucose is converted to the molecule putrescine, by metabolic steps in the mitochondrion and cytosolic fluid. In module II, enzymatic reactions in the peroxisome and the membrane of the endoplasmic reticulum (ER) catalyse the conversion of putrescine to tropine. In module III (which occurs in parallel with modules I and II), glucose and the amino acid phenylalanine are converted to phenyllactic acid glucoside (PLA glucoside). In module V, tropine and PLA glucoside are transported into the vacuole and together converted to littorine. Finally, in module IV, part of which occurs in the ER membrane, littorine is converted to hyoscyamine, which is then converted to scopolamine.

littorine is converted to hyoscyamine and then to scopolamine. The final step in hyoscyamine production involves the enzyme hyoscyamine dehydrogenase (HDH), but the gene that encodes this enzyme was unknown. The authors analysed a data set of gene-expression profiles from A. belladonna to generate 12 candidate genes. They expressed each of these candidates in yeast strains to determine which had the desired enzymatic activity. They then compared the activity of the HDH-encoding gene from A. belladonna with equivalents from other plants, and finally selected the gene from Circe’s jimsonweed as being optimal for hyoscyamine and scopolamine production. Alongside these steps, Srinivasan and Smolke deleted enzymes native to S. cerevisiae that consume key intermediate metabolite molecules, and overexpressed others to increase the production of metabolites required by the biosynthetic pathway. Together, their work is a major achievement that demonstrates the potential of microbial platforms to enable cheaper, faster, more-reliable and more-sustainable means of producing ­pharmaceuticals. Their yeast strain produced only a few micrograms to milligrams of tropane alkaloids per litre of yeast culture — not yet sufficient to replace our current methods of production through plant extraction. Nonetheless, it is an essential milestone towards this goal. To further increase production, it will be necessary to optimize each module of the pathway, as one would optimize each reaction in a chemical factory. This will involve increasing the rate of tropane-alkaloid biosynthesis by upregulating or downregulating native enzymes, boosting metabolite transport across subcellular compartments, and

improving the activity of enzymes at metabolic bottlenecks (key steps in the pathway that impede faster biosynthesis). Going forward, researchers should explore possible permutations of Srinivasan and Smolke’s biosynthetic pathway. Perhaps variations in the pathway could lead to the discovery of new drugs that have improved efficacy and reduced side effects. We might even discover drugs to treat other ailments. José Montaño López and José L. Avalos are in the Department of Chemical and Biological Engineering, Princeton University, Princeton, New Jersey 08544, USA. J.L.A. is also in the Andlinger Center for Energy and the Environment, the Department of Molecular Biology and the Princeton Environmental Institute, Princeton University. e-mail: [email protected]

1. Kaplan, M. Science Of The Magical: From the Holy Grail to Love Potions to Superpowers (Simon & Schuster, 2015). 2. Cravens, A., Payne, J. & Smolke, C. D. Nature Commun. 10, 2142 (2019). 3. Li, S., Li, Y. & Smolke, C. D. Nature Chem. 10, 395–404 (2018). 4. Srinivasan, P. & Smolke, C. D. Nature 585, 614–619 (2020). 5. Ro, D.-K. et al. Nature 440, 940–943 (2006). 6. Ajikumar, P. K. et al. Science 330, 70–74 (2010). 7. Brown, S., Clastre, M., Courdavault, V. & O’Connor, S. E. Proc. Natl Acad. Sci. USA 112, 3205–3210 (2015). 8. Kohnen-Johannsen, K. L. & Kayser, O. Molecules 24, 796 (2019). 9. Kukula-Koch, W. A. & Widelski, J. in Pharmacognosy: Fundamentals, Applications and Strategy (eds Badal, S. & Delgoda, R.) 163–198 (Elsevier, 2017). 10. Grynkiewicz, G. & Gadzikowska, M. Pharmacol. Rep. 60, 439–463 (2008). 11. Hammer, S. K. & Avalos, J. L. Nature Chem. Biol. 13, 823–832 (2017). 12. Srinivasan, P. & Smolke, C. D. Nature Commun. 10, 3634 (2019). This article was published online on 2 September 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 585 | 24 September 2020 | 505

News & views Materials science

Elusive photonic crystals come a step closer John C. Crocker

Researchers have long sought materials in which light behaves the way electrons do in semiconductors. A workable approach for growing such materials in bulk now seems at hand, and could lead to advances in computing. See p.524 In 1987, the physicist Eli Yablonovitch predicted that materials called photonic bandgap crystals (PBCs) would enable light to be handled in the way existing microcircuits handled electrical signals1. Since then, one- and two-dimensional cousins of PBCs have been microfabricated2, for which several applications have been found. Although some small PBCs have been formed by direct microfabrication3, a bulk 3D PBC material has been elusive, along with its potential applications — including next-generation computing technology. On page 524, He et al.4 report the growth of opal-like crystals that have the unusual structure required for PBCs: transparent micro­particles arranged in a manner akin to the carbon atoms in a diamond crystal. For a working PBC, these materials will need to be used as moulds to form Swiss-cheese-like ‘inverse opals’ that have holes where the current crystals have particles. To understand the physics of materials such as PBCs and semiconductors, imagine trying to run across a furrowed field. If your stride matched the spacing between the furrows, a

you might find that you can run at two speeds: quickly, by skipping along the tops of the furrows; or more slowly, by letting your feet fall in the muddy troughs. Analogously, when a wave passes through a periodic medium that has alternating more- or less-dense ‘furrows’, it can propagate in two ways: with its crests on the peaks of the furrows, or with its crests falling between these peaks. In general, such a wave has two possible energies, corresponding to the two modes of propagation; it is not possible for any such wave to have an energy in the gap between these values. In a 3D crystal, the spacing of the furrows and the gap energies depend on the direction of the wave’s motion with respect to the axes of the crystal lattice. However, for certain kinds of crystal, there can be a range of wave energies, known as a bandgap, for which waves cannot propagate in any direction at all. In a silicon-crystal semiconductor, the waves are electrons, and the bandgap means that electrons of certain energies cannot exist, enabling devices such as transistors — the tiny switches that are ubiquitous in modern electronics.

Yablonovitch showed theoretically that a similar bandgap phenomenon could occur for light waves, but only for a few crystal structures resembling the diamond lattice, and formed of microscopic particles made of certain transparent materials. Fortuitously, microparticles of the required size will often spontaneously arrange themselves into analogous ordered structures, termed colloidal crystals. Indeed, opals are naturally formed, fossilized colloidal crystals of silica particles, and the sparkle of opals is caused by the energy gaps described above. When light shines on an opal, some of the photons will have an energy (associated with a colour) in the gap. Such photons cannot enter the crystal, resulting in nearly 100% reflection. The gap energies (and therefore the reflected colours) depend on the direction of the incident light, giving opals their characteristic ‘fire’. Despite optimism in the 1990s that a simple method would yield diamond-like colloidal crystals, more than two decades and several innovations5 would be required as a prelude to He and colleagues’ achievement. In a diamond lattice, every particle is connected to four equally spaced nearest neighbours. But making particles that attach to only four neighbours does not suffice to form diamond. When two such particles come together, they must also be rotated such that the other six particles they bind to are in the correct relative orientation. To achieve this feat, He et al. synthesized microscopic plastic building blocks that resemble chubby balloon animals. Each building block consists of four merged spheres in the shape of a triangular pyramid, with a recessed sticky patch in the centre of each pyramid face (Fig. 1a). When suspended in a drop of water, particles that dock together through their sticky patches are forced into the required angular configuration. These

b

Figure 1 | Growth of opal-like crystals with a long-sought structure analogous to that of diamond. a, He et al.4 synthesized microscopic plastic particles consisting of four merged spheres in the shape of a triangular pyramid, with a recessed sticky patch in the centre of each pyramid face. Some of these patches are highlighted in blue. b, When suspended in water, these particles dock through their sticky patches to spontaneously form opal-like ordered materials, in which the particles are arranged in a manner analogous to the atoms in a crystal. In the crystal shown, the particles mimic the arrangement of carbon atoms in diamond. Scale bars, 1 µm.

. 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 ©

506 | Nature | Vol 585 | 24 September 2020

particles then spontaneously form highly ordered, stable crystals that have the longsought diamond structure (Fig. 1b). The authors have so far produced crystals containing only about 100,000 particles and weighing less than one microgram. However, scaling up their process should be straightforward. Then, all that remains to form large 3D PBCs is to chemically fill the empty space in these crystals with pure silicon or titanium dioxide (for use with infrared or visible light, respectively) and then dissolve the building blocks. One of the most exciting possible applications of PBCs is for quantum computers. In these devices, the digital bits that store values of ‘0’ or ‘1’ in a conventional computer are replaced with quantum bits (qubits) that can be both ‘0’ and ‘1’ at the same time. This replacement enables impressively faster computation of many difficult combinatorial problems that can be encountered in code-breaking. The challenge of building practical quantum computers lies in connecting many qubits together, typically using photonic signals, as well as isolating the qubits so that they do not get scrambled by interference from the outside world. The piping around of photons in a PBC microcircuit is a solution to the first problem, and 2D PBCs have already been used to build prototype quantum devices6. But because current quantum photonic circuits are thin 2D sheets, their performance is limited — photons can leak out and disturbances can leak in. A simple solution to both problems would be to sandwich these circuits between two slabs of 3D PBC. More generally, bulk PBCs will enable a broad range of technologies in the production of large quantum systems7, their controlled manipulation using light, and interfacing with conventional electronics8. The ultimate potential and applications of such technologies challenge our imagination. John C. Crocker is in the Department of Chemical and Biomolecular Engineering, University of Pennsylvania, Philadelphia, Pennsylvania 19104, USA. e-mail: [email protected]

Yablonovitch, E. Phys. Rev. Lett. 58, 2059–2062 (1987). Painter, O. et al. Science 284, 1819–1821 (1999). Subramania, G. et al. Nano Lett. 11, 4591–4596 (2011). He, M. et al. Nature 585, 524–529 (2020). Wang, Y. et al. Nature 491, 51–55 (2012). Olthaus, J., Schrinner, P. P. J., Reiter, D. E. & Schuck, C. Adv. Quantum Technol. 3, 1900084 (2020). 7. Jiang, J.-H. & John, S. Sci. Rep. 4, 7432 (2014). 8. Rudolph, T. APL Photon. 2, 030901 (2017).

1. 2. 3. 4. 5. 6.

Plant biology

Calcium channel helps shut the door on intruders Keiko Yoshioka & Wolfgang Moeder

Disease-causing microorganisms can invade plants through leaf pores called stomata, which close rapidly in a calciumdependent manner on detecting such danger. The calcium channels involved have now finally been identified. See p.569 In plants, calcium ions (Ca2+) function as a central signal for diverse stimuli, ranging from internal developmental cues to physical or biological stresses such as infection. However, the transient nature of Ca2+ signals and the enigmatic identities of plant Ca2+ channels have made the role of these ions difficult to study. Moreover, the connection between Ca2+ channels and specific plant responses is often unclear. On page 569, Thor et al.1 now clarify one such connection, and report their finding of a type of Ca2+ channel that is activated during a specific response against infection. Two specialized, moon-shaped cells, called guard cells, form a leaf pore called a stoma (Fig. 1a). Stomata allow gas exchange, including the entry of carbon dioxide for the energy-generating process of photosynthesis. They are thus essential for plant survival. However, disease-causing microorganisms (pathogens) can use stomata as a gateway for invasion. To limit infection, plants close stomata on recognizing such an attack, in a defence response called stomatal immunity2. The surfaces of the cells of both plants and animals have receptor proteins containing regions called kinase domains, and these proteins can recognize evolutionarily conserved microbial molecular motifs called pathogen-associated molecular patterns (PAMPs) and initiate signalling pathways needed for defence. In the model plant Arabidopsis thaliana, a receptor protein called FLS2, which has a kinase domain, binds to the bacterial protein flagellin, recognizing a region of this PAMP called flg22. This recognition event causes FLS2 to form an active receptor complex with another cell-surface receptor kinase called BAK1. The complex adds a phosphate group to a cytoplasmic kinase called BIK1. This phosphorylation of BIK1 activates immune responses3, such as the production of reactive oxygen species by the protein RBOHD. BIK1 is required for stomatal immunity4, and if guard cells contain a mutant version of the gene that encodes this kinase, the plant cannot respond

to flg22. However, the link between PAMP recognition and Ca 2+-mediated stomatal closure regulated by BIK1 has been unclear. To join the dots, Thor et al. speculated that, through direct phosphorylation, BIK1 controls the Ca2+ channel(s) required for stomatal immunity. The authors focused on an ion channel called OSCA1.3, which is phosphorylated on sensing flg22. Thor and colleagues report that OSCA1.3 is permeable to Ca2+, and that phosphorylation of OSCA1.3 by BIK1 at serine amino-acid residue 54 (in the same type of motif as that phosphorylated by BIK1 in RBOHD) activates this channel on pathogen recognition (Fig. 1b). Furthermore, the authors’ observation that the gene that encodes OSCA1.3 is specifically expressed in stomata is consistent with a role for the channel in stomatal immunity. The OSCA family of proteins are evolutionarily conserved ion channels, and A. thaliana contains 15 members of this family. Each ion channel is probably formed of two OSCA proteins. The largest group of these proteins, clade  1, includes OSCA1.1, OSCA1.2 (also known as OSCA1) and OSCA1.3 (refs 5–7). OSCA1.1 and OSCA1.2 are Ca2+-permeable channels that are also permeable to several other types of positively charged ion (cations), and they are activated by an ionic imbalance known as osmotic stress5,7 . Thor et al. observed no clear effect on the immune response to flg22 in a mutant plant in which the gene OSCA1.3 was disabled. However, in a plant engineered also to have a mutant version of another clade 1 member — the gene OSCA1.7 — stomatal closure on perceiving flg22 was impaired and susceptibility to bacterial infection was enhanced, compared with the response in the wild-type plant. OSCA1.7 has a similar protein motif to the one phosphorylated on OSCA1.3 by BIK1, and is activated through phosphorylation by BIK1 to generate a Ca2+ influx into cells. Thus, it seems that OSCA1.3 and OSCA1.7 are Ca2+ channels that regulate stomatal immunity and they probably function in a redundant manner,

. 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 585 | 24 September 2020 | 507

particles then spontaneously form highly ordered, stable crystals that have the longsought diamond structure (Fig. 1b). The authors have so far produced crystals containing only about 100,000 particles and weighing less than one microgram. However, scaling up their process should be straightforward. Then, all that remains to form large 3D PBCs is to chemically fill the empty space in these crystals with pure silicon or titanium dioxide (for use with infrared or visible light, respectively) and then dissolve the building blocks. One of the most exciting possible applications of PBCs is for quantum computers. In these devices, the digital bits that store values of ‘0’ or ‘1’ in a conventional computer are replaced with quantum bits (qubits) that can be both ‘0’ and ‘1’ at the same time. This replacement enables impressively faster computation of many difficult combinatorial problems that can be encountered in code-breaking. The challenge of building practical quantum computers lies in connecting many qubits together, typically using photonic signals, as well as isolating the qubits so that they do not get scrambled by interference from the outside world. The piping around of photons in a PBC microcircuit is a solution to the first problem, and 2D PBCs have already been used to build prototype quantum devices6. But because current quantum photonic circuits are thin 2D sheets, their performance is limited — photons can leak out and disturbances can leak in. A simple solution to both problems would be to sandwich these circuits between two slabs of 3D PBC. More generally, bulk PBCs will enable a broad range of technologies in the production of large quantum systems7, their controlled manipulation using light, and interfacing with conventional electronics8. The ultimate potential and applications of such technologies challenge our imagination. John C. Crocker is in the Department of Chemical and Biomolecular Engineering, University of Pennsylvania, Philadelphia, Pennsylvania 19104, USA. e-mail: [email protected]

Yablonovitch, E. Phys. Rev. Lett. 58, 2059–2062 (1987). Painter, O. et al. Science 284, 1819–1821 (1999). Subramania, G. et al. Nano Lett. 11, 4591–4596 (2011). He, M. et al. Nature 585, 524–529 (2020). Wang, Y. et al. Nature 491, 51–55 (2012). Olthaus, J., Schrinner, P. P. J., Reiter, D. E. & Schuck, C. Adv. Quantum Technol. 3, 1900084 (2020). 7. Jiang, J.-H. & John, S. Sci. Rep. 4, 7432 (2014). 8. Rudolph, T. APL Photon. 2, 030901 (2017).

1. 2. 3. 4. 5. 6.

Plant biology

Calcium channel helps shut the door on intruders Keiko Yoshioka & Wolfgang Moeder

Disease-causing microorganisms can invade plants through leaf pores called stomata, which close rapidly in a calciumdependent manner on detecting such danger. The calcium channels involved have now finally been identified. See p.569 In plants, calcium ions (Ca2+) function as a central signal for diverse stimuli, ranging from internal developmental cues to physical or biological stresses such as infection. However, the transient nature of Ca2+ signals and the enigmatic identities of plant Ca2+ channels have made the role of these ions difficult to study. Moreover, the connection between Ca2+ channels and specific plant responses is often unclear. On page 569, Thor et al.1 now clarify one such connection, and report their finding of a type of Ca2+ channel that is activated during a specific response against infection. Two specialized, moon-shaped cells, called guard cells, form a leaf pore called a stoma (Fig. 1a). Stomata allow gas exchange, including the entry of carbon dioxide for the energy-generating process of photosynthesis. They are thus essential for plant survival. However, disease-causing microorganisms (pathogens) can use stomata as a gateway for invasion. To limit infection, plants close stomata on recognizing such an attack, in a defence response called stomatal immunity2. The surfaces of the cells of both plants and animals have receptor proteins containing regions called kinase domains, and these proteins can recognize evolutionarily conserved microbial molecular motifs called pathogen-associated molecular patterns (PAMPs) and initiate signalling pathways needed for defence. In the model plant Arabidopsis thaliana, a receptor protein called FLS2, which has a kinase domain, binds to the bacterial protein flagellin, recognizing a region of this PAMP called flg22. This recognition event causes FLS2 to form an active receptor complex with another cell-surface receptor kinase called BAK1. The complex adds a phosphate group to a cytoplasmic kinase called BIK1. This phosphorylation of BIK1 activates immune responses3, such as the production of reactive oxygen species by the protein RBOHD. BIK1 is required for stomatal immunity4, and if guard cells contain a mutant version of the gene that encodes this kinase, the plant cannot respond

to flg22. However, the link between PAMP recognition and Ca 2+-mediated stomatal closure regulated by BIK1 has been unclear. To join the dots, Thor et al. speculated that, through direct phosphorylation, BIK1 controls the Ca2+ channel(s) required for stomatal immunity. The authors focused on an ion channel called OSCA1.3, which is phosphorylated on sensing flg22. Thor and colleagues report that OSCA1.3 is permeable to Ca2+, and that phosphorylation of OSCA1.3 by BIK1 at serine amino-acid residue 54 (in the same type of motif as that phosphorylated by BIK1 in RBOHD) activates this channel on pathogen recognition (Fig. 1b). Furthermore, the authors’ observation that the gene that encodes OSCA1.3 is specifically expressed in stomata is consistent with a role for the channel in stomatal immunity. The OSCA family of proteins are evolutionarily conserved ion channels, and A. thaliana contains 15 members of this family. Each ion channel is probably formed of two OSCA proteins. The largest group of these proteins, clade  1, includes OSCA1.1, OSCA1.2 (also known as OSCA1) and OSCA1.3 (refs 5–7). OSCA1.1 and OSCA1.2 are Ca2+-permeable channels that are also permeable to several other types of positively charged ion (cations), and they are activated by an ionic imbalance known as osmotic stress5,7 . Thor et al. observed no clear effect on the immune response to flg22 in a mutant plant in which the gene OSCA1.3 was disabled. However, in a plant engineered also to have a mutant version of another clade 1 member — the gene OSCA1.7 — stomatal closure on perceiving flg22 was impaired and susceptibility to bacterial infection was enhanced, compared with the response in the wild-type plant. OSCA1.7 has a similar protein motif to the one phosphorylated on OSCA1.3 by BIK1, and is activated through phosphorylation by BIK1 to generate a Ca2+ influx into cells. Thus, it seems that OSCA1.3 and OSCA1.7 are Ca2+ channels that regulate stomatal immunity and they probably function in a redundant manner,

. 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 585 | 24 September 2020 | 507

News & views a

b

Ca2+

Arabidopsis thaliana

Open stomata

Water loss closes stomata

flg22

Guard cell Cell exterior

FLS2

BAK1

OSCA calcium channel

SLAC

Pathogen sensed Closed stomata

Cytoplasm of guard cell

P BIK1

Figure 1 | A calcium channel that regulates closure of stomata. a, Plants, such as the model species Arabidopsis thaliana, have leaf pores called stomata. If the plant senses a disease-causing agent (termed a pathogen), stomatal guard cells rapidly close in response. b, Thor et al.1 describe the identification of the calcium-ion (Ca2+) channel in the pathway that leads to stomatal closure. Pathogens are sensed by the receptor protein FLS2, which forms a complex with the BAK1 protein. When this complex senses a bacterial-protein fragment, termed flg22, it adds a phosphate group (P) to the protein BIK1, thereby

such that if OSCA1.3 is absent, OSCA1.7 can fulfil its role. Whether just one or both of these proteins together form Ca2+ channels that act in stomatal immunity is unknown. In addition to identifying these Ca2+ channels, Thor et al. explored the role of the plant hormone abscisic acid (ABA), which regulates stomatal closure when the plant senses a water deficit. This hormone also controls stomatal defences, because stomata of ABA-deficient plants do not close effectively on perceiving pathogens2. However, the authors found that a plant with mutations in the genes encoding both OSCA1.3 and OSCA1.7 is fully responsive to ABA, indicating that these channels are not involved in ABA-mediated stomatal closure. This observation corroborates previous evidence4 that the regulation of stomatal immunity by BIK1 does not require ABA. Furthermore, Thor et al. report that the overall Ca2+-signal activation by flg22 in leaves that had mutations in the genes encoding both OSCA1.3 and OSCA1.7 was not impaired; guard cells make up only a small fraction of leaf cells. This result strongly supports the specific role of these channels in stomatal immunity, rather than general immunity, even though BIK1 is required for both types of response. How changes in Ca 2+ concentration deliver stimulus-specific cellular responses is a central question in this area of research. One proposed idea is that stimulus-specific temporal patterns of cytoplasmic Ca2+ levels might provide a key cue, and that these ‘Ca2+ signatures’ might be generated and decoded by specific Ca2+-binding components, such as calmodulin proteins or calcium-dependent protein kinases8. Thor and colleagues’ results suggest instead that the Ca2+ channels themselves might determine specificity, at least for stomatal immunity. Although OSCA proteins allow stomata to close independently of ABA involvement,

H 2O Anion

CDPK

activating it. BIK1 then phosphorylates the Ca2+ channel. Thor et al. report that two proteins of the OSCA family, OSCA1.3 and OSCA1.7, can function as Ca2+ channels during this response (whether one or both of these together fulfil this role is unknown). How an influx of Ca2+ through the channel causes stomatal closure is unclear. One possibility is that enzymes called calcium-dependent protein kinases (CDPKs) activate S-type anion channels (SLACs). SLACs enable anions (negatively charged ions) to exit the cell, which leads to the water loss that drives stomatal closure.

closure mediated either by ABA or in response to infection probably involves the same mechanism, which eventually closes stomata through water movement out of guard cells. Therefore, both pathways should converge at some point. The activation of channels that enable negatively charged ions (anions) to exit the cell, such as S-type anion channels, termed SLACs, is a crucial step in stomatal movement9. The protein kinase OPEN STOMATA1, which is a component of an ABA-mediated signalling pathway, activates SLACs and has been proposed2 as a point of convergence for defence responses and ABA signalling. However, some calcium-dependent protein kinases also activate SLACs10, and such Ca2+-signal decoders, or perhaps even the anion channels themselves, might

“An emerging theme in studies of plant calcium-ion channels is their regulation by phosphorylation.” be the convergence point instead. An emerging theme in studies of plant Ca2+ channels is their regulation by phosphorylation. Previous studies11,12 reported that BIK1 and BAK1 phosphorylate members of another group of plant Ca2+ channels, the cyclic nucleo­ tide-gated ion channels, to regulate their function or stability. The phosphorylation of OSCA1.3 and OSCA1.7 by BIK1 underscores the connection between receptor kinases and Ca2+ channels, presumably to generate stimulus-specific Ca2+ signals. It will be interesting to determine whether OSCA-family proteins interact with other components on the surface of cells to form a structure called a channelo­some — a group of signalling molecules surrounding an ion channel13.

. 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 ©

508 | Nature | Vol 585 | 24 September 2020

BIK1-mediated channel activation

?

?

OSCA proteins have so far been linked mostly to the sensing of osmotic stress. They are categorized as a type of mechanosensing channel, one that converts physical forces into biochemical signals14,15. Are OSCA1.3 and OSCA1.7 activated by osmotic stress or mechanical stimulation, in addition to their activation by BIK1? Did the two proteins evolve a defence-specific role, or do they also have other functions in stomata? Understanding the biological function of each OSCA and the Ca2+ signals they generate will shed light on stomatal biology. Such insights could be crucial for the bioengineering of plants to meet future challenges in crop production. Keiko Yoshioka and Wolfgang Moeder are in the Department of Cell and Systems Biology, University of Toronto, Toronto M5S 3B2, Canada. e-mails: [email protected]; [email protected]

1. Thor, K. et al. Nature 585, 569–573 (2020). 2. Melotto, M., Zhang, L., Oblessuc, P. R. & He, S. Y. Plant Physiol. 174, 561–571 (2017). 3. Couto, D. & Zipfel, C. Nature Rev. Immunol. 16, 537–552 (2016). 4. Li, L. et al. Cell Host Microbe 15, 329–338 (2014). 5. Yuan, F. et al. Nature 514, 367–371 (2014). 6. Hou, C. et al. Cell Res. 24, 632–635 (2014). 7. Murthy, S. E. et al. eLife 7, e41844. (2018). 8. McAinsh, M. R. & Pittman, J. K. New Phytol. 181, 275–294 (2009). 9. Jezek, M. & Blatt, M. R. Plant Physiol. 174, 487–519 (2017). 10. Geiger, D. et al. Proc. Natl Acad. Sci. USA 107, 8023–8028 (2010). 11. Yu, X. et al. Curr. Biol. 29, 3778–3790 (2019). 12. Tian, W. et al. Nature 572, 131–135 (2019). 13. Dietrich, P., Moeder, W. & Yoshioka, K. Plant Physiol. https://doi.org/10.1104/pp.20.00425 (2020). 14. Liu, X., Wang, J. & Sun, L. Nature Commun. 9, 5060 (2018). 15. Zhang, M. et al. Nature Struct. Mol. Biol. 25, 850–858 (2018). This article was published online on 7 September 2020.

Review

Host–microbiota maladaptation in colorectal cancer https://doi.org/10.1038/s41586-020-2729-3

Alina Janney1, Fiona Powrie1 ✉ & Elizabeth H. Mann1

Received: 7 May 2020 Accepted: 29 July 2020 Published online: 23 September 2020 Check for updates

Colorectal cancer (CRC) is a heterogeneous disease of the intestinal epithelium that is characterized by the accumulation of mutations and a dysregulated immune response. Up to 90% of disease risk is thought to be due to environmental factors such as diet, which is consistent with a growing body of literature that describes an ‘oncogenic’ CRC-associated microbiota. Whether this dysbiosis contributes to disease or merely represents a bystander effect remains unclear. To prove causation, it will be necessary to decipher which specific taxa or metabolites drive CRC biology and to fully characterize the underlying mechanisms. Here we discuss the host–microbiota interactions in CRC that have been reported so far, with particular focus on mechanisms that are linked to intestinal barrier disruption, genotoxicity and deleterious inflammation. We further comment on unknowns and on the outstanding challenges in the field, and how cutting-edge technological advances might help to overcome these. More detailed mechanistic insights into the complex CRC-associated microbiota would potentially reveal avenues that can be exploited for clinical benefit.

Colorectal cancer (CRC) is the second leading cause of cancer-related death worldwide, and its incidence is increasing, particularly in individuals under 50 years of age1,2. As such, new diagnostic and treatment options are needed. Unlike the well-described causal role of Helicobacter pylori in gastric cancer, a specific microorganism that triggers CRC has not been identified. Nevertheless, an emerging strategy is to target the altered—and putatively oncogenic—microbiota that has been reproducibly identified in patients with disease. However, we must first better understand what underlies the microbiota associations that are observed in CRC, as well as their functional consequences. Here we review published mechanisms pertaining to host–microbiota interactions in CRC, and examine how microorganisms and metabolites can elicit the hallmarks of cancer. We further consider host maladaptations that promote tumorigenesis by enabling the penetration of microorganisms through the gut wall and deleterious inflammation. Throughout this Review, we discuss potential explanations for seemingly contrasting data and identify unresolved questions. We finish by outlining technological advances that can be harnessed to accelerate progress.

The intestinal host–microbiota interface The mammalian colon co-evolves with a diverse microbial ecosystem, in which symbiotic interactions facilitate the development of both host and microbiota, in part by shaping a robust immune system in the host3. Complex dialogue between the host and the microbiota ensures peaceful coexistence with dietary components and beneficial commensals, as well as defence against potentially harmful pathogens4,5. Central to sustaining this balance are intestinal epithelial cells (IECs), which form a single-layer physical barrier at the host–microbiota interface that is continuously replenished by multipotent intestinal stem cells residing in intestinal crypts6 (Fig. 1). IEC proliferation is regulated by a

stem cell niche and involves paracrine signalling from the underlying mesenchyme7, whereas the integrity of the barrier is further reinforced by inter-epithelial tight-junction proteins alongside the secretion of protective mucins6. In addition to their function as a physical barrier, IECs sense the microbiota through pattern recognition receptors. Stimulation of these receptors promotes epithelial cell repair and the upregulation of tight-junction proteins, and triggers the production of various cytokines that signal to cells in the lamina propria6. Critically, the epithelium does not stand alone, but rather as a triad with immune cells and the mesenchyme (Fig. 1). If epithelial barrier defences are breached by microorganisms or their products, immune and mesenchymal cells act as a second line of defence by initiating a cascade of signalling networks that uphold epithelial integrity6,8. For example, activation of the NF-κB and STAT3 pathways can trigger the production of growth factors and cytokines, which support homeostatic tissue repair9. Such responses simultaneously shape the microbiota by exerting a selective pressure for particular microorganisms and by triggering the secretion of epithelial-derived factors such as mucins and antimicrobial peptides8. Although mesenchymal cells are dynamic and are critical in maintaining homeostasis10, further work is required to determine whether they interact directly with the microbiota. Nevertheless, tight-knit regulation of these interconnected pathways is essential, because excessive stimulation can transform a well-balanced network into an inflamed wound that will not heal11.

The aetiology of CRC CRC is a disease of the intestinal stem cells that is characterized by dysregulation of multiple components within the intestine (Fig. 1). Although the underlying cause is often unclear, inflammation associated with inflammatory bowel disease (IBD) is a known risk factor

Kennedy Institute of Rheumatology, University of Oxford, Oxford, UK. ✉e-mail: [email protected]

1

Nature | Vol 585 | 24 September 2020 | 509

Review

Niche modification

a

b

Barrier function

Microbiota

c

f Barrier maintenance

Immune

Tissue repair

Host

d activation

Dysbiosis in CRC

e

Mesenchyme

Immune cells

Fig. 1 | A schematic of the host–microbiota interactions in health and in colorectal cancer. The host–microbiota interface comprises a continuously evolving microbial ecosystem neighbouring host immune and mesenchymal cells, which underlie a single-layer epithelial-cell barrier. Under homeostatic conditions, bi-directional communication between host cells and the microbiota facilitates critical symbiotic functions and maintains the structural integrity of the intestine (arrows). For example, microorganisms and metabolites can trigger responses that maintain barrier function and tissue repair. This response is simultaneously affected by the genetic makeup of host cells, which in turn shapes microbiota niche formation. However, such interactions can become maladapted in CRC, in which the microbiota composition is different (a), including an increase in bacteria with genotoxic abilities (b). These changes are accompanied by deterioration of the epithelial cell barrier (c); this enables an influx of microorganisms (d) that can trigger inflammation (e) and promote established hallmarks of cancer, such as DNA damage (f).

for CRC12, as are heritable genetic defects such as familial adenomatous polyposis and Lynch syndrome13. Beyond such inherited conditions, mathematical modelling attributes 70–90% of disease risk to environmental factors—most notably diets that are low in fibre and high in red meat2,14. Pathogenic infections are also well-established environmental triggers of many cancers, with a link between H. pylori and gastric cancer first identified in 199115,16. Although evidence for a single disease-triggering pathogen in CRC is lacking, these observations have sparked a growing interest in the role of the microbiota, in which cancer-initiating microorganisms might prosper owing to ecological changes. Irrespective of what underlies the initiation of CRC, it typically involves the disruption of homeostatic immune and microbiota-derived signals, stimulating responses that have evolved to permit epithelial restitution to excess17,18. Whether such responses initiate tumorigenesis depends on a succession of genomic alterations within intestinal stem cells, often in the form of mutations in genes such as APC, KRAS or PIK3CA19. Notably, the incidence of such genetic alterations can be increased by a maladaptation of the host–microbiota interface (Fig. 1), which can contribute to the hallmarks of cancer—including sustained epithelial-cell proliferation, resistance to cell death, invasion and immune evasion20. Like the delicate balance in protective versus pathogenic epithelial responses, intestinal inflammation is considered a ‘double-edged 510 | Nature | Vol 585 | 24 September 2020

sword’ in CRC. The numbers of in situ T cells correlate with beneficial CRC outcome21, and harnessing the immune system for antigen-specific elimination remains the goal of many therapeutic strategies. However, tumours often escape such immune-mediated destruction by immunoediting tumour antigens, rendering them undetectable22. Although innate immune cells such as neutrophils and macrophages can aid tumour clearance, under certain circumstances they release reactive oxygen species, which can potentially initiate a carcinogenic cascade by damaging the genomic integrity of IECs23. Tumour cells thereby create a vicious cycle in which numerous cells in the microenvironment—including resident fibroblasts—can be reprogrammed to produce additional growth factors, cytokines and pro-angiogenic factors that sustain unrestrained proliferation and invasion17,18,24,25.

Initial evidence for host–microbiota interactions in CRC emerged in 1975, when it was shown that the carcinogen dimethylhydrazine triggered significantly less colonic tumorigenesis in germ-free rats than in those with intestinal microbiota26. More functionally, mice transplanted with a faecal microbiota from patients with CRC developed more intestinal polyps than those receiving microbiota from healthy controls27. Thanks to developments in microbiome profiling—including 16S rRNA and shotgun metagenomics—it is indisputable that individuals with CRC have a different taxonomic composition relative to healthy controls, referred to as ‘dysbiosis’28. These metagenomic and metataxonomic studies show that patients with CRC have a greater overall taxonomic diversity of faecal species and an outgrowth of certain species, the nature of which has been described extensively elsewhere29–33. In summary, a higher relative abundance of putatively pro-carcinogenic microbial members—including Fusobacterium nucleatum, Escherichia coli, Bacteroides fragilis, Enterococcus faecalis, Streptococcus gallolyticus and Peptostreptococcus spp.—has been detected in CRC tumour tissue, whereas so-called protective genera—including Roseburia, Clostridium, Faecalibacterium and Bifidobacterium—are reduced29–34. Some of these differences, most consistently levels of Fusobacterium, have been correlated with clinical outcomes and chemosensitivity, and therefore such bacteria have potential as biomarkers33. Recent analysis of biopsies from 100 patients with Lynch syndrome highlighted early microbial changes in colonic neoplasia, including a shift in flagellin contributors34,35. However, determining whether such changes are a cause or an effect of cancer, and attributing the initiation and/or progression of CRC to certain so-called ‘oncogenic’ microorganisms, remains challenging—in part due to considerable inter-individual differences within the microbiota3. Nonetheless, a recently published CRC signature identified an enrichment of 29 species across 8 geographic locations, taking the field closer to defining a ‘CRC microbiome’34. Important mechanistic questions remain as to whether the observed pathophysiology is due to the activity of the collective ‘oncogenic’ community, for example via metabolite production, or to single strains with certain genotoxic and/or yet to be discovered characteristics.

Genotoxicity induced by CRC-associated bacteria For a single microorganism or community to be considered oncogenic, they must elicit carcinogenic effects, such as causing DNA damage. Pathogens associated with cancer—including H. pylori—are known to trigger cancerous mutations36, and a number of recent studies have reported the genotoxicity of species associated with CRC, including colibactin-producing polyketide synthase (pks)+ E. coli37–39, enterotoxigenic B. fragilis (ETBF)40, E. faecalis41 and cytolethal distending toxin-producing Campylobacter jejuni42. In particular, pks+ E. coli can induce double-strand breaks, aneuploidy and improper cellular division37, an effect driven by the mutagen colibactin43,44. This was

particularly evident in a study of ApcMin/+ mice that lacked the autophagy gene Atg16l1 and thereby failed to recruit the DNA repair protein RAD51 after infection. As a result, DNA double-strand breaks accumulated and tumour burden increased45. Direct causality has been demonstrated between this CRC-associated microorganism and intestinal stem cell mutations by injecting pks+ E. coli into human organoids— self-organizing, three-dimensional, in vitro culture systems of epithelial cells. This induced a mutation signature that notably included 112 established CRC-driver mutations46. In mouse models, pks+ E. coli, ETBF and E. faecalis can similarly induce DNA damage through the induction of inflammation and oxidative stress40,47–49. In genetically susceptible Il10−/− mice, pks+ E. coli and ETBF induced 8-oxoguanine DNA lesions that correlated with a higher incidence of colonic tumours48. More recently, tungstate administration was found to reduce DNA damage and tumorigenesis in Il10−/− mice, at least in part through the inhibition of metabolic pathways that are relied on by the putatively pathogenic E. coli50. The ETBF toxin also increased spermine oxidase levels, which led to the generation of reactive oxygen species and the induction of the DNA-damage marker γ-H2A.x40.

The effect of microorganism-driven metabolism Beyond the outgrowth and functions of specific strains, the collective activities of the microbiota—which is seemingly more stable than taxonomic readouts3—deserve further attention. Several products of bacterial metabolism have been implicated in CRC, many of which are associated with dietary intake14 or drug metabolism (for example, aspirin51). These include products of protein fermentation, secondary bile acids from high fat intake, and short-chain fatty acids (SCFAs) metabolized from carbohydrates and phytochemicals52. Although the diet can be directly carcinogenic53, it can also alter the ecosystem by skewing the abundance of specific species and metabolites. To this end, a diet associated with increased levels of high-sulfur-metabolizing bacteria correlates with increased risk of distal colon and rectal cancer, an effect that is thought to be due to genotoxic activities54. Independently, a ‘Westernized’ high-fat diet was found to correlate with CRC recurrence as well as with the collagenolytic activity of certain microorganisms51.To identify candidate disease-associated metabolites, metaproteomic analysis has been performed on stools from individuals with Lynch syndrome35, adenomas and CRC55,56. These studies identified a heightened oxidative metabolic microenvironment, which is thought to be due to increased levels of reactive oxygen species and reactive nitrogen species in the colon of patients with CRC56 as well as to increased concentrations of the DNA-damaging bile acid deoxycholic acid55,57,58. Because these observations are primarily correlative in nature, in-depth mechanistic follow-up studies are required in order to establish causality. More functionally, increased tumorigenesis is observed in mice that lack free-fatty acid receptor 2 (FFAR2), a SCFA receptor. This effect is thought to be driven by reduced IEC integrity, enabling increased bacterial influx, overactivated dendritic cells and an exhausted CD8+ T cell phenotype59. Much work has focused on the putatively anti-tumorigenic SCFA butyrate, which is fermented from dietary fibre52,60, and on aryl hydrocarbon receptor ligands such as indole-3-carbinol, which is derived from vegetables61. In mouse models of CRC, the latter reduced IEC proliferation by restricting the accumulation of β-catenin62. Butyrate reportedly functions intracellularly as a histone deacetylase inhibitor to downregulate IL-663 and enhance the antimicrobial functions of macrophages64. Recently, mouse Faecalibaculum rodentium and its human counterpart Holdemanella biformis were identified as protective species in CRC, in particular due to their release of butyrate and its subsequent downstream histone deacetylase inhibitory activity that dampens tumour proliferation65. By binding to the receptor GPR109A, butyrate can promote the differentiation of anti-inflammatory IL-10-expressing T cells66 and, in the context of

a CRC model, induce apoptosis of IECs67. In a mouse model of colitis, butyrate was found to act through the transcription factor FOXO3 to suppress ISC proliferation. Although this is thought to be protective in established cancers, delaying proper wound repair could make it detrimental at earlier stages of the disease68. Alongside potential temporal dependencies, the genetic landscape of the host could be critical in determining whether butyrate has a pro- or an anti-tumorigenic role in CRC. To this end, butyrate induced IEC hyperproliferation and increased production of mitochondrial reactive oxygen species specifically in mice that were deficient in the mismatch-repair apparatus48,69.

Influx of immune-stimulating microorganisms Inflammation is a well-established driver of colorectal carcinogenesis18,25, such that individuals with IBD have an increased risk of CRC12. Bacterially induced inflammation has also been shown to positively correlate with tumour multiplicity70. However, pathways through which the microbiota shapes the immune environment of the tumour and, in turn, how that modulates the surrounding microbiota, is the focus of ongoing research (Fig. 2). Disruption of the epithelial barrier enables an influx of previously compartmentalized, and potentially harmful, microorganisms into the tissue. Enhanced tumour multiplicity and IEC permeability were reported in mice deficient in the pattern recognition receptor-associated genes Nod1, Nod2 and Ripk2, highlighting the importance of host defence and IEC function in preventing tumorigenesis71,72. Similarly, loss of the SCFA-sensing receptor FFAR2 in ApcMin/+  mice increased colonic permeability, which was associated with reduced expression of the tight-junction protein E-cadherin59. Such loss of barrier integrity within tumours is reportedly accompanied by location-specific bacterial influx73 or formation of invasive bacterial aggregates74. In humans, these aggregates of microorganisms—termed ‘biofilms’—consist of a defined community that includes B. fragilis, E. coli and F. nucleatum enclosed in an exo-polymeric matrix75,76. Biofilms are also detected in benign polyps from patients with familial adenomatous polyposis76, indicating a potential role in the development of adenomas from polyps. Although understanding of the precise function of biofilms is limited, inoculating ApcMin/+  mice with bacterial slurries from human biofilm-positive—but not biofilm-negative—tissues was found to induce tumours. Notably, tumorigenesis was observed irrespective of whether biofilm-positive inoculants were derived from patients with CRC or from cancer-free individuals, suggesting that these microorganism aggregates may be an indicator of disease risk77,78. Several questions remain here, including whether the carcinogenic properties of biofilms result from the presence of specific strains, the local microecology of the community, and/ or a maladapted response renders the host unable to react effectively to a biofilm. Bacterial translocation caused by deficiencies in the intestinal barrier correlates with—and is thought to trigger—the production of several cancer-related pro-inflammatory cytokines, including IL-1β, IL-23, IL-22, IL-27 and—of particular importance—IL-17A and IL-659,71–74,76,79–82. In agreement with this, in mice and humans, the expression of IL-6, IL-23 and IL-17A was found to increase within tumours that were infiltrated by bacteria73,81. IL-23 and IL-6 are potent inducers of T helper type 17 (TH17)-derived cytokines IL-17A and IL-2283; in addition, IL-6 binds to its receptor on IECs to trigger aberrant proliferation71,74,79. IL-6 and IL-17A therefore have intriguing translational potential, although future studies are required to determine whether invasive microorganisms unquestionably trigger increased expression at the protein level. Specific cellular sources should also be investigated, because these cytokines can be produced by—and act on—immune, epithelial and mesenchymal cells. The importance of receptor expression patterns is exemplified by a study showing that CD4+ T-cell-specific ablation of the IL-1 receptor decreased tumour-elicited inflammation, whereas lack of Nature | Vol 585 | 24 September 2020 | 511

Review Homeostasis

Maladaptation

Barrier dysfunction

Inflammation

Hyperplasia

Tumour

Microbiota Alistipes spp. A. muciniphila

F. nucleatum

pks+ E. coli

Enterotoxigenic B. fragilis

H. hepaticus ?

Mucus

P. anaerobius

FFAR2 NOD 1/2

SCFAs

a

IgA

IL-33

IL-6

Plasma cell FOXP3+ Treg cell

ATG7

d

b

c

IL-23a IL-1β

IL-10

IL-17a IL-22

Type 17 lymphoid cells (CD4+ TH17 cells and ILC3s)

CXCL1 CXCL2 CXCL5

f

e

NK cell

CD4+ TH1 cell

TAM

CD4+ T cell

TAN Pro-inflammatory myeloid cells

CCL5 CCL20

CD8+ T cell

MDSC

Fig. 2 | Known inflammatory mechanisms by which the microbiota contributes to CRC. The microbiota contributes to CRC pathogenesis via a range of emerging mechanisms. a, Microbiota-derived SCFAs induce IL-10, which can dampen inflammation. IL-33 upregulates the production of IgA, which limits the activity of mucus-degrading A. muciniphila. b, Inflammation triggered by IL-10 deficiency enables the outgrowth of harmful (Alistipes spp.), genotoxic (pks+ E. coli) and other undefined species that further enhance inflammation and contribute to genomic instability. c, Increased epithelial permeability enables the influx of various species that stimulate pro-inflammatory cytokine production, affecting epithelial cell proliferation and promoting CD4+ TH17 responses. Altered innate immunity is permissive for

H. hepaticus-induced IL-22 production by ILC3 and leads to epithelial cell proliferation. d, Biofilms containing ETBF, pks+ E. coli and F. nucleatum coat tumours and polyps and coincide with increased mRNA levels of IL6, IL23A and IL17A. e, Specific species alter the immune-cell composition of the tumour microenvironment. For example, ETBF and P. anaerobius trigger the secretion of chemokines that recruit immunosuppressive MDSCs, tumour-associated macrophages (TAMs) and tumour-associated neutrophils (TANs), whereas F. nucleatum can directly inhibit the cytotoxic activity of natural killer (NK) cells. Elements of the microbiota can be protective in CRC, such as certain Clostridia species, which promote CD4 TH1 and CD8 T cell responses (f). ILC, innate lymphoid cell.

this receptor on myeloid cells increased bacterial invasion and IL-17A secretion, as well as the number and size of proliferating tumours84. To add further complexity, many cytokines display pleiotropic functions within CRC25. For example, IL-22—which is known to be triggered by the microbiome85,86 and produced by T cells and innate lymphoid cells87—can sustain IEC proliferation and induce anti-apoptotic proteins during inflammation (pro-tumorigenic), but can also promote barrier integrity and induce the repair of DNA damage after genotoxic insults (anti-tumorigenic)87–90. It should be noted that some of these mechanistic insights are derived from model systems using mouse-adapted pathogens such as Helicobacter hepaticus87. Such disparate roles for cytokines in carcinogenesis depend on their concentration and their temporal dynamics during disease progression, as well as the context of other cell types, cytokines and mutations—variables that are often overlooked in more reductionist studies.

The abundance of specific ‘oncogenic’ species seems to depend in part on the anti-inflammatory cytokine IL-10, illustrating the detrimental consequence of immune maladaptation during carcinogenesis. Indeed, increased amounts of genotoxic pks+ E. coli were observed in the intestine of IL-10-knockout mice, and tumour multiplicity was dependent on unrestrained inflammation92,93. IL-10 deficiency is also reported to coincide with increased abundance of the Bacteroides, Prevotella and Rikenella genera94. The combined inhibition of IL-10 and the bacterial scavenger protein lipocalin in vivo leads to the colonic outgrowth of Alistipes spp., a microorganism that induces spontaneous colitis-associated cancer in an IL-6-dependent manner95 and is over-represented in cohorts of patients with CRC96. Importantly, the microbial community can feed back to alter levels of IL-10, for example through the production of butyrate. Signalling downstream of butyrate-specific receptor GPR109A suppressed inflammatory-driven carcinogenesis and enabled macrophages and dendritic cells to skew CD4+ T cells into IL-10-producing T regulatory cells66. Additionally, Clostridium butyricum, a butyrate-producing strain that is used clinically for its anti-inflammatory effects in IBD, can supress IEC proliferation in tumour-bearing ApcMin/+ mice by decreasing expression of the Wnt pathway component β-catenin97. Like IL-10, a lack of IL-33 also enhances susceptibility to inflammation-driven colon cancer. Genetic ablation of IL-33 coincides with an early increase of pro-inflammatory IL-1α, preceding a later wave of inflammation driven by IL-6, IL-17 and IL-1β in the mouse colon. In vivo, IL-33 triggered production of immunoglobulin (Ig)A,

Inflammation-driven bacterial niche Studies of IBD91—and now the CRC models discussed below—have illustrated that uncontrolled inflammation creates a different ecological niche within the colon, leading to the outgrowth of bacterial species that are better adapted. Although this is typically regarded as detrimental, such a nuanced process may be evolutionarily important. When studying putatively harmful species, it should therefore be considered that the shift towards a microbiota that has adapted to cope with the new harsh environment might be critical for host regeneration. 512 | Nature | Vol 585 | 24 September 2020

which counteracted a dysbiosis characterized by potentially harmful mucus-degrading Akkermansia species98. Notably, mice deficient in mucin 2—which lack an intact mucus barrier—spontaneously develop inflammation-driven CRC99. Low-fibre diets have also been shown to promote the expansion of mucus-degrading bacteria that can cause erosion of the intestinal barrier100. IL-18 prevents spontaneous colitis in Il10−/− mice, at least in part by inhibiting the colonization of Akkermansia muciniphila101. In support of these data, Il18−/− mice display both enhanced dextran sodium sulfate (DSS)-induced colitis102 and exacerbated inflammation-induced CRC79,103. Notably, co-housing IL-18-deficient and wild-type mice to enable microbiota transfer led to an increased incidence of inflammation-induced CRC in the wild-type mice in comparison with those that were single-housed79; further work will be required to determine the duration of this effect and the severity of the resulting disease. By contrast, other studies report that IL-18 can promote DSS-induced colitis, possibly by inhibiting the maturation of mucus-producing goblet cells102. Whether these seemingly contradictory data arise as a result of taking measurements at different time points, or from differences in doses or housing facilities, is unknown.

‘Oncomicrobes’ alter immune composition Tumours create a permissive tumour microenvironment by recruiting certain immune cells22, the density and composition of which can be skewed by the microbiota. For example, levels of the lymphocyte-attracting chemokines CCL5, CCL20 and CXCL11 correlate with members of the Bacteroidetes and Firmicutes phyla and can be induced in vitro by F. nucleatum and E. coli104. However, whether these chemoattractants recruit protective or pathogenic T cell subsets is unclear. Such decisions may be affected by the genetic landscape, because a distinct microbiota correlated with cytotoxic CD8+ T cell and CD4+ TH1 responses specifically in mice that lacked the autophagy-regulating gene Atg7 in IECs105. Altering the microbiota to skew towards a ‘hot’ tumour microenvironment, which is often characterized by infiltrating CD8+ T cells and beneficial patient outcomes, is therefore an intriguing therapeutic avenue. However, care must be taken, because a recent study has shown that CD8+ T cells are not always anti-tumorigenic. A distinct mouse microbiota, characterized by an increased abundance of Prevotellaceae, correlated with a higher tumour burden and a skewing towards exhausted instead of IFNγ-producing intratumoral CD8+ T cells106. Although correlative, this study highlights the importance of characterizing the activation status of cells alongside their abundance, as dysfunctional T cells may aid tumour evasion107. ETBF was found to induce a pro-carcinogenic TH17 response driven by STAT3 activation in ApcMin/+ mice80. This was accompanied by the recruitment and differentiation of iNoshigh monocytic-like myeloid-derived suppressor cells (M-MDSCs), with Nos2 upregulation driven by intratumoral IL-17A binding to IL-17R+ myeloid cells. These M-MDSCs supressed the activity of cytotoxic CD8+ T cells while inducing the expression of genes involved in tumour growth (Mmp9) and angiogenesis (Vegfa)108. Notably, ETBF caused MDSC accumulation specifically in the distal colon, where locally restricted NF-κb signalling in the epithelium triggered confined production of the myeloid-recruiting chemokines CXCL1 and CXCL2109. That study is one of few taking into consideration tumour location, an important variable given evidence for distinct microbial and immune niches throughout the colon110 and location-specific effects of the microbiota in preclinical CRC models43. In humans, ETBF similarly induced the expression and secretion of the CXCL1-orthologue IL-8 from epithelial cells in an NF-κb-dependent manner111,112. Similar to ETBF, oral gavage with the CRC-associated bacterium Streptococcus gallolyticus in the DSS-induced mouse model of CRC led to an increased tumour burden, the selective recruitment of immunosuppressive CD11b+ myeloid populations and increased levels of

myeloid-derived cytokines, including IL-6 and IL-8113. A greater tumour load in ApcMin/+ mice fed with F. nucleatum was also accompanied by an increased number of immunosuppressive intratumoral myeloid cells—characterized as mononuclear and granulocytic  MDSCs, tumour-associated macrophages, tumour-associated neutrophils and dendritic cells. More functionally, the MDSCs and tumour-associated macrophages isolated from mouse tumours suppressed the proliferation of CD4+ T cells in ex vivo co-culture experiments114. The CRC-associated bacterium Peptostreptococcus anaerobius has also been reported to trigger an expansion of pro-tumour myeloid populations in ApcMin/+ mice, by selectively adhering to colon cancer cells and inducing NF-κb. A simultaneous increase of T regulatory cells, TH17 and cytotoxic CD8+ T cell frequencies was also observed115. F. nucleatum can additionally act in a cytokine-independent manner and directly inhibit natural killer (NK)-cell cytotoxicity, enabling the tumour to evade immune destruction. Pre-treating CRC cell lines with F. nucleatum reduced the activity of co-cultured NK cells via binding of the receptor TIGIT to the F. nucleatum-derived protein Fap2116. The ability of distinct species to drive similar downstream effector functions, including NF-κB activation, could help to explain heterogeneity in the disease-associated microbiome. Therefore, targeting the consequences of dysbiosis, such as the recruitment of certain immunosuppressive populations, might be more therapeutically efficacious than inhibiting specific bacteria. However, caution must be taken given the pleiotropy of immune-derived factors and the fact that the ultimate outcome depends on the density, activation status and localization of different cells.

Technologies to investigate microbiome causality Despite substantial progress in the field, there is a lack of evidence for therapeutically tractable causal interactions between the altered microbiota and host. Compiling a complete CRC-associated microbiome that includes the abundance and function of bacteria, fungi, viruses, archaea and metabolites—as well as considering the multi-layer crosstalk between these factors and the host—is a daunting task that is yet to be tackled. This is in part due to limitations in sequencing depths, in the ability to culture certain species, and of models to recapitulate the human tumour microenvironment—issues that technological advances are addressing (Fig. 3). Organoids are three-dimensional culture systems of epithelial tissue that enable the stable in vitro culture of, for example, patient-derived tumours117. By using this technique, the effect of various stimuli can be tested, but poor accessibility to the enclosed apical surface of organoids renders culture with microbiota components challenging118. However, the development of intra-organoid microinjections of bacteria119 and reversing the epithelial polarity118 have begun to solve this issue. Indeed, injection of pks+ E. coli into human organoids revealed the capacity of this bacterium to directly cause clinically relevant CRC-driver mutations46. Despite the value of such reductionist approaches, it is important to know the capacity of oncomicrobes to act on the genetic landscape and in the context of other cells from the tumour microenvironment. To this end, organoid experiments can be expanded to co-culture with immune and mesenchymal cells117, while targeted mutation combinations are introduced using CRISPR–Cas9117. The effect of such manipulations on post-translational modifications can also now be determined using mass cytometry120. To further account for the fluidic and mechanistic properties of an organ, a microfluidic in vitro system termed ‘organ-on-a-chip’ was developed that utilizes specific channel permeabilities to integrate several cellular elements121. Such a microfluidic model, termed ‘HuMIX’, has provided functional insight into CRC-related metabolites through the observation that simultaneous co-culture of non-digestible nutrients, live microorganisms and human epithelial cells was required to generate a distinct SCFA ratio that limited the self-renewal capacity of IECs122. Nature | Vol 585 | 24 September 2020 | 513

Review Longitudinal international cohorts (‘Multi-omic’ approaches, host genetic traits, environmental factors)

GEMMs Gnotobiotic/antibiotics ‘Wilding’ of models

Microbiota

Organoids (± CRISPR–CAS9) Microinjections (microorganisms, metabolites) Co-culture systems (cells, soluble mediators) Organ-on-a-chip

Cancer cells Soluble mediators Immune cells, stromal cells

• Biomarkers • Diagnostics • Therapies Fig. 3 | Approaches to advance the translation of microbiome-based therapeutics in CRC. Combining ‘multi-omic’ analysis of samples taken from large, international patient cohorts (left) with improved mechanistic in vitro (right) and in vivo (middle) approaches aims to distinguish between correlative or causative observations. In vitro approaches that have shown recent improvements include co-culturing patient-derived organoids with immune and stromal cells, intra-organoid injections of microorganisms, reversing

organoid epithelial cell polarity and microfluidic ‘organs-on-a-chip’. Patient-derived and CRISPR–Cas9-edited organoids can be injected into the rectum of recipient mice to expedite in vivo tumour formation in the context of the tumour microenvironment. These and other GEMMs of CRC enable reductionist approaches to identify targetable host–microbe interactions through longitudinal sequencing, antibiotic, gnotobiotic, faecal transplantation and ‘wilding’ experiments.

Importantly, this technology has been extended to allow for modelling of anaerobic and aerobic microorganisms while maintaining viable IECs123. Such advances will also facilitate the study of intra-microbe competition for essential nutrients, which may underpin outgrowths of oncomicrobes. Genetically engineered mouse models (GEMMs) form the basis of many of the studies discussed in this Review (for reviews of the different models, see refs. 25,124); however, several such models have a long latency period and fail to recapitulate late-stage human disease. To reduce the duration of models and thereby allow for antibiotic treatment without the occurrence of bacterial resistance, tumorigenic organoids can be locally injected into the rectum of experimental animals117,125. Ultimately, however, gnotobiotic GEMMs will be indispensable for deciphering—in a reductionist in vivo approach—the complexity of microbiota–host interactions and proving causality in CRC. Humanizing mice through faecal transplantation27, or using ‘wilding’ mice that have a more natural microbiota126, will aid the study of inter-individual microbiota heterogeneity and help to tackle some issues with translatability127. To study late-stage CRC, metastatic models have been developed and provide a much-needed tool to investigate potential roles for the microbiota in modulating the epithelial–mesenchymal transition128. Further GEMM models are under development, and will hopefully help to resolve questions such as whether F. nucleatum—associated with a primary tumour and identified in matched metastasis sites129—drives metastasis or is merely a passenger, and whether this phenomenon applies to other species. Experimental vigilance is essential to ensure the reproducibility of findings, particularly because the microbiota can be affected by the animals’ diet, the cages in which they are housed and the specific facility in which they are reared130. Although antibiotics are a useful tool for modulating the microbiota during disease progression, they often target certain groups of microorganisms without adequate control or guarantee of eradication, making data interpretation challenging. For example, in the DSS–azoxymethane model of cancer, enhanced tumorigenesis was observed in germ-free mice131 but the opposite reported when the microbiota was depleted by treatment with antibiotics82. Such seemingly contrasting data could be due to an outgrowth of uncharacterized and antibiotic-resistant species in the colon, differences in the time points

at which the microorganisms were eliminated, or differences in—for example—the immune system of germ-free mice132. Beyond bacteria and metabolites, there is emerging evidence of fungal133, archaeal134 and viral135 changes in CRC, such as an increased abundance of fungi from the genus Malassezia30. Moreover, induction of IL-18 by fungal commensals was recently shown to inhibit colitis-associated CRC in mice, thus highlighting a protective role136. Particular consideration should also be given to the specific strains and culture techniques used, owing to variable intra-strain virulence potential137,138 and the stark differences in cancer-related gene expression that are observed when comparing viable and heat-inactivated F. nucleatum139. Moving forwards, precise characterization of the numerous species within a genus is needed to uncover the full functional diversity of the microbiota. Such knowledge may help to explain apparent contradictions, such as certain species of the putatively protective Clostridia genus producing CRC-associated secondary bile acids140. For such integrated analysis pipelines, broad and systematic ‘multi-omic’ approaches will be paramount to building an interactome141,142. These could also include spatial analysis using in situ imaging alongside single-cell sequencing and in silico computation modelling approaches that mathematically reconstruct the metabolome143–145. Another emerging area is the analysis of the epigenetic landscape in the context of the microbiome146, such as Fusobacterium correlating with distinctive methylation patterns in CRC tumours147. Finally, given extensive inter-individual heterogeneity and limitations within model systems, large international patient cohorts that comprise integrated data on the microbiota and genetic traits of the host will be crucial in order to pinpoint clinically relevant findings.

514 | Nature | Vol 585 | 24 September 2020

Future outlook Rapidly accumulating studies over the past decade have detected differences in the composition of the intestinal microbiota in individuals with early-stage pre-cancerous lesions ranging to those with metastatic CRC. Observations such as the increased abundance of F. nucleatum, E. coli or B. fragilis in CRC have created a new dimension of oncology, and such microorganisms could be incorporated as biomarkers alongside

routine diagnostics, as has been reviewed recently30,148. Therapeutics that target the microbiota are also emerging—including selective elimination of oncomicrobes, faecal transplantation of anti-tumorigenic species and skewing the microbiota by oral supplementations or targeted bacteriophage therapy30,149,150. However, careful consideration of any therapeutic avenue is required owing to the numerous effects of the microbiome on the biology of the host, many of which are currently unknown. To uncover the desired clinical potential of the microbiome in CRC, we must continue to distinguish correlation and causation by systematically adding knowledge to the layers of complexity. For example, CRC-associated mutations are known to affect the way in which cells respond to stimuli—such as a mutation in KRAS that increases responsiveness to IL-22151—but whether such mutations affect host colonization, dysbiosis and intercellular communications is unclear. Importantly, we must decipher whether characteristics of specific strains or the collective capabilities of an ‘oncogenic’ microbiota— which encompasses bacteria, fungi, viruses, archaea and their produced metabolites—alter disease. Owing to the vast intra-individual heterogeneity in the microbiota, it is plausible that distinct strains in different individuals will trigger a similar pathology through common pathways. Therefore, targeting wide-spanning functions, rather than specific taxa, might be the most effective strategy152. Finally, although we have focused on the microbiome in CRC, many of the underlying mechanisms that we have discussed could be relevant to other diseases such as IBD, in which certain bacteria—including E. coli and B. fragilis— are similarly over-represented153.

Conclusion Various mechanisms have been revealed that begin to explain how elements of the microbiota modulate tumorigenesis, broadly ranging from alteration of the intestinal barrier and/or immune landscape to reshaping the colonic ecological niche and the provision of genotoxic insults (Fig. 1). Although host maladaptation and consequential barrier breakdown is most probably critical in order to facilitate close proximity between the host and the microbiota, longitudinal multi-omics are required to decipher the chronology of these ‘phases’ in tumour development. To provide answers to the many open questions, an in-depth understanding of how the microbiota mediates its effects in the context of the tumour microenvironment—whether through direct effects on DNA damage and inflammation or through other host-derived mechanisms—is required. Fortunately, technological advances provide us with revised tools to study the microbiota in the context of increasingly physiological CRC model systems, in order to decipher the challenging complexity of a colonic tumour microenvironment. 1.

Keum, N. & Giovannucci, E. Global burden of colorectal cancer: emerging trends, risk factors and prevention strategies. Nat. Rev. Gastroenterol. Hepatol. 16, 713–732 (2019). 2. GBD 2017 Colorectal Cancer Collaborators. The global, regional, and national burden of colorectal cancer and its attributable risk factors in 195 countries and territories, 1990–2017: a systematic analysis for the Global Burden of Disease Study 2017. Lancet Gastroenterol. Hepatol. 4, 913–933 (2019). 3. Lloyd-Price, J., Abu-Ali, G. & Huttenhower, C. The healthy human microbiome. Genome Med. 8, 51 (2016). 4. Garrett, W. S., Gordon, J. I. & Glimcher, L. H. Homeostasis and inflammation in the intestine. Cell 140, 859–870 (2010). 5. Belkaid, Y. & Hand, T. W. Role of the microbiota in immunity and inflammation. Cell 157, 121–141 (2014). 6. Peterson, L. W. & Artis, D. Intestinal epithelial cells: regulators of barrier function and immune homeostasis. Nat. Rev. Immunol. 14, 141–153 (2014). 7. Gehart, H. & Clevers, H. Tales from the crypt: new insights into intestinal stem cells. Nat. Rev. Gastroenterol. Hepatol. 16, 19–34 (2019). 8. Zhou, L. & Sonnenberg, G. F. Essential immunologic orchestrators of intestinal homeostasis. Sci. Immunol. 3, eaao1605 (2018). 9. Karin, M. & Clevers, H. Reparative inflammation takes charge of tissue regeneration. Nature 529, 307–315 (2016). 10. Kinchen, J., et al. Structural remodeling of the human colonic mesenchyme in inflammatory bowel disease. Cell 175, 372–386.e17 (2018). 11. Dvorak, H. F. Tumors: wounds that do not heal. Similarities between tumor stroma generation and wound healing. N. Engl. J. Med. 315, 1650–1659 (1986).

12. Jess, T., Rungoe, C. & Peyrin-Biroulet, L. Risk of colorectal cancer in patients with ulcerative colitis: a meta-analysis of population-based cohort studies. Clin. Gastroenterol. Hepatol. 10, 639–645 (2012). 13. Bogaert, J. & Prenen, H. Molecular genetics of colorectal cancer. Ann. Gastroenterol. 27, 9–14 (2014). 14. Wu, S., Powers, S., Zhu, W. & Hannun, Y. A. Substantial contribution of extrinsic risk factors to cancer development. Nature 529, 43–47 (2016). 15. Nomura, A. et al. Helicobacter pylori infection and gastric carcinoma among Japanese Americans in Hawaii. N. Engl. J. Med. 325, 1132–1136 (1991). 16. Parsonnet, J. et al. Helicobacter pylori infection and the risk of gastric carcinoma. N. Engl. J. Med. 325, 1127–1131 (1991). 17. Elinav, E. et al. Inflammation-induced cancer: crosstalk between tumours, immune cells and microorganisms. Nat. Rev. Cancer 13, 759–771 (2013). 18. Grivennikov, S. I., Greten, F. R. & Karin, M. Immunity, inflammation, and cancer. Cell 140, 883–899 (2010). 19. Punt, C. J., Koopman, M. & Vermeulen, L. From tumour heterogeneity to advances in precision treatment of colorectal cancer. Nat. Rev. Clin. Oncol. 14, 235–246 (2017). 20. Hanahan, D. & Weinberg, R. A. Hallmarks of cancer: the next generation. Cell 144, 646–674 (2011). 21. Galon, J. et al. Type, density, and location of immune cells within human colorectal tumors predict clinical outcome. Science 313, 1960–1964 (2006). 22. Galon, J. & Bruni, D. Tumor immunology and tumor evolution: intertwined histories. Immunity 52, 55–81 (2020). 23. Canli, O. et al. Myeloid cell-derived reactive oxygen species induce epithelial mutagenesis. Cancer Cell 32, 869–883.e5 (2017). 24. Sahai, E. et al. A framework for advancing our understanding of cancer-associated fibroblasts. Nat. Rev. Cancer 20, 174–186 (2020). 25. West, N. R., McCuaig, S., Franchini, F. & Powrie, F. Emerging cytokine networks in colorectal cancer. Nat. Rev. Immunol. 15, 615–629 (2015). 26. Reddy, B. S., Weisburger, J. H., Narisawa, T. & Wynder, E. L. Colon carcinogenesis in germ-free rats with 1,2-dimethylhydrazine and N-methyl-N′-nitro-N-nitrosoguanidine. Cancer Res. 34, 2368–2372 (1974). 27. Wong, S. H. et al. Gavage of fecal samples from patients with colorectal cancer promotes intestinal carcinogenesis in germ-free and conventional mice. Gastroenterology 153, 1621–1633.e6 (2017). Faecal transplantation of stools from patients with CRC into mice illustrates a carcinogenic role of CRC-associated microbiota. 28. Olesen, S. W. & Alm, E. J. Dysbiosis is not an answer. Nat. Microbiol. 1, 16228 (2016). 29. Saus, E., Iraola-Guzmán, S., Willis, J. R., Brunet-Vega, A. & Gabaldón, T. Microbiome and colorectal cancer: roles in carcinogenesis and clinical potential. Mol. Aspects Med. 69, 93–106 (2019). 30. Wong, S. H. & Yu, J. Gut microbiota in colorectal cancer: mechanisms of action and clinical applications. Nat. Rev. Gastroenterol. Hepatol. 16, 690–704 (2019). 31. Tilg, H., Adolph, T. E., Gerner, R. R. & Moschen, A. R. The intestinal microbiota in colorectal cancer. Cancer Cell 33, 954–964 (2018). 32. Garrett, W. S. The gut microbiota and colon cancer. Science 364, 1133–1135 (2019). 33. Ternes, D. et al. Microbiome in colorectal cancer: how to get from meta-omics to mechanism? Trends Microbiol. 28, 401–423 (2020). 34. Wirbel, J. et al. Meta-analysis of fecal metagenomes reveals global microbial signatures that are specific for colorectal cancer. Nat. Med. 25, 679–689 (2019). 35. Yan, Y. et al. Structure of the mucosal and stool microbiome in Lynch syndrome. Cell Host Microbe 27, 585–600.e4 (2020). 36. Touati, E. et al. Chronic Helicobacter pylori infections induce gastric mutations in mice. Gastroenterology 124, 1408–1419 (2003). 37. Cuevas-Ramos, G. et al. Escherichia coli induces DNA damage in vivo and triggers genomic instability in mammalian cells. Proc. Natl Acad. Sci. USA 107, 11537–11542 (2010). 38. Nougayrède, J. P. et al. Escherichia coli induces DNA double-strand breaks in eukaryotic cells. Science 313, 848–851 (2006). 39. Cougnoux, A. et al. Bacterial genotoxin colibactin promotes colon tumour growth by inducing a senescence-associated secretory phenotype. Gut 63, 1932–1942 (2014). 40. Goodwin, A. C. et al. Polyamine catabolism contributes to enterotoxigenic Bacteroides fragilis-induced colon tumorigenesis. Proc. Natl Acad. Sci. USA 108, 15354–15359 (2011). 41. Wang, X. et al. Enterococcus faecalis induces aneuploidy and tetraploidy in colonic epithelial cells through a bystander effect. Cancer Res. 68, 9909–9917 (2008). 42. He, Z. et al. Campylobacter jejuni promotes colorectal tumorigenesis through the action of cytolethal distending toxin. Gut 68, 289–300 (2019). 43. Tomkovich, S. et al. Locoregional effects of microbiota in a preclinical model of colon carcinogenesis. Cancer Res. 77, 2620–2632 (2017). 44. Cougnoux, A. et al. Small-molecule inhibitors prevent the genotoxic and protumoural effects induced by colibactin-producing bacteria. Gut 65, 278–285 (2016). 45. Lucas, C. et al. Autophagy of intestinal epithelial cells inhibits colorectal carcinogenesis induced by colibactin-producing Escherichia coli in ApcMin/+ mice. Gastroenterology 158, 1373–1388 (2020). 46. Pleguezuelos-Manzano, C. et al. Mutational signature in colorectal cancer caused by genotoxic pks+ E. coli. Nature 580, 269–273 (2020). This pioneering study provides direct evidence of causality between pks+ E. coli and established CRC driver mutations. 47. Wang, X., Yang, Y. & Huycke, M. M. Commensal bacteria drive endogenous transformation and tumour stem cell marker expression through a bystander effect. Gut 64, 459–468 (2015). 48. Irrazabal, T. et al. Limiting oxidative DNA damage reduces microbe-induced colitis-associated colorectal cancer. Nat. Commun. 11, 1802 (2020). 49. Veziant, J. et al. Association of colorectal cancer with pathogenic Escherichia coli: focus on mechanisms using optical imaging. World J. Clin. Oncol. 7, 293–301 (2016). 50. Zhu, W. et al. Editing of the gut microbiota reduces carcinogenesis in mouse models of colitis-associated colorectal cancer. J. Exp. Med. 216, 2378–2393 (2019).

Nature | Vol 585 | 24 September 2020 | 515

Review 51. Gaines, S. et al. Western diet promotes intestinal colonization by collagenolytic microbes and promotes tumor formation after colorectal surgery. Gastroenterology, 158, 958–970 (2020). 52. Louis, P., Hold, G. L. & Flint, H. J. The gut microbiota, bacterial metabolites and colorectal cancer. Nat. Rev. Microbiol. 12, 661–672 (2014). 53. De Almeida, C. V., de Camargo, M. R., Russo, E. & Amedei, A. Role of diet and gut microbiota on colorectal cancer immunomodulation. World J. Gastroenterol. 25, 151–162 (2019). 54. Nguyen, L. H. et al. Association between sulfur-metabolizing bacterial communities in stool and risk of distal colorectal cancer in men. Gastroenterology 158, 1313–1325 (2020). 55. Yachida, S. et al. Metagenomic and metabolomic analyses reveal distinct stage-specific phenotypes of the gut microbiota in colorectal cancer. Nat. Med. 25, 968–976 (2019). 56. Long, S. et al. Metaproteomics characterizes human gut microbiome function in colorectal cancer. NPJ Biofilms Microbiomes 6, 14 (2020). 57. Kawalek, J. C., Hallmark, R. K. & Andrews, A. W. Effect of lithocholic acid on the mutagenicity of some substituted aromatic amines. J. Natl Cancer Inst. 71, 293–298 (1983). 58. Shibuya, N. et al. Co-mutagenicity of glyco- and tauro-deoxycholic acids in the Ames test. Mutat. Res. 395, 1–7 (1997). 59. Lavoie, S. et al. Expression of FFAR2 by dendritic cells prevents their expression of IL27 and is required for maintenance of mucosal barrier and immune response against colorectal tumors in mice. Gastroenterology 158, 1359–1372.e9 (2020). 60. Liu, T. et al. High-fat diet-induced dysbiosis mediates MCP-1/CCR2 axis-dependent M2 macrophage polarization and promotes intestinal adenoma-adenocarcinoma sequence. J. Cell. Mol. Med. 24, 2648–2662 (2020). 61. Denison, M. S. & Nagy, S. R. Activation of the aryl hydrocarbon receptor by structurally diverse exogenous and endogenous chemicals. Annu. Rev. Pharmacol. Toxicol. 43, 309–334 (2003). 62. Metidji, A. et al. The environmental sensor AHR protects from inflammatory damage by maintaining intestinal stem cell homeostasis and barrier integrity. Immunity 49, 353–362 (2018). 63. Chang, P. V., Hao, L., Offermanns, S. & Medzhitov, R. The microbial metabolite butyrate regulates intestinal macrophage function via histone deacetylase inhibition. Proc. Natl Acad. Sci. USA 111, 2247–2252 (2014). 64. Schulthess, J. et al. The short chain fatty acid butyrate imprints an antimicrobial program in macrophages. Immunity 50, 432–445.e7 (2019). 65. Zagato, E. et al. Endogenous murine microbiota member Faecalibaculum rodentium and its human homologue protect from intestinal tumour growth. Nat. Microbiol. 5, 511–524 (2020). 66. Singh, N. et al. Activation of Gpr109a, receptor for niacin and the commensal metabolite butyrate, suppresses colonic inflammation and carcinogenesis. Immunity 40, 128–139 (2014). 67. Donohoe, D. R. et al. A gnotobiotic mouse model demonstrates that dietary fiber protects against colorectal tumorigenesis in a microbiota- and butyrate-dependent manner. Cancer Discov. 4, 1387–1397 (2014). 68. Kaiko, G. E. et al. The colonic crypt protects stem cells from microbiota-derived metabolites. Cell 165, 1708–1720 (2016). 69. Belcheva, A. et al. Gut microbial metabolism drives transformation of MSH2-deficient colon epithelial cells. Cell 158, 288–299 (2014). 70. Uronis, J. M. et al. Modulation of the intestinal microbiota alters colitis-associated colorectal cancer susceptibility. PLoS ONE 4, e6026 (2009). 71. Couturier-Maillard, A. et al. NOD2-mediated dysbiosis predisposes mice to transmissible colitis and colorectal cancer. J. Clin. Invest. 123, 700–711 (2013). 72. Chen, G. Y., Shaw, M. H., Redondo, G. & Núñez, G. The innate immune receptor Nod1 protects the intestine from inflammation-induced tumorigenesis. Cancer Res. 68, 10060–10067 (2008). 73. Grivennikov, S. I. et al. Adenoma-linked barrier defects and microbial products drive IL-23/IL-17-mediated tumour growth. Nature 491, 254–258 (2012). 74. Dejea, C. M. et al. Microbiota organization is a distinct feature of proximal colorectal cancers. Proc. Natl Acad. Sci. USA 111, 18321–18326 (2014). 75. Drewes, J. L. et al. High-resolution bacterial 16S rRNA gene profile meta-analysis and biofilm status reveal common colorectal cancer consortia. NPJ Biofilms Microbiomes 3, 34 (2017). 76. Dejea, C. M. et al. Patients with familial adenomatous polyposis harbor colonic biofilms containing tumorigenic bacteria. Science 359, 592–597 (2018). 77. Tomkovich, S. et al. Human colon mucosal biofilms from healthy or colon cancer hosts are carcinogenic. J. Clin. Invest. 129, 1699–1712 (2019). 78. Domingue, J. C., Drewes, J. L., Merlo, C. A., Housseau, F. & Sears, C. L. Host responses to mucosal biofilms in the lung and gut. Mucosal Immunol. 13, 413–422 (2020). 79. Hu, B. et al. Microbiota-induced activation of epithelial IL-6 signaling links inflammasome-driven inflammation with transmissible cancer. Proc. Natl Acad. Sci. USA 110, 9862–9867 (2013). 80. Wu, S. et al. A human colonic commensal promotes colon tumorigenesis via activation of T helper type 17 T cell responses. Nat. Med. 15, 1016–1022 (2009). 81. Wu, P. et al. γδT17 cells promote the accumulation and expansion of myeloid-derived suppressor cells in human colorectal cancer. Immunity 40, 785–800 (2014). 82. Zackular, J. P. et al. The gut microbiome modulates colon tumorigenesis. MBio 4, e0069213 (2013). 83. Ghoreschi, K. et al. Generation of pathogenic TH17 cells in the absence of TGF-β signalling. Nature 467, 967–971 (2010). 84. Dmitrieva-Posocco, O. et al. Cell-type-specific responses to interleukin-1 control microbial invasion and tumor-elicited inflammation in colorectal cancer. Immunity 50, 166–180.e7 (2019). Using mouse models, this study highlights how cell-type-specific cytokine responses differentially influence tumorigenesis.

516 | Nature | Vol 585 | 24 September 2020

85. Zelante, T. et al. Tryptophan catabolites from microbiota engage aryl hydrocarbon receptor and balance mucosal reactivity via interleukin-22. Immunity 39, 372–385 (2013). 86. Busbee, P. B. et al. Indole-3-carbinol prevents colitis and associated microbial dysbiosis in an IL-22-dependent manner. JCI Insight 5, e127551 (2020). 87. Kirchberger, S. et al. Innate lymphoid cells sustain colon cancer through production of interleukin-22 in a mouse model. J. Exp. Med. 210, 917–931 (2013). 88. Hernandez, P., Gronke, K. & Diefenbach, A. A catch-22: interleukin-22 and cancer. Eur. J. Immunol. 48, 15–31 (2018). 89. Gronke, K. et al. Interleukin-22 protects intestinal stem cells against genotoxic stress. Nature 566, 249–253 (2019). This study shows how dietary stimuli mediates IL-22 signalling, which can in turn protect stem cells from genotoxic stress. 90. Perez, L. G. et al. TGF-β signaling in TH17 cells promotes IL-22 production and colitis-associated colon cancer. Nat. Commun. 11, 2608 (2020). 91. Lloyd-Price, J. et al. Multi-omics of the gut microbial ecosystem in inflammatory bowel diseases. Nature 569, 655–662 (2019). 92. Arthur, J. C. et al. Intestinal inflammation targets cancer-inducing activity of the microbiota. Science 338, 120–123 (2012). 93. Arthur, J. C. et al. Microbial genomic analysis reveals the essential role of inflammation in bacteria-induced colorectal cancer. Nat. Commun. 5, 4724 (2014). 94. Dennis, K. L. et al. Adenomatous polyps are driven by microbe-instigated focal inflammation and are controlled by IL-10-producing T cells. Cancer Res. 73, 5905–5913 (2013). 95. Moschen, A. R. et al. Lipocalin 2 protects from inflammation and tumorigenesis associated with gut microbiota alterations. Cell Host Microbe 19, 455–469 (2016). 96. Feng, Q. et al. Gut microbiome development along the colorectal adenoma-carcinoma sequence. Nat. Commun. 6, 6528 (2015). 97. Chen, D. et al. Clostridium butyricum, a butyrate-producing probiotic, inhibits intestinal tumor development through modulating Wnt signaling and gut microbiota. Cancer Lett. 469, 456–467 (2020). 98. Malik, A. et al. IL-33 regulates the IgA-microbiota axis to restrain IL-1α-dependent colitis and tumorigenesis. J. Clin. Invest. 126, 4469–4481 (2016). 99. Velcich, A. et al. Colorectal cancer in mice genetically deficient in the mucin Muc2. Science 295, 1726–1729 (2002). 100. Desai, M. S. et al. A dietary fiber-deprived gut microbiota degrades the colonic mucus barrier and enhances pathogen susceptibility. Cell 167, 1339–1353.e21 (2016). Using a gnotobiotic mouse model, this study correlates a low-fibre diet to the expansion of a mucus-degrading bacteria and aggressive colitis. 101. Seregin, S. S. et al. NLRP6 protects Il10−/− mice from colitis by limiting colonization of Akkermansia muciniphila. Cell Rep. 19, 733–745 (2017). 102. Nowarski, R. et al. Epithelial IL-18 equilibrium controls barrier function in colitis. Cell 163, 1444–1456 (2015). 103. Salcedo, R. et al. MyD88-mediated signaling prevents development of adenocarcinomas of the colon: role of interleukin 18. J. Exp. Med. 207, 1625–1636 (2010). 104. Cremonesi, E. et al. Gut microbiota modulate T cell trafficking into human colorectal cancer. Gut 67, 1984–1994 (2018). 105. Lévy, J. et al. Intestinal inhibition of Atg7 prevents tumour initiation through a microbiome-influenced immune response and suppresses tumour growth. Nat. Cell Biol. 17, 1062–1073 (2015). 106. Yu, A. I. et al. Gut microbiota modulate CD8 T cell responses to influence colitis-associated tumorigenesis. Cell Rep. 31, 107471 (2020). 107. Thommen, D. S. & Schumacher, T. N. T cell dysfunction in cancer. Cancer Cell 33, 547–562 (2018). 108. Thiele Orberg, E. et al. The myeloid immune signature of enterotoxigenic Bacteroides fragilis-induced murine colon tumorigenesis. Mucosal Immunol. 10, 421–433 (2017). 109. Chung, L. et al. Bacteroides fragilis toxin coordinates a pro-carcinogenic inflammatory cascade via targeting of colonic epithelial cells. Cell Host Microbe 23, 203–214.e5 (2018). In this study, it is shown that enterotoxic B. fragilis creates a pro-tumorigenic environment in the distal colon by driving locally restricted chemokine expression in mice. 110. James, K. R. et al. Distinct microbial and immune niches of the human colon. Nat. Immunol. 21, 343–353 (2020). 111. Kim, J. M., Oh, Y. K., Kim, Y. J., Oh, H. B. & Cho, Y. J. Polarized secretion of CXC chemokines by human intestinal epithelial cells in response to Bacteroides fragilis enterotoxin: NF-κB plays a major role in the regulation of IL-8 expression. Clin. Exp. Immunol. 123, 421–427 (2001). 112. Sanfilippo, L. et al. Bacteroides fragilis enterotoxin induces the expression of IL-8 and transforming growth factor-beta (TGF-β) by human colonic epithelial cells. Clin. Exp. Immunol. 119, 456–463 (2000). 113. Zhang, Y., Weng, Y., Gan, H., Zhao, X. & Zhi, F. Streptococcus gallolyticus conspires myeloid cells to promote tumorigenesis of inflammatory bowel disease. Biochem. Biophys. Res. Commun. 506, 907–911 (2018). 114. Kostic, A. D. et al. Fusobacterium nucleatum potentiates intestinal tumorigenesis and modulates the tumor-immune microenvironment. Cell Host Microbe 14, 207–215 (2013). 115. Long, X. et al. Peptostreptococcus anaerobius promotes colorectal carcinogenesis and modulates tumour immunity. Nat. Microbiol. 4, 2319–2330 (2019). 116. Gur, C. et al. Binding of the Fap2 protein of Fusobacterium nucleatum to human inhibitory receptor TIGIT protects tumors from immune cell attack. Immunity 42, 344–355 (2015). 117. Tuveson, D. & Clevers, H. Cancer modeling meets human organoid technology. Science 364, 952–955 (2019). 118. Co, J. Y. et al. Controlling epithelial polarity: a human enteroid model for host–pathogen interactions. Cell Rep. 26, 2509–2520.e4 (2019). 119. Dutta, D. & Clevers, H. Organoid culture systems to study host–pathogen interactions. Curr. Opin. Immunol. 48, 15–22 (2017). 120. Qin, X. et al. Cell-type-specific signaling networks in heterocellular organoids. Nat. Methods 17, 335–342 (2020).

121. Zhang, B. et al. Advances in organ-on-a-chip engineering. Nat. Rev. Mater. 3, 257–278 (2018). 122. Greenhalgh, K. et al. Integrated in vitro and in silico modeling delineates the molecular effects of a synbiotic regimen on colorectal-cancer-derived cells. Cell Rep. 27, 1621–1632 (2019). 123. Jalili-Firoozinezhad, S. et al. A complex human gut microbiome cultured in an anaerobic intestine-on-a-chip. Nat. Biomed. Eng. 3, 520–531 (2019). 124. Romano, G., Chagani, S. & Kwong, L. N. The path to metastatic mouse models of colorectal cancer. Oncogene 37, 2481–2489 (2018). 125. Fumagalli, A. et al. A surgical orthotopic organoid transplantation approach in mice to visualize and study colorectal cancer progression. Nat. Protoc. 13, 235–247 (2018). 126. Rosshart, S. P. et al. Wild mouse gut microbiota promotes host fitness and improves disease resistance. Cell 171, 1015–1028.e13 (2017). 127. Rosshart, S. P. et al. Laboratory mice born to wild mice have natural microbiota and model human immune responses. Science 365, eaaw4361 (2019). 128. Jackstadt, R. et al. Epithelial NOTCH signaling rewires the tumor microenvironment of colorectal cancer to drive poor-prognosis subtypes and metastasis. Cancer Cell 36, 319–336.e7 (2019). 129. Bullman, S. et al. Analysis of Fusobacterium persistence and antibiotic response in colorectal cancer. Science 358, 1443–1448 (2017). 130. Laukens, D., Brinkman, B. M., Raes, J., De Vos, M. & Vandenabeele, P. Heterogeneity of the gut microbiome in mice: guidelines for optimizing experimental design. FEMS Microbiol. Rev. 40, 117–132 (2016). 131. Zhan, Y. et al. Gut microbiota protects against gastrointestinal tumorigenesis caused by epithelial injury. Cancer Res. 73, 7199–7210 (2013). 132. Kennedy, E. A., King, K. Y. & Baldridge, M. T. Mouse microbiota models: comparing germ-free mice and antibiotics treatment as tools for modifying gut bacteria. Front. Physiol. 9, 1534 (2018). 133. Coker, O. O. et al. Enteric fungal microbiota dysbiosis and ecological alterations in colorectal cancer. Gut 68, 654–662 (2019). 134. Coker, O. O., Wu, W. K. K., Wong, S. H., Sung, J. J. & Yu, J. Altered gut archaea composition and interaction with bacteria are associated with colorectal cancer. Gastroenterology https://doi.org/10.1053/j.gastro.2020.06.042 (2020). 135. Nakatsu, G. et al. Alterations in enteric virome are associated with colorectal cancer and survival outcomes. Gastroenterology 155, 529–541.e5 (2018). 136. Malik, A. et al. SYK–CARD9 signaling axis promotes gut fungi-mediated inflammasome activation to restrict colitis and colon cancer. Immunity 49, 515–530.e5 (2018). This study shows that fungal signalling via the Syk–CARD9 pathway promotes IL-18 maturation and CRC protection. 137. Allen-Vercoe, E., Strauss, J. & Chadee, K. Fusobacterium nucleatum: an emerging gut pathogen? Gut Microbes 2, 294–298 (2011). 138. McGuire, A. M. et al. Evolution of invasion in a diverse set of Fusobacterium species. MBio 5, e01864-14 (2014). 139. Kasper, S. H. et al. Colorectal cancer-associated anaerobic bacteria proliferate in tumor spheroids and alter the microenvironment. Sci. Rep. 10, 5321 (2020). This study shows that viable and heat-inactivated F. nucleatum induce different gene-expression profiles in three-dimensional tumour spheroids. 140. Ridlon, J. M., Kang, D. J., Hylemon, P. B. & Bajaj, J. S. Bile acids and the gut microbiome. Curr. Opin. Gastroenterol. 30, 332–338 (2014). 141. Narayanasamy, S. et al. IMP: a pipeline for reproducible reference-independent integrated metagenomic and metatranscriptomic analyses. Genome Biol. 17, 260 (2016). 142. Yimagou, E. K. et al. Full-repertoire comparison of the microscopic objects composing the human gut microbiome with sequenced and cultured communities. J. Microbiol. 58, 377–386 (2020).

143. Bauer, E. & Thiele, I. From metagenomic data to personalized in silico microbiotas: predicting dietary supplements for Crohn’s disease. NPJ Syst. Biol. Appl. 4, 27 (2018). 144. Tropini, C., Earle, K. A., Huang, K. C. & Sonnenburg, J. L. The gut microbiome: connecting spatial organization to function. Cell Host Microbe 21, 433–442 (2017). 145. Pacheco, M. P. & Sauter, T. The FASTCORE family: for the fast reconstruction of compact context-specific metabolic networks models. Methods Mol. Biol. 1716, 101–110 (2018). 146. Sobhani, I. et al. Colorectal cancer-associated microbiota contributes to oncogenic epigenetic signatures. Proc. Natl Acad. Sci. USA 116, 24285–24295 (2019). 147. Wang, Q. et al. Multi-omic profiling reveals associations between the gut mucosal microbiome, the metabolome, and host DNA methylation associated gene expression in patients with colorectal cancer. BMC Microbiol. 20, 83 (2020). 148. Kim, D. J. et al. Colorectal cancer diagnostic model utilizing metagenomic and metabolomic data of stool microbial extracellular vesicles. Sci. Rep. 10, 2860 (2020). 149. Zheng, D. W. et al. Phage-guided modulation of the gut microbiota of mouse models of colorectal cancer augments their responses to chemotherapy. Nat. Biomed. Eng. 3, 717–728 (2019). 150. Fong, W., Li, Q. & Yu, J. Gut microbiota modulation: a novel strategy for prevention and treatment of colorectal cancer. Oncogene 39, 4925–4943 (2020). 151. McCuaig, S. et al. The interleukin 22 pathway interacts with mutant KRAS to promote poor prognosis in colon cancer. Clin. Cancer Res. https://doi.org/10.1158/1078-0432.CCR-191086 (2020). 152. Heintz-Buschart, A. & Wilmes, P. Human gut microbiome: function matters. Trends Microbiol. 26, 563–574 (2018). 153. Schirmer, M., Garner, A., Vlamakis, H. & Xavier, R. J. Microbial genes and pathways in inflammatory bowel disease. Nat. Rev. Microbiol. 17, 497–511 (2019). Acknowledgements We thanks N. Ilott for conceiving and generating Fig. 1, as well as M. Friedrich, T. Griseri, S. Leedham, C. Pearson and M. Pohin for their input after reading the manuscript. This work was supported by the Cancer Research UK (CRUK) grant OPTIMISTICC (C10674/A27140). A.J. has received support from the Oxford–Medical Research Council Doctoral Training Partnership (MRC DTP) and the Kennedy Trust for Rheumatology Research. E.H.M. is supported by an MRC Experimental Medicine Grant (MR/N02690X/1). Author contributions A.J. reviewed the literature and designed Figs. 2 and 3. A.J. and E.H.M. contributed equally to the writing of this manuscript. A.J., F.P. and E.H.M. discussed the content and edited the manuscript. Competing interests F.P. discloses the receipt of grants and research support from Roche and Janssen, and consulting fees from GSK, Genentech and Kintai Therapeutics. Additional information Correspondence and requests for materials should be addressed to F.P. Peer review information Nature thanks Nicola Segata and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Reprints and permissions information is available at http://www.nature.com/reprints. Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. © Springer Nature Limited 2020

Nature | Vol 585 | 24 September 2020 | 517

Article

Third-order nanocircuit elements for neuromorphic engineering https://doi.org/10.1038/s41586-020-2735-5

Suhas Kumar1 ✉, R. Stanley Williams2 & Ziwen Wang3

Received: 28 January 2020 Accepted: 3 August 2020 Published online: 23 September 2020 Check for updates

Current hardware approaches to biomimetic or neuromorphic artificial intelligence rely on elaborate transistor circuits to simulate biological functions. However, these can instead be more faithfully emulated by higher-order circuit elements that naturally express neuromorphic nonlinear dynamics1–4. Generating neuromorphic action potentials in a circuit element theoretically requires a minimum of third-order complexity (for example, three dynamical electrophysical processes)5, but there have been few examples of second-order neuromorphic elements, and no previous demonstration of any isolated third-order element6–8. Using both experiments and modelling, here we show how multiple electrophysical processes—including Mott transition dynamics—form a nanoscale third-order circuit element. We demonstrate simple transistorless networks of third-order elements that perform Boolean operations and find analogue solutions to a computationally hard graph-partitioning problem. This work paves a way towards very compact and densely functional neuromorphic computing primitives, and energy-efficient validation of neuroscientific models.

The exponential growth of both the volume of data and also the demand for computing, coupled with performance saturation of transistor-based computing systems, has fuelled interest in alternative computing primitives9. Neuromorphic, neuron-like or biomimetic computing may yield dramatic performance improvements over digital computation in the rapidly growing areas of identification and classification of information buried within massive datasets, and solving computationally hard problems such as viral genome sequencing. However, the efficient hardware implementation of brain-inspired and neural network algorithms is still a major challenge9,10. Communication and processing of data via neuromorphic dynamics is a key goal for brain-inspired computers11, but a single electronic component that can mimic a neuron does not exist. Periodic spiking requires only second-order complexity, but full neuromorphic action-potential functionality (including phasic and periodic spiking, bursting, self-sustained oscillations, chaos and sub/super-threshold active dynamics) requires a minimum of third-order complexity (three state variables or equivalent)5. There have been few successful second-order elements (exhibiting only periodic spiking and oscillations)3,8,12, and the efforts aimed at achieving extended neuromorphic properties used circuits with multiple elements13,14. There have been no previous demonstrations of an isolated third-order electronic element, neuromorphic or otherwise. Digital transistor-based chips attempt to simulate the complex equations representing the rich nonlinear dynamics of neurons, thereby making them complicated, bulky and energy-inefficient15. The design and realization of higher-order electronic elements will enable extremely efficient implementations of neuromorphic artificial intelligence. Such realizations may also provide a platform on which to explore models of higher-order brain functions (for example, psychiatric conditions), which are currently impeded by computing bottlenecks16,17.

Here we fabricated sub-100-nm components, each of which incorporated a NbO2 volatile Mott memristive switch, an internal parallel capacitor defined by the metal contacts sandwiching a dielectric, and an internal series resistor defined by an electrode interface (Fig. 1a–c, Supplementary Figs. 2–4). The quasistatic current–voltage behaviour (Fig. 1d) of the element measured by sourcing a current consists of an S-type negative differential resistance (NDR) at lower currents, followed by a box-shaped hysteresis at higher currents. The NDR is known to originate from a positive feedback mechanism in which Joule heating is enhanced by the super-linear thermally activated transport of NbO218–20. The box-shaped hysteresis in quasistatic measurements has been observed recently and attributed to a Mott transition, but other mechanisms such as local temperature redistribution have also been suggested20–22. Although interesting nonlinear phenomena have been reported in materials exhibiting a Mott transition (such as chaos driven by thermal noise)3, the unique dynamics associated with the Mott transition itself have not been previously characterized, and as we will show here, the transition contributes an additional state variable that can be harnessed to produce neuromorphic functions. The hysteresis may have different causes in various components, and so we performed additional physical measurements to determine its origins in our element. Cross-sectional transmission electron micrographs and electron-diffraction patterns within the active as-grown amorphous NbO2+δ (over-oxidized NbO2) layer of an element that had been operated once past the hysteresis exhibited crystallized NbO2 within a 10-nm region near the centre of the structure (Fig. 1e, f, Supplementary Figs. 5–7). The crystallization temperature is close to the local temperature expected at the power level of the Mott transition18,20,23, and so we infer that Joule heating in a localized active region forms the ordered structure. Material surrounding the active region had a higher oxygen concentration and was probably not Mott active24.

Hewlett Packard Labs, Palo Alto, CA, USA. 2Texas A&M University, College Station, TX, USA. 3Stanford University, Stanford, CA, USA. ✉e-mail: [email protected]

1

518 | Nature | Vol 585 | 24 September 2020

a

Cint im RNbO (T, iNbO ) 2

vext or iext

2

Rint vm Integrated circuit element vdev

c

b

Pt T TiN N NbO2 SiO2 S TiN T S SiNx W

g

e

Nb 4d* σ*

Nb 5s*

50 1 0

Si1-Si2 0

im (mA)

0.6

Quasi-static Mott transition: the missing dynamics

π*

Data Fit

d||* 530

[001] 0.0 0.0

0.2 Vm (V)

0.4

U (×10–3 Ω m)

h

f

NDR

0.3

540 Energy (eV)

ΔOD (×10–3)

0.9

XAS in OD (×10–3)

d

–1

550

Static Mott 10 NDR 1 No NDR 0.1 1.0

Mott dynamics 1.5

2.0 x in NbOx

2.5

Fig. 1 | Element construction and static measurements. a, Circuit model of the integrated element. b, Schematic illustration of the structure, indicating the presence of the different electronic constituents. c, Cross-sectional transmission electron micrograph displaying a part of the structure of the element. Scale bar is 30 nm. d, Quasistatic current–voltage behaviour of the element. e, High-resolution transmission electron micrograph of a region within an element (indicated by dotted rectangle in c) that was operated past the box-shaped hysteresis in d. Dashed green box contains obvious crystal ordering. Scale bar is 8 nm. f, Electron diffraction pattern of the crystalline region indicating monoclinic [001] ordering. g, In operando X-ray absorption spectrum (XAS) (measured in optical density, OD) of a prototypical component identical in material stack and behaviour to the element under study, along with the spectral difference corresponding to two current levels (i1 = 0.6 mA and i2 = 1 mA) on either side of the hysteresis in the element’s current–voltage behaviour. The constituent bands are marked. The spectral difference (S i1 − S i2) (in change in OD, ΔOD) indicates the occurrence of a Mott transition specifically within the hysteresis (see Supplementary Information sections 2–4 for additional details). h, Resistivity plotted against the stoichiometry of NbOx, with corresponding electrical behaviours: (I) NDR, (II) hysteresis in the current–voltage behaviour, (III) dynamical neuromorphic properties when operated within the hysteresis, and (IV) no NDR. The blue (NDR) region exhibits only behaviour (I), green (Static Mott) exhibits only behaviours (I) and (II), red (Mott dynamics) exhibits behaviours (I)–(III), and the grey region exhibits only behaviour (IV).

We also searched for spectro-microscopy signatures of a Mott transition by applying a modified in operando synchrotron X-ray absorption technique25 specifically within the hysteresis, by removing signals from operating regions outside of the hysteresis. We observed lowering of

the π* band and disappearance of a d⁎ band at higher temperatures, consistent with both an electronic Mott insulator-to-metal transition and also the associated changes to the structural phase ordering in NbO2 (Fig.  1g, Supplementary Fig.  8) 26. To induce the desired higher-order Mott transition dynamics, the geometric structure of the element was optimized for both electrical and thermal properties27, and the material composition was also carefully tuned (Fig. 1h). Neuromorphic functions occurring only within a narrow subset of material compositions that support NDR is predicted by Chua’s theory of local activity (see Supplementary Information section 1, Supplementary Fig. 1)2. The element was powered by connecting a tunable constant voltage across the terminals that could access different parts of its current– voltage curve via load-lines determined by the internal resistor and the applied voltage (Fig. 2a). When biased just below the hysteresis (vext = 1.8 V), we observed self-sustained sinusoidal oscillations, but when biased within the hysteresis (vext = 1.95 V), we observed periodic two-spike bursting, which changed into periodic single spikes at a higher voltage (vext = 2.05 V), similar to a neuron’s action potential (Fig. 2b, c). At even higher voltages (vext = 2.1 V, load-line biasing just above the hysteresis), the spikes abruptly transitioned to low-intensity periodic features similar to the super-threshold damped spiking of neurons5. In an additional experiment, we slowed down the dynamics using an external capacitor and a resistor along with the element to observe the entire extent of behaviours noted above within a single time series by tuning the voltage across the terminals. Including the behaviours noted here, we identified a total of 15 different neuromorphic responses of our third-order element that originate by tuning the voltage across the element (Supplementary Information section 5, Supplementary Figs. 9–19). We constructed a simple compact model for the element to explore the dynamics in simulation. We used a Schottky equation (equation (1)) for the state-dependent conduction model, because any sufficiently nonlinear thermally activated transport can produce NDR. We represented the dynamics of the temperature state variable T using Newton’s law of cooling (equation (2)), and included the dynamics of the capacitor via the Kirchhoff current and voltage laws (equation (3)). To represent the Mott transition, we included a switchable Ohmic conductor (Rmet) in parallel with the Schottky transport when the NbO2 is in its metallic state at higher power levels (equation (4)). Equation (5) models a hysteresis in the switchable Ohmic resistor.

 q q0v m − ϕ    q0vm   0 dκ   iSch = AT 1 − exp −   exp   k T k B BT      

(1)

dT ioxvm T − Tamb = − dt Cth RthCth

(2)

dvm 1  1  =  (vext − vm) − iox  dt C  RS 

(3)

2

iox = iSch +

vm Rmet

      di  Rmet = R0 − tanh α iox + sgn  ox  I1 − I2   + β     dt     

(4)

(5)

In equations (1)–(5), iSch is the Schottky current; A, q0, κ, d, ϕ, kB, α, R0 and β are constants; vm is the voltage across NbO2; iox is the current through NbO2; t is time; T is the absolute temperature of NbO2 and Tamb is the ambient temperature; Cth and Rth are the thermal capacitance and resistance, respectively; RS is the series resistor (RS = Rint when Nature | Vol 585 | 24 September 2020 | 519

Article vext = 1.8 V

b

vext = 2.1 V

0.3

Beats

Periodic oscillations

0.0 0.0

d

0.2 0.4 Vm (V)

im (mA)

0.5

3

0

3

0

3

0

0.0 0.0

0.3

0.6 0.75 t (μs)

0.3

1.00 0.50

vext (V)

im (mA)

0.0

1.40

im (mA)

f

0.6

60

2

t (μs) vext = 1.7 V vext = 1.9 V

100

120

vext = 2.1 V

vext = 2.2 V

0.2 0.4 Vm (V)

i

1 Resting state

2 Depolarization

Outside cell Na+ Cell wall

K+

Na+

Inside cell ≈ Undercooling

≈ Inactivity 4 Refractory period

3 Re- and hyper-polarization

Na+ K+

K+

≈ Cooling

3

0

3

0

3

0

3

0 0.0

–1 –2

1

≈ Heating and Mott transition

0

j

k 1,500 T (K)

Periodic oscillations 0.0 0.0

80

0

0.3

t

Inter-cell transmission

1

g 20 im (mA)

im (mA)

Simulation of model

40

Axon terminals and synapses

0.75

1.45

Action potential

1

4

Electrical impulse (action potential)

0.5

0.9

3

0.5

1.0

Beyond-threshold activity

2 1

1.0

1.50

20

Potential

Axon

3

0

1.5

0

0V

Nucleus

0.0

c

Physico-chemical stimuli

Cell body

Analogy to a biological neuron

Chaos

im (mA)

0.6

e

h

1.0 Action potential

im (mA)

vext = 2.05 V

Dendrites

0.9

Experimental data

vext = 1.95 V

–im (mA)

Beyond-threshold activity

a

1

2

3

4

1

1,000 500

0.3 0.0

0.3

0.25 t (μs)

0.50 0.0

0.3

0.0

0.1 t (μs)

0.2

Fig. 2 | Experimental measurements and modelling of action potentials. a, Quasistatic current–voltage behaviour of the circuit element (data repeated from Fig. 1d). The dashed load-lines correspond to the colour-coded voltage biases shown in b. The overlaid ellipses indicate biasing regions that exhibit a range of qualitatively different oscillatory behaviours in the third-order element. b, Measured temporal dynamics of the element’s current (im) at different applied external voltages, as labelled. c, Data from b magnified in time. d, The temporal dynamics of the element’s current at different applied external voltages, measured in a network of the element with an external

capacitor and resistor. e–g, Simulation results corresponding to a–c, obtained from the third-order compact model. h, Illustration of a biological neuron, labelling its parts (left) and their functions (right). Four dynamical events during the electrical impulse (1–4) are marked. i, Illustration of the biological origins of the four events 1–4, along with equivalent physical processes in the third-order element (indicated by ‘≈’). j, Modelled current through the element (the polarity is reversed because the current in the biological neuron is ionic, not electronic). k, As in j, corresponding temperature dynamics. The equivalents of the four events 1–4 are marked as time windows.

no external series resistor is used); I1 and I2 are constants defining the bounds of the hysteresis; and sgn is the sign function. Additional details are provided in Supplementary Information section 7, Supplementary Figs. 20–25. This is a third-order model that contains three state variables, each of which has a distinct dynamical equation: T, vm and Rmet. Note that each order of complexity is a physical process with its associated state variable; for example, the Mott transition dynamics is represented by Rmet, where equation (5) is written in integral form instead of differential form. Although the underlying physics and the electrical behaviour can be represented in more accurate and elaborate forms, we chose a simple

representation to provide an intuitive illustration that a third-order system can enable neuromorphic dynamics. Numerical simulations of this model were in agreement with experimental data (Fig. 2e–g), thus demonstrating how a Mott insulator, which is not a typical electronic element, can be the basis for a nonlinear response that produces dynamical behaviour in a circuit. An action potential in a biological neuron consists of three events: up-shoot, lowering and under-shoot in potential, relative to the resting state of the neuron, driven by ionic transport (Fig. 2h, i)14. The dynamics of the current flowing into the neuromorphic element contains similar shapes within each spike (Fig. 2j). Using our model, we correlate each

520 | Nature | Vol 585 | 24 September 2020

a

iin1

N1

vin1

Rout Nout

N2

vin2

iin1 (mA)

b

iin2

c

iout = iin1 NAND iin2 1.4

iout

iout = iin1 NOR iin2

0

0

1

1

0

0

0

1

1

0

1

0

0

1

1

0

0

0

1

1

1

1

1

0

1

1

0

0

0

20

40 60 t (μs)

0.7

iin2 (mA)

0.0 1.4 0.7

iout (mA)

0.0 1

1.4 0.7 0.0 0

80

0

20

40 t (μs)

60

Fig. 3 | Experimental demonstration of universal Boolean logic via nonmonotonic spiking behaviour. a, Schematic of the network of three neuromorphic elements (N1, N2, Nout), with each constructed using a third-order element, an external capacitor and a resistor. b, Temporal dynamics of the three currents marked in a for Rout = 427 Ω, showing the NAND operation. c, Temporal dynamics of the three currents marked in a for Rout = 389 Ω, showing the NOR operation.

of these events to specific temperature responses driven by the Mott transition (Fig. 2i–k), which provides an illustration of how the physics of our nanoscale element produced a neuromorphic action potential. Any new electronic component faces stiff competition from entrenched transistor circuits for chip-level integration. Here we demonstrate that simple networks of our third-order elements can perform nonmonotonic operations and transistorless all-analogue computations. First, we constructed a simple network of three third-order elements (Fig. 3a) and included a tunable coupling resistor. By holding the input voltages at one of two levels (to produce spiking or damped spiking), and by carefully tuning the coupling resistor, spiking (or damped spiking) at both the inputs may propagate such that, the output element received a high (or low) signal, producing damped spiking (or spiking). Spiking in only one of the two inputs may propagate to the output as a high or a low signal depending on the coupling resistor. Thus, by tuning the coupling resistor (Fig. 3b, c), we obtained behaviours corresponding to NAND and NOR operations. Boolean operations using simple networks of neuromorphic elements have been modelled before and are not intended as hardware primitives to replace transistor-based digital logic, which is unlikely to be surpassed in efficiency at large scales28. However, the nonmonotonic operations based on the bifurcations of the spiking are being studied as a basis for transistorless neuromorphic primitives such as cellular automata28. To experimentally demonstrate analogue computing, we constructed an integrated array of 24 nanocircuit elements coupled by an impedance matrix defined by a programmable crossbar array of non-volatile memristive switches exhibiting pseudo-memcapacitance (Fig. 4c). In the pseudo-memcapacitors, constructed using a material stack consisting of two back-to-back metal–insulator–metal structures,

the resistance switching is accompanied by changes in capacitance (Fig. 4d), and thus the passive crossbar array is programmed with a problem (represented by a weight matrix), just as arrays of non-volatile resistive memories are programmed. The oscillators are powered by a single bias voltage, and their phases are monitored for convergence. The solution of the problem represented by the connection matrix is encoded in the phase of the oscillations29. This system has approximate similarities to thalamo-cortical computations in the brain, which occur in networks of oscillating neurons connected either via tunable synapses or a hub/thalamus that processes and routes neural signals (Fig. 4a)30. This leads to synchronization within the dynamics of the neural oscillations (for example, phase alignment), resulting in spatiotemporal classification, for instance, natural language and face recognition (Fig. 4b)31–33. Although the idea of using phase synchronization of oscillators to identify data, and to a limited extent to perform optimization, has been previously demonstrated33–35, the combination of third-order elements with pseudo-memcapacitors used here enables a highly efficient and compact hardware implementation. Most digital or analogue–digital hybrid approaches to neural networks require clocking of the circuit, explicit feedback in the case of recurrent networks (that usually involves digital–analogue conversions, amplification, and so on), and often in the case of oscillator-based computing, construction of bulky transistor-based oscillators and a connection matrix, all of which impede scalability34–37. Many of these limitations continue to exist despite recent efforts on two-terminal-memory-based Boltzmann machines and Hopfield networks38,39. The neuromorphic-element and pseudo-memcapacitor network is a transistorless all-analogue reprogrammable system with all-to-all connectivity that does not require clocking or explicit feedback, and is simpler to construct relative to simulating these functions via more elaborate digital circuits. To illustrate the system’s operation, we programmed toy instances of the viral quasispecies reconstruction problem (generically formulated as a maximum-cut graph problem), which seeks to identify genetic diversities of intra-host viral populations (Fig. 4g)40, and is important in ensuring the effectiveness of medications especially against viral species exhibiting diversity41. The problem is solved by repeatedly partitioning a graph (network of genomic reads and conflicts) into two sets of vertices (reads) by maximizing the edges (conflicts) between the sets (known as a maximum graph cut), so that the eventual solution consists of sets (inferred mutations) of reads with minimized intra-set conflicts42. The problem is nondeterministic polynomial-time (NP)-hard and is not efficiently addressed by traditional von Neumann computers, necessitating novel software algorithms40,42. Generic formulations of graph-partitioning problems are being used for benchmarking performance of optical, quantum and electronic NP-hard optimization solvers, enabling future performance comparisons43. In addition to the illustrative example chosen here, any problem that can be represented by an Ising formulation can be programmed on this hardware. The experimental solution to a subset of the programmed problem displays phase synchronization consistent with the optimal solution to division of the corresponding graph (Fig. 4h–j). In a similar fashion to solving optimization problems with recurrent neural networks, this approach is prone to converge to incorrect solutions, which may be only locally optimal (local minimum of a non-convex function). Transient chaotic dynamics—one of the properties of neuromorphic elements—can excite the system out of local minima and enable convergence to a global minimum (optimal solution)3. A single third-order element exhibits a range of behaviours—including periodic oscillations and chaotic dynamics—and so biasing the neuromorphic elements in a region that exhibits chaos (vext = 1.7 V) leads to clearly better statistics in minimizing the errors in solutions, relative to an operational region that produces non-chaotic periodic oscillations (vext = 1.3 V) (Fig. 4k, l). This also enables modulation of the input bias in order to tune the degree of chaos while approaching convergence, similar to computational simulated annealing. We further provide two benchmark metrics: time Nature | Vol 585 | 24 September 2020 | 521

Article c

d Conn

ectio

n ma

Os

cil

lat

e

0

or

s

0 Vc (V)

Viral genome Reads

2

C-graph

Mutations

Cc (pF)

Gc (mS)

A

A′

i

50

0.1

l

Err (%)

t (μs)

N=8

10

t 0

0.2

vext = 1.3 V

20

Err = 0

j

A

≈

0 0.0

k

Spatio-temporal synchronization

60

0

1

0

–2

vext = 1.7 V Frequency (a.u.)

0.3

m 10

0 10 110 0110 01110 1

0.1

1

ps

Potential

b

g

f

h

Hub (thalamus)-mediated neural connections

30

Conflict

Thalamo-cortical computing

Computing with neuromorphic elements Vo (mV)

Neuron groups (cortical columns)

trix

60 Cc (pF)

vext

tsol (μs)

a

0.01 0.1

2

N

10

20

Fig. 4 | Experimental demonstration of neuromorphic analogue computing. a, Illustration of simplified thalamo-cortical neural topology in a brain. b, Illustration of the phase synchronization of neuron oscillations resulting from the thalamo-cortical network. c, Schematic illustration of the experimental system with the neuromorphic oscillators and the connection matrix formed by a crossbar array of pseudo-memcapacitors. d, Quasistatic capacitance as a function of the applied voltage of a prototypical pseudo-memcapacitor. e, f, Conductance, Gc (e) and capacitance, Cc (f) maps of a crossbar array of linear size N = 24. g, Depiction of the viral quasispecies reconstruction problem, where reads of a viral genome are represented by a conflict graph (C graph), the graph partitions of which produce the inferred mutations. h, Oscillations from five neuromorphic oscillators connected to the part of the crossbar array represented by the dashed squares in e and f. Two phase groups (A and A′) are identified. Colours of the data correspond to the

node colours of the graph in i. i, The problem graph corresponding to the matrix within the dashed squares in e and f. The minimum conflict partitioning solution—also represented by the phase group A in h—is marked with a pink dashed ellipse. j, The adjacency matrix representing the graph in i. k, Illustration of minimization of errors (Err, percentage error from an optimal solution). The blue curve with no perturbations converges to a local minimum, whereas the orange curve with perturbations escapes all local minima to converge to the global minimum (Err = 0). l, Error distribution histograms of two sets of solutions obtained by applying two different vext to the oscillators. m, Time to solution (tsol, time for phase differences to settle) and probability of solution ps (fraction of convergences to the globally optimal solution), as a function of N. The shaded regions are the ranges over 10 to 60 experiments at different N. a.u., arbitrary units.

to solution (or convergence) and the accuracy (percentage of attempts converging on the optimal solution), as a function of the problem size, N (Fig. 4m). A comprehensive performance comparison to competing digital and hybrid technologies crucially awaits widespread reporting of generic benchmark metrics such as the energy required to reach a solution or the number of solutions obtained for a joule of energy. Additional information is provided in Supplementary Information section 9, Supplementary Figs. 26–31. In conclusion, we have demonstrated that it is possible to incorporate the Mott transition in NbO2 as an additional dynamical process to construct an isolated nanoscale electronic circuit element with third-order complexity. The element can be designed to produce optimal interactions among its constituent electrical and thermal components, such that it produces neuromorphic action-potential behaviours when powered by a constant voltage source. We demonstrate two nonmonotonic and complete logic operations using one simple network of our elements, and we further demonstrate a transistorless all-analogue network of neuromorphic elements to solve computationally hard

problems that have far-reaching applications in alleviating the von Neumann bottleneck of present digital computers. This result enables extremely compact and highly functional neuromorphic computing primitives.

522 | Nature | Vol 585 | 24 September 2020

Online content Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41586-020-2735-5. 1. 2. 3. 4.

Mead, C. Neuromorphic electronic systems. Proc. IEEE 78, 1629–1636 (1990). Mainzer, K. & Chua, L. Local Activity Principle (Imperial College Press, 2013). Kumar, S., Strachan, J. P. & Williams, R. S. Chaotic dynamics in nanoscale NbO2 Mott memristors for analogue computing. Nature 548, 318–321 (2017). Chua, L. Memristor, Hodgkin–Huxley, and edge of chaos. Nanotechnology 24, 383001 (2013).

5. 6. 7. 8. 9. 10. 11. 12.

13. 14. 15.

16. 17. 18. 19. 20. 21.

22. 23. 24. 25. 26.

Izhikevich, E. M. Dynamical Systems in Neuroscience. (MIT Press, 2007). Chua, L. Everything you wish to know about memristors but are afraid to ask. Radioengineering 24, 319–368 (2015). Chua, L. Handbook of Memristor Networks (Springer Nature, 2019). Bohaichuk, S. M. et al. Fast spiking of a Mott VO2–carbon nanotube composite device. Nano Lett. 19, 6751–6755 (2019). Kendall, J. D. & Kumar, S. The building blocks of a brain-inspired computer. Appl. Phys. Rev. 7, 011305 (2020). Zidan, M. A., Strachan, J. P. & Lu, W. D. The future of electronics based on memristive systems. Nature Electronics 1, 22 (2018). Paugam-Moisy, H. & Bohte, S. Computing with spiking neuron networks. In Handbook of Natural Computing (eds Rozenberg, G. et al.) 335–376 (Springer, 2012). Pickett, M. D., Borghetti, J., Yang, J. J., Medeiros-Ribeiro, G. & Williams, R. S. Coexistence of memristance and negative differential resistance in a nanoscale metal–oxide–metal system. Adv. Mater. 23, 1730–1733 (2011). Pickett, M. D., Medeiros-Ribeiro, G. & Williams, R. S. A scalable neuristor built with Mott memristors. Nat. Mater. 12, 114–117 (2013). Yi, W. et al. Biological plausibility and stochasticity in scalable VO2 active memristor neurons. Nat. Commun. 9, 4661 (2018). Khanday, F. A., Kant, N. A., Dar, M. R., Zulkifli, T. Z. A. & Psychalinos, C. Low-voltage low-power integrable CMOS circuit implementation of integer- and fractional-order FitzHugh–Nagumo neuron model. IEEE Trans. Neural Netw. Learn. Syst. 30, 2108–2122 (2018). Markram, H. Seven challenges for neuroscience. Funct. Neurol. 28, 145–151 (2013). Palmer, T. Modelling: build imprecise supercomputers. Nature 526, 32 (2015). Gibson, G. A. et al. An accurate locally active memristor model for S-type negative differential resistance in NbOx. Appl. Phys. Lett. 108, 023505 (2016). Slesazeck, S. et al. Physical model of threshold switching in NbO2-based memristors. RSC Adv. 5, 102318–102322 (2015). Kumar, S. et al. Physical origins of current- and temperature-controlled negative differential resistances in NbO2. Nat. Commun. 8, 658 (2017). Li, S., Liu, X., Nandi, S. K., Nath, S. K. & Elliman, R. G. Origin of current-controlled negative differential resistance modes and the emergence of composite characteristics with high complexity. Adv. Funct. Mater. 29, 1905060 (2019). Goodwill, J. M. et al. Spontaneous current constriction in threshold switching devices. Nat. Commun. 10, 1628 (2019). Zhang, J. et al. Thermally induced crystallization in NbO2 thin films. Sci. Rep. 6, 34294 (2016). Seta, K. & Naito, K. Calorimetric study of the phase transition in NbO2. J. Chem. Thermodyn. 14, 921–935 (1982). Kumar, S. et al. Spatially uniform resistance switching of low current, high endurance titanium–niobium–oxide memristors. Nanoscale 9, 1793 (2017). Kumar, S. et al. The phase transition in VO2 probed using X-ray, visible and infrared radiations. Appl. Phys. Lett. 108, 073102 (2016).

27. Gibson, G. A. Designing negative differential resistance devices based on self-heating. Adv. Funct. Mater. 28, 1704175 (2018). 28. Pickett, M. D. & Williams, R. S. Phase transitions enable computational universality in neuristor-based cellular automata. Nanotechnology 24, 384002 (2013). 29. Kopell, N. & Somers, D. Anti-phase solutions in relaxation oscillators coupled through excitatory interactions. J. Math. Biol. 33, 261–280 (1995). 30. Hoppensteadt, F. C. & Izhikevich, E. M. Thalamo-cortical interactions modeled by weakly connected oscillators: could the brain use FM radio principles? Biosystems 48, 85–94 (1998). 31. Bansal, K. et al. Cognitive chimera states in human brain networks. Sci. Adv. 5, eaau8535 (2019). 32. Steriade, M. Synchronized activities of coupled oscillators in the cerebral cortex and thalamus at different levels of vigilance. Cereb. Cortex 7, 583–604 (1997). 33. Csaba, G. & Porod, W. Coupled oscillators for computing: a review and perspective. Appl. Phys. Rev. 7, 011302 (2020). 34. Chou, J., Bramhavar, S., Ghosh, S. & Herzog, W. Analog coupled oscillator based weighted Ising machine. Sci. Rep. 9, 14786 (2019). 35. Parihar, A., Shukla, N., Jerry, M., Datta, S. & Raychowdhury, A. Vertex coloring of graphs via phase dynamics of coupled oscillatory networks. Sci. Rep. 7, 911 (2017); correction 8, 6120 (2018). 36. Maffezzoni, P., Bahr, B., Zhang, Z. & Daniel, L. Oscillator array models for associative memory and pattern recognition. IEEE Trans. Circuits Syst. I 62, 1591–1598 (2015). 37. Romera, M. et al. Vowel recognition with four coupled spin-torque nano-oscillators. Nature 563, 230–234 (2018). 38. Mahmoodi, M., Prezioso, M. & Strukov, D. Versatile stochastic dot product circuits based on nonvolatile memories for high performance neurocomputing and neurooptimization. Nat. Commun. 10, 5113 (2019). 39. Cai, F. et al. Power-efficient combinatorial optimization using intrinsic noise in memristor Hopfield neural networks. Nat. Electron. 3, 409–418 (2020). 40. Huang, A., Kantor, R., DeLong, A., Schreier, L. & Istrail, S. QColors: an algorithm for conservative viral quasispecies reconstruction from short and non-contiguous next generation sequencing reads. In Silico Biol. 11, 193–201 (2011). 41. Pang, J. et al. Potential rapid diagnostics, vaccine and therapeutics for 2019 novel coronavirus (2019-nCoV): a systematic review. J. Clin. Med. 9, 623 (2020). 42. Mangul, S. et al. Accurate viral population assembly from ultra-deep sequencing data. Bioinformatics 30, i329–i337 (2014). 43. Hamerly, R. et al. Experimental investigation of performance differences between coherent Ising machines and a quantum annealer. Sci. Adv. 5, eaau0823 (2019). Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. © The Author(s), under exclusive licence to Springer Nature Limited 2020

Nature | Vol 585 | 24 September 2020 | 523

Article Data availability Data presented in the main figures of the manuscript are available from the authors upon reasonable request. Acknowledgements We thank L. O. Chua and H. S. P. Wong for comments on the manuscript. We also thank K. J. Cremata, A. L. D. Kilcoyne, T. Tyliszczak, D. Shapiro, G. Gibson, X. Sheng and J. Zhang for assistance in collecting experimental data or construction of the models. We acknowledge S. M. Bohaichuk, J. C. Nino, A. Conklin and J. L. Andrews for discussions on specific topics and suggestions for illustrations. Work was performed in part in the nano@Stanford labs, which are supported by the National Science Foundation under award ECCS-1542152. Synchrotron measurements were conducted at the Advanced Light Source, a US DOE Office of Science User Facility under contract no. DE-AC02-05CH11231, and at the Stanford Synchrotron Radiation Lightsource, SLAC National Accelerator Laboratory, a US DOE Office of Science User Facility under contract no.

DE-AC02-76SF00515. R.S.W. acknowledges the X-Grants Program of the President’s Excellence Fund at Texas A&M University. Author contributions S.K. and R.S.W. conceived the project and planned the various measurements. Z.W. performed the materials growth, compositional analysis and parts of the chip fabrication. S.K. and Z.W. performed the in operando spectroscopic measurements and collected the electrical data. S.K. and R.S.W. constructed the model and wrote the manuscript. Competing interests The authors declare no competing interests. Additional information Supplementary information is available for this paper at https://doi.org/10.1038/s41586-0202735-5. Correspondence and requests for materials should be addressed to S.K. Peer review information Nature thanks Adnan Mehonic, Syed Ghazi Sarwat and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Reprints and permissions information is available at http://www.nature.com/reprints.

Article

Colloidal diamond https://doi.org/10.1038/s41586-020-2718-6 Received: 9 February 2020

Mingxin He1,2, Johnathon P. Gales2, Étienne Ducrot2,3, Zhe Gong4, Gi-Ra Yi5, Stefano Sacanna4 ✉ & David J. Pine1,2 ✉

Accepted: 22 July 2020 Published online: 23 September 2020 Check for updates

Self-assembling colloidal particles in the cubic diamond crystal structure could potentially be used to make materials with a photonic bandgap1–3. Such materials are beneficial because they suppress spontaneous emission of light1 and are valued for their applications as optical waveguides, filters and laser resonators4, for improving light-harvesting technologies5–7 and for other applications4,8. Cubic diamond is preferred for these applications over more easily self-assembled structures, such as face-centred-cubic structures9,10, because diamond has a much wider bandgap and is less sensitive to imperfections11,12. In addition, the bandgap in diamond crystals appears at a refractive index contrast of about 2, which means that a photonic bandgap could be achieved using known materials at optical frequencies; this does not seem to be possible for face-centred-cubic crystals3,13. However, self-assembly of colloidal diamond is challenging. Because particles in a diamond lattice are tetrahedrally coordinated, one approach has been to self-assemble spherical particles with tetrahedral sticky patches14–16. But this approach lacks a mechanism to ensure that the patchy spheres select the staggered orientation of tetrahedral bonds on nearest-neighbour particles, which is required for cubic diamond15,17. Here we show that by using partially compressed tetrahedral clusters with retracted sticky patches, colloidal cubic diamond can be self-assembled using patch–patch adhesion in combination with a steric interlock mechanism that selects the required staggered bond orientation. Photonic bandstructure calculations reveal that the resulting lattices (direct and inverse) have promising optical properties, including a wide and complete photonic bandgap. The colloidal particles in the self-assembled cubic diamond structure are highly constrained and mechanically stable, which makes it possible to dry the suspension and retain the diamond structure. This makes these structures suitable templates for forming high-dielectric-contrast photonic crystals with cubic diamond symmetry.

The superior optical properties of cubic diamond compared to other self-assembled structures has led to investigations of the possibility of self-assembling a diamond lattice from colloidal spheres14,16,18,19. However, the diamond lattice poses a challenge for colloidal self-assembly. The spheres in a diamond lattice are tetrahedrally coordinated (Fig. 1a), which means that they have two fewer constraints than the six required for mechanical stability and a maximum packing fraction of π 3 /16 ≈ 0.34. Unlike face-centred-cubic colloidal crystals9, in which the spheres have 12 nearest neighbours and a maximum packing fraction of π/ 18 ≈ 0.74 , diamond crystals cannot be stabilized by entropy alone. One way to address this challenge is to self-assemble a superlattice of two or more colloidal species, with one of the sublattices being diamond19–22. This solves the low-packing-density problem by backfilling the voids with a temporary lattice that is ultimately removed, but doing so is delicate and has yet to be demonstrated. Another approach is to build a three-dimensional DNA scaffold and tether small gold

nanoparticles within the scaffold23, but the length scales are too small and the particles are disconnected, precluding the formation of photonic bandgaps. Yet another approach is to use faceted particles with attractive interactions, which has yielded some surprising results, such as colloidal clathrate24. This approach is related to an earlier one, suggested on the basis of simulations25, which involves triangular di-patches on spheres and can produce colloidal clathrate or diamond, depending on the relative orientations of the di-patches. In other simulations, certain truncated tetrahedra are predicted to have diamond phases26, but it is not clear whether they could serve as templates for photonic crystals. In devising any method to make diamond photonic crystals, it is important to distinguish between cubic and hexagonal diamond. Cubic diamond has a photonic bandgap; hexagonal diamond does not. An important difference between cubic and hexagonal diamond is the way each particle is connected to its four nearest neighbours15. For cubic diamond, all four nearest neighbours are connected in the staggered

Department of Chemical and Biomolecular Engineering, New York University, Brooklyn, NY, USA. 2Department of Physics, Center for Soft Matter Research, New York University, New York, NY, USA. 3University of Bordeaux, CNRS, Centre de Recherche Paul Pascal, Pessac, France. 4Department of Chemistry, Molecular Design Institute, New York University, New York, NY, USA. 5School of Chemical Engineering, Sungkyunkwan University, Suwon, South Korea. ✉e-mail: [email protected]; [email protected]

1

524 | Nature | Vol 585 | 24 September 2020

a

b

c

Here we show that the rotational information needed to select the staggered conformation can be written into the shape of the particles. The idea of using shaped colloidal clusters to direct colloidal self-assembly has been well investigated18,20,27–31. Figure 1c illustrates our particle design strategy. Each particle consists of four tetrahedrally coordinated, partially overlapping spherical lobes, shown in purple or white. At the centre of each of the four triangular faces is a DNA-coated patch, shown in light blue. The DNA on the patches is designed with self-complementary sticky ends so that patches on different particles are attractive below the melting temperature Tm of DNA patch. The radial extent of the patches is retracted from the plane formed by the convex hull of the spherical lobes. This means that the DNA on the patches of different particles can reach each other and bind only if the lobes on different particles are oriented in the staggered conformation, as shown in Fig. 1c. Below, we show with simulations and experiments that this is sufficient to stabilize the cubic diamond structure. Figure 1d, e and Supplementary Video 2 show the diamond unit cell formed by these particles.

Particle synthesis

d

e

Fig. 1 | Schematic and space-filling models of a colloidal diamond lattice. a, Unit cell of a cubic diamond crystal of spheres: a face-centred-cubic Bravais lattice with a two-particle basis (white and purple). Yellow rods show the tetrahedral bonds between atoms. b, Top right, eclipsed conformation; bottom right, staggered conformation; left, corresponding Newman projections. c, Bound spherical patchy particles (left; patches shown in blue) are completely free to rotate about the axis that connects their centres, without any preferred orientation. The finite patch size means that the bond angle is also flexible, making rings of 5, 6 and 7 particles possible. The patches in tetrahedral cluster particles (middle) can reach each other and bind only when the spherical lobes interlock, which fixes their patches in the staggered conformation. Bound patchy tetrahedral particles (right) form only six-membered rings, which are in the desired ‘chair’ conformation of cubic diamond, not the ‘boat’ conformation that occurs in hexagonal diamond. d, Unit cell of a cubic diamond crystal of patchy clusters, made artificially smaller to make the bonding between patches visible. e, Unit cell of a cubic diamond crystal of patchy clusters with correct sizing.

conformation (Fig. 1b). For hexagonal diamond, only three of the four nearest neighbours are connected in the staggered conformation; the fourth is connected in the eclipsed conformation (Fig. 1b)15. In atomic crystals, the staggered conformation is preferred because the sp3 bonding electrons between next-nearest-neighbour bonds are further away from each other than they are in the eclipsed conformation, minimizing the Coulomb energy. In a colloidal system, it is difficult to achieve both an attractive interaction that binds patches together and, simultaneously, a long-range interaction that produces either the staggered or eclipsed conformation. Therefore, for spherical patchy colloids, no conformation—staggered, eclipsed or anything in between—is energetically favoured (Fig. 1c, left, Supplementary Video 1).

The synthesis of our patchy compressed clusters builds on a colloidal fusion protocol reported recently32. In this protocol (Fig. 2a), solid non-crosslinked polystyrene particles are mixed with smaller droplets of a polymerizable oil, 3-trimethoxysilyl propyl methacrylate (TPM). When the ratio of the diameters of the solid particles and liquid droplets is near α = 1 + 2 ≈ 2.41, the stochastic aggregation of solid particles onto the smaller liquid droplets results in tetrahedral clusters—four solid particles bound to a liquid droplet—with nearly 100% yield32,33 (Methods). Density gradient centrifugation removes the small number of non-tetrahedral clusters. In the end, fewer than 1 in 1,000 particles are not tetrahedra. The next step is the controlled deformation of the polystyrene spheres by the addition of a plasticizer to the suspension; we use tetrahydrofuran (THF). The deformation of the spheres extrudes the liquid core of the clusters such that the core protrudes out of the interstices between each set of three polystyrene particles that form the four faces of the clusters. This is performed at room temperature, which allows us to finely tune the degree to which the polystyrene spheres are compressed and the liquid core is extruded (Fig. 2b). To characterize the geometry of the partially deformed clusters, two parameters are introduced: the compression ratio of the polystyrene spheres and the size ratio of the patches to the spheres (Fig. 2c, d). The compression and size ratios can be finely tuned by varying the concentration of plasticizer (THF) and the types of surfactant used (Extended Data Fig. 1). A compression ratio of 0 means that the four original polystyrene particles have coalesced into a single sphere; a compression ratio of 1 means that the clusters are not compressed at all. The clusters depicted in Fig. 2 have a compression ratio of 0.78, which is typical for our experiments. Although the size and compression ratios are closely linked, they can be independently adjusted to some degree.  By using different surfactants, which control the wetting angle between a TPM droplet and its polystyrene cluster, the size ratio can be changed. We find that using sodium dodecyl sulfate (SDS) gives the right amount of wetting34 (Methods). To fix the geometry of the patchy cluster, the plasticizer is evaporated to harden the polystyrene cluster and the liquid cores are solidified by free radical polymerization. Before clustering, the TPM oil is functionalized with epoxy groups by introducing (3-glycidyloxypropyl) trimethoxysilane. After deformation and polymerization, these epoxy groups are converted to azide groups, which can further react with dibenzocyclooctyne (DBCO)-functionalized DNA by strain-promoted azide-alkyne cycloaddition chemistry35. Because only the TPM patches have these surface functional groups, we can selectively functionalize the TPM patches with single-stranded Nature | Vol 585 | 24 September 2020 | 525

Article a (1)

(2)

(3)

b

c

d

dcc

a

b

a

Fig. 2 | Synthesis of compressed tetrahedral patchy clusters. a, (1) Aggregation of four polystyrene particles (white) around a smaller oil droplet (blue), followed by (2) controlled deformation of the polystyrene particles with THF, which extrudes the central oil droplet. (3) The THF is then removed, the oil polymerized and coated with DNA to produce solid compressed tetrahedral clusters with DNA-coated patches (red). b, Scanning electron microscopy (SEM) image of compressed tetrahedral clusters. Some TPM patches are highlighted in light blue. Scale bar, 1 μm. c, The compression ratio dcc/(2a) is defined as the distance between the centres of the spherical lobes divided by their diameter. d, The size ratio b/a is defined as the radial extent of the patches from the cluster centre divided by the radius of the spherical lobes. For the particles shown here, dcc/(2a) = 0.78 and b/a = 1.22.

DNA and leave the surfaces of the polystyrene spheres nearly bare (Extended Data Fig. 2). Although the preparation of these tetrahedral patchy colloidal clusters is complex, the experimental design is simple: it uses a single type of particle, with patches functionalized with a single type of DNA. Moreover, the self-assembly is robust, crystallization is relatively fast, and cubic diamond is the only product.

Particle design and crystallization To guide the design of our particles and verify the conditions under which they might crystallize into cubic diamond, we performed simulations using the HOOMD-blue simulation package36,37 (Methods). The phase diagram determined by the simulations is shown in Extended Data Fig. 3. We also performed numerical calculations of the photonic bandstructures of the resultant direct and inverse lattices using the MIT Photonic Bands software38 (see ‘Calculation of photonic bandgap’). The outcomes of these two sets of calculations led us to explore particles with compression ratios between 0.63 and 0.78, and size ratios near 1.2. The size of the primary polystyrene particles from which the compressed tetrahedral clusters are made was chosen to be 1.0 μm, as 526 | Nature | Vol 585 | 24 September 2020

this leads to a photonic bandgap centred at the technologically interesting wavelength of 1.5 μm, at which most optical communications networks work. Starting from 1.0-μm primary polystyrene particles, the resulting compressed clusters are slightly smaller than 2.0 μm across, resulting in substantial sedimentation, with a gravitational height of 2 μm in water. As shown in Extended Data Fig. 4, the particles bind, interlock and form small crystals after annealing overnight. To grow bigger crystals, the particles are nearly density-matched by suspending them in a mixture of H2O and D2O (with PBS buffer), which increases the gravitational height from about 2 μm to 20 μm. The suspension is loaded into a glass capillary and sealed, with typical dimensions of 100 μm × 2 mm × 50 mm. The capillary is tilted at 20° along the 2-mm dimension to provide an exponential atmosphere of particles and promote slow growth and annealing. A temperature gradient of about 1 °C, which spans the melting temperature of the DNA-coated patches, is applied along the long, 50-mm length of the capillary. The compressed clusters crystallize overnight, with typical crystal sizes of 40 μm, and some extending to 100 μm or more (Fig. 3a). To examine the crystal structure, the TPM cores are fluorescently labelled before polymerization when the clusters are prepared. Figure 3a–c shows images taken in the horizontal plane with a fluorescent microscope. Figure 3a, b reveals the honeycomb pattern characteristic of the 111 plane of diamond; the polystyrene lobes of the particles are not visible as they are not dyed. Figure 3c shows a crystal in which the 110 plane of cubic diamond can be seen. Whereas the 111 plane (Fig. 3a, b) appears in both the cubic and hexagonal versions of diamond, the 110 plane (Fig. 3c) is unique to the cubic diamond lattice. To further examine the structure of the self-assembled crystals, the hybridized DNA bonds that link the patches of neighbouring particles are permanently crosslinked via exposure to ultraviolet radiation in the presence of 8-methoxypsoralen (Methods)39. This allows the samples to be removed from the capillary and dried without disturbing their structure. To facilitate optical measurements in three dimensions, a sample is immersed in index-matching oil and viewed with a confocal microscope. The confocal z stacks reveal an ABC stacking of the 111 honeycomb planes, which confirms that the crystals are cubic diamond (Supplementary Videos 3, 4) and not hexagonal diamond, which has an AB stacking of the 111 planes. The confocal images show that the crystals are typically 10 or more layers thick, isotropic and fully three dimensional. We also view the psoralen-crosslinked dried crystals with a scanning electron microscope. Figure 3f confirms that the crystal is well preserved after drying. Figure 3g shows a side view of the dried crystal and reveals that the thickness of the diamond crystal is about 10 layers.

Calculation of photonic bandgap It is well established that diamond lattices of spheres exhibit a photonic bandgap. But the question of whether diamond lattices assembled from tetrahedral patchy clusters could also exhibit photonic bandgaps has not previously been considered. To address this question, we performed a series of photonic bandstructure calculations using the MIT Photonic Bands software38. We consider the direct and inverse lattices. The direct lattice is a cubic diamond lattice made of tetrahedral clusters like the ones we used, but with a higher refractive index. The inverse lattice is obtained by backfilling the interstices of the direct lattice with a high-index dielectric material, after which the original tetrahedral clusters are removed, leaving only air behind. The inset in Fig. 4 shows the unit cell of the inverse lattice (see also Extended Data Fig. 5a, Supplementary Video 5). We choose refractive indices of 2.6 and 3.4, corresponding to TiO2 and silicon, respectively, as these materials have high refractive indices in the visible and near-infrared, exhibit very little absorption of light in their respective frequency ranges and can be fabricated experimentally.

a

b

c

d

e

2 μm

f

g

h

Fig. 3 | Crystallization of cubic diamond colloidal crystals. a, Confocal microscope image showing the hexagonal symmetry of the 111 plane characteristic of diamond crystals. The signal originates from the fluorescently labelled TPM cores of the tetrahedral cluster patchy particles. b, c, Magnified confocal images showing the 111 (b) and 110 (c) planes of the cubic diamond crystal. d, e, Computer-generated images of the TPM patterns expected for the 111 (d) and 110 (e) planes for cubic diamond crystals. A single TPM core within one compressed cluster is highlighted in white. Inset, SEM image of the TPM core, in which polystyrene is dissolved and washed away by

THF. f, SEM images of the 111 plane of colloidal diamond crystals. The crystals are about 40 μm across, with grain boundaries and point defects. Inset, a computer-generated image showing the 111 plane of a colloidal diamond crystal for dcc/(2a) = 0.74, consistent with the SEM image. g, Side view of a crystal edge. The thickness of the crystal is about 10–20 particles. h, Magnified SEM image of the 111 plane showing the interlocking of particles, as designed. For particles shown in a, c, f and g, dcc/(2a) = 0.73 and b/a = 1.20; in b, dcc/ (2a) = 0.69 and b/a = 1.18; in h, dcc/(2a) = 0.75 and b/a = 1.19. In a–d, f–h, scale bars are 5 μm.

These calculations reveal that the direct and inverse (Extended Data Fig. 5c) cluster diamond lattices both have complete photonic bandgaps between the second and third bands, consistent with the photonic properties of conventional diamond lattices of spheres. Figure 4 shows how the relative width of the bandgap changes as the compression ratio is varied from 0 (a diamond lattice of non-overlapping spheres) to 1 (a diamond lattice of uncompressed

clusters), for refractive indices of 2.6 (TiO2) and 3.4 (silicon). These calculations reveal that the use of compressed clusters (0.1 ≤ dcc/ (2a) ≤ 0.8) opens up a bandgap for the inverse lattice, whereas no bandgap appears for non-overlapping spheres (dcc/(2a) = 0). The widest bandgap is achieved slightly below or above a compression ratio of 0.6, depending on the value of the refractive index. This is very near the compression ratios that we have already found result in crystallization Nature | Vol 585 | 24 September 2020 | 527

Article

Relative bandgap, Δf/fc

0.2

Inverse, 2.6 Inverse, 3.4 Direct, 2.6 Direct, 3.4

0.1

0.0 0.0

0.2

0.4

0.6

0.8

1.0

Compression ratio, dcc/(2a) Fig. 4 | Relative bandgap versus compression ratio. The relative bandgap is the width of the bandgap Δf divided by its centre frequency fc. Circles and triangles correspond to direct and inverse lattices, respectively. Blue and red points correspond to TiO2 (refractive index n = 2.6) and silicon (n = 3.4), respectively. The experimental range for the compression ratio found to lead to crystallization in experiments is highlighted in light blue; the maximum bandgaps for the inverse cluster cubic diamond are realized near the experimental conditions. The grey line shows the bandgap obtained for an inverse silicon lattice where a protective oxide layer of thickness 0.1a is used to coat the colloidal template before backfilling with silicon. The protective layer is then removed. Inset, inverse cubic diamond unit cell for a compression ratio of particles in the direct lattice of dcc/(2a) = 0.76.

experimentally: 0.63 ≤ dcc/(2a) ≤ 0.78 (Fig. 4, light blue). By contrast, using compressed clusters improves the bandgap only slightly for the direct lattice, with the widest bandgap occurring near dcc/(2a) = 0.15 and diminishing to 0 when dcc/(2a) = 0.6. In all of these calculations, the size ratio is fixed so that the patches touch precisely when the spherical lobes touch for neighbouring particles in the staggered conformation. Variations in the size ratio have very little effect on the bandgap. On the basis of previous numerical studies of the photonic bandgap of diamond crystals11, we expect the bandgap of the cubic diamond crystals described here to be robust with respect to disorder and various kinds of defects. Moreover, the crystals grown experimentally show good order, aided by the steric interlocking of clusters. In the geometric limit of ideal packing, each of the four faces of a colloidal cluster has six points of contact with its neighbour—seven including the sticky patch. This very large number of contact points per particle (28) helps to ensure orientational order.

Next steps The cubic diamond colloidal crystals described here are made from polystyrene and TPM, which have refractive indices of 1.6 and 1.4, respectively, too low to open up a photonic bandgap. Materials with refractive indices larger than 2 are needed to realize a photonic bandgap (Extended Data Fig. 5d). The simplest strategy to achieve this is to use our colloidal crystals as templates to make an inverse diamond structure by backfilling the interstices with a high-refractive-index material and then removing the colloidal template. Sol–gel chemistry40–42 or atomic layer deposition43,44 can be used to backfill a colloidal crystal with TiO2, which has a refractive index of around 2.6 in the visible and near-infrared. Similarly, chemical vapour deposition10,45 can be used to backfill a colloidal crystal with silicon, which has a refractive index of 3.4 in the near-infrared. Because chemical vapour deposition takes place at temperatures above the glass transition temperature of polystyrene of 105 °C, a low-temperature process such as atomic layer deposition 528 | Nature | Vol 585 | 24 September 2020

should be used to first coat the colloidal crystal with a protective oxide layer, after which the template can be coated with silicon using chemical vapour deposition45. The protective oxide layer can be left in place or removed (along with any remnant of the colloidal template) after the backfilling with silicon is complete. Extended Data Fig. 5b shows a rendering of the inverse lattice with the oxide layer removed. Removing the protective oxide layer increases the bandgap substantially. The grey line in Fig. 4 shows the bandgap obtained using a protective layer with a thickness of 0.1a, followed by backfilling with silicon. For a compression ratio of 0.65, the bandgap increases from 14% to 22%. Even greater increases can be achieved using thicker protective oxide layers. To realize the optical properties of photonic bandgaps, crystals that are ten or more unit cells thick are desirable. We have grown colloidal diamond crystals with lateral dimensions of up to 80 μm (about 30 unit cells) and thicknesses of up to 40 μm (about 15 unit cells). Although this is sufficiently large to investigate the photonic bandgap properties of these materials, crystals of larger lateral extent (several millimetres or more in size) would be more suitable for optical waveguides, lasers and other optical applications. There are well established methods to grow large colloidal crystals, such as epitaxial growth from a templated surface46. The inverse structure of the crystals reported here would have a photonic bandgap centred in the infrared, around a wavelength of 1.5 μm. As the photonic bandstructure scales with the crystal lattice constant, the particle size would need to be reduced by a factor of two to realize a bandgap in the visible range. The smaller size would make following the crystallization using an optical microscope more difficult, making experiments more challenging. The particles would probably be more polydisperse, but it should be feasible. Our approach combines directional interactions with a steric interlock mechanism that orients the attractive patches in the desired staggered conformation. We note that particle shape alone is insufficient to form diamond; removing the attractive interaction between patches results in amorphous structures. Our work suggests that, similarly to our DNA hybridization approach, any attractive interaction between patches—such as depletion47, hydrophobic48 or critical Casimir interactions49—should yield cubic diamond colloidal crystals.

Online content Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41586-020-2718-6.

1.

Yablonovitch, E. Inhibited spontaneous emission in solid-state physics and electronics. Phys. Rev. Lett. 58, 2059–2062 (1987). 2. John, S. Strong localization of photons in certain disordered dielectric superlattices. Phys. Rev. Lett. 58, 2486–2489 (1987). 3. Ho, K. M., Chan, C. T. & Soukoulis, C. M. Existence of a photonic gap in periodic dielectric structures. Phys. Rev. Lett. 65, 3152–3155 (1990). 4. Joannopoulos, J. D., Johnson, S. G., Winn, J. N. & Meade, R. D. Photonic Crystals: Molding the Flow of Light 2nd edn (Princeton Univ. Press, 2008). 5. Halaoui, L. I., Abrams, N. M. & Mallouk, T. E. Increasing the conversion efficiency of dye-sensitized TiO2 photoelectrochemical cells by coupling to photonic crystals. J. Phys. Chem. B 109, 6334–6342 (2005). 6. Bermel, P., Luo, C., Zeng, L., Kimerling, L. C. & Joannopoulos, J. D. Improving thin-film crystalline silicon solar cell efficiencies with photonic crystals. Opt. Express 15, 16986–17000 (2007). 7. Zhang, Z., Zhang, L., Hedhili, M. N., Zhang, H. & Wang, P. Plasmonic gold nanocrystals coupled with photonic crystal seamlessly on TiO2 nanotube photoelectrodes for efficient visible light photoelectrochemical water splitting. Nano Lett. 13, 14–20 (2013). 8. Yanik, M. F., Fan, S., Soljacić, M. & Joannopoulos, J. D. All-optical transistor action with bistable switching in a photonic crystal cross-waveguide geometry. Opt. Lett. 28, 2506–2508 (2003). 9. Pusey, P. N. & van Megen, W. Phase behaviour of concentrated suspensions of nearly hard colloidal spheres. Nature 320, 340–342 (1986). 10. Vlasov, Y. A., Bo, X. Z., Sturm, J. C. & Norris, D. J. On-chip natural assembly of silicon photonic bandgap crystals. Nature 414, 289–293 (2001).

11. 12. 13. 14. 15. 16. 17. 18. 19.

20.

21. 22. 23. 24. 25. 26.

27. 28. 29. 30.

31. 32.

Sigalas, M. M., Soukoulis, C. M., Chan, C. T., Biswas, R. & Ho, K. M. Effect of disorder on photonic band gaps. Phys. Rev. B 59, 12767–12770 (1999). Li, Z.-Y. & Zhang, Z.-Q. Fragility of photonic band gaps in inverse-opal photonic crystals. Phys. Rev. B 62, 1516–1519 (2000). Busch, K. & John, S. Photonic band gap formation in certain self-organizing systems. Phys. Rev. E 58, 3896–3908 (1998). Zhang, Z., Keys, A. S., Chen, T. & Glotzer, S. C. Self-assembly of patchy particles into diamond structures through molecular mimicry. Langmuir 21, 11547–11551 (2005). Romano, F., Sanz, E. & Sciortino, F. Crystallization of tetrahedral patchy particles in silico. J. Chem. Phys. 134, 174502 (2011). Wang, Y. et al. Colloids with valence and specific directional bonding. Nature 491, 51–55 (2012). Noya, E. G., Zubieta, I., Pine, D. J. & Sciortino, F. Assembly of clathrates from tetrahedral patchy colloids with narrow patches. J. Chem. Phys. 151, 094502 (2019). Manoharan, V. N., Elsesser, M. T. & Pine, D. J. Dense packing and symmetry in small clusters of microspheres. Science 301, 483–487 (2003). Hynninen, A.-P., Thijssen, J. H. J., Vermolen, E. C. M., Dijkstra, M. & van Blaaderen, A. Self-assembly route for photonic crystals with a bandgap in the visible region. Nat. Mater. 6, 202–205 (2007). Zanjani, M. B., Jenkins, I. C., Crocker, J. C. & Sinno, T. Colloidal cluster assembly into ordered superstructures via engineered directional binding. ACS Nano 10, 11280–11289 (2016). Wang, Y., Jenkins, I. C., McGinley, J. T., Sinno, T. & Crocker, J. C. Colloidal crystals with diamond symmetry at optical lengthscales. Nat. Commun. 8, 14173 (2017). Ducrot, É., He, M., Yi, G. R. & Pine, D. J. Colloidal alloys with preassembled clusters and spheres. Nat. Mater. 16, 652–657 (2017). Liu, W. et al. Diamond family of nanoparticle superlattices. Science 351, 582–586 (2016). Lin, H. et al. Clathrate colloidal crystals. Science 355, 931–935 (2017). Romano, F. & Sciortino, F. Patterning symmetry in the rational design of colloidal crystals. Nat. Commun. 3, 975 (2012). Damasceno, P. F., Engel, M. & Glotzer, S. C. Crystalline assemblies and densest packings of a family of truncated tetrahedra and the role of directional entropic forces. ACS Nano 6, 609–614 (2012). Perro, A. et al. A chemical synthetic route towards “colloidal molecules”. Angew. Chem. Int. Ed. 48, 361–365 (2009). Sacanna, S. & Pine, D. J. Shape-anisotropic colloids: building blocks for complex assemblies. Curr. Opin. Colloid Interface Sci. 16, 96–105 (2011). Grünwald, M. & Geissler, P. L. Patterns without patches: hierarchical self-assembly of complex structures from simple building blocks. ACS Nano 8, 5891–5897 (2014). McGinley, J. T., Wang, Y., Jenkins, I. C., Sinno, T. & Crocker, J. C. Crystal-templated colloidal clusters exhibit directional DNA interactions. ACS Nano 9, 10817–10825 (2015). Ducrot, É., Gales, J., Yi, G.-R. & Pine, D. J. Pyrochlore lattice, self-assembly and photonic band gap optimizations. Opt. Express 26, 30052–30060 (2018). Gong, Z., Hueckel, T., Yi, G. R. & Sacanna, S. Patchy particles made by colloidal fusion. Nature 550, 234–238 (2017).

33. Schade, N. B. et al. Tetrahedral colloidal clusters from random parking of bidisperse spheres. Phys. Rev. Lett. 110, 148303 (2013). 34. Wang, Z. et al. Active patchy colloids with shape-tunable dynamics. J. Am. Chem. Soc. 141, 14853–14863 (2019). 35. Agard, N. J., Prescher, J. A. & Bertozzi, C. R. A strain-promoted [3 + 2] azide-alkyne cycloaddition for covalent modification of biomolecules in living systems. J. Am. Chem. Soc. 126, 15046–15047 (2004). 36. Anderson, J. A., Lorenz, C. D. & Travesset, A. General purpose molecular dynamics simulations fully implemented on graphics processing units. J. Comput. Phys. 227, 5342–5359 (2008). 37. Glaser, J. et al. Strong scaling of general-purpose molecular dynamics simulations on GPUs. Comput. Phys. Commun. 192, 97–107 (2015). 38. Johnson, S. & Joannopoulos, J. Block-iterative frequency-domain methods for Maxwell’s equations in a planewave basis. Opt. Express 8, 173–190 (2001). 39. Lee, S., Zheng, C. Y., Bujold, K. E. & Mirkin, C. A. A cross-linking approach to stabilizing stimuli-responsive colloidal crystals engineered with DNA. J. Am. Chem. Soc. 141, 11827–11831 (2019). 40. Imhof, A. & Pine, D. J. Ordered macroporous materials by emulsion templating. Nature 389, 948–951 (1997). 41. Wijnhoven, J. E. G. J. & Vos, W. L. Preparation of photonic crystals made of air spheres in titania. Science 281, 802–804 (1998). 42. Holland, B. T., Blanford, C. F. & Stein, A. Synthesis of macroporous minerals with highly ordered three-dimensional arrays of spheroidal voids. Science 281, 538–540 (1998). 43. von Freymann, G. et al. Three-dimensional nanostructures for photonics. Adv. Funct. Mater. 20, 1038–1052 (2010). 44. Liu, L., Karuturi, S. K., Su, L. T. & Tok, A. I. Y. TiO2 inverse-opal electrode fabricated by atomic layer deposition for dye-sensitized solar cell applications. Energy Environ. Sci. 4, 209–215 (2011). 45. Gratson, G. M. et al. Direct-write assembly of three-dimensional photonic crystals: Conversion of polymer scaffolds to silicon hollow-woodpile structures. Adv. Mater. 18, 461–465 (2006). 46. van Blaaderen, A., Ruel, R. & Wiltzius, P. Template-directed colloidal crystallization. Nature 385, 321–324 (1997). 47. Leal-Calderon, F., Mondain-Monval, O., Pays, K., Royer, N. & Bibette, J. Water-in-oil emulsions: role of the solvent molecular size on droplet interactions. Langmuir 13, 7008–7011 (1997). 48. Chen, Q., Bae, S. C. & Granick, S. Directed self-assembly of a colloidal kagome lattice. Nature 469, 381–384 (2011). 49. Bonn, D. et al. Direct observation of colloidal aggregation by critical Casimir forces. Phys. Rev. Lett. 103, 156101 (2009). Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. © The Author(s), under exclusive licence to Springer Nature Limited 2020

Nature | Vol 585 | 24 September 2020 | 529

Article Methods Synthesis of liquid cores Monodisperse oil droplets50 are prepared by adding 200 μl of NH3 (28 wt%) to 100 ml deionized water, followed by the addition of 166 μl of a 4:1 (v/v) mixture of 3-(trimethoxysilyl)propyl methacrylate (≥98%, Sigma-Aldrich) and (3-glycidyloxypropyl) trimethoxysilane (≥98%, Sigma-Aldrich). This mixture is kept under mild magnetic stirring for 2 h at room temperature to allow the oil droplets to nucleate and grow to a final diameter of approximately 500 nm. If the droplet size deviates substantially for the target size, the procedure is repeated. To limit diffusion of TPM into polystyrene during the THF-induced compression, we reduce the solubility of the TPM in the THF-plasticized polystyrene by letting the condensation of TPM in ammonia proceed for a longer period of time, thus obtaining higher-molecular-weight oligomers. However, as the molecular weight of TPM increases, its viscosity also increases. If it becomes too viscous, the TPM oil cannot be extruded to form suitable patches, so this limits the extent to which we can polymerize the TPM. We find that 2–3 h of condensation gives the best results. Next, the emulsion is fluorescently labelled using rhodamine-B isothiocyanate (RITC). The dye is bonded to the oil droplets via siloxane linkages using 3-aminopropyl trimethoxysilane as a coupling agent (RITC-APS). Typically, 500 μl of surfactant (Pluronics F-108, 2 wt% in water) are added to the emulsion, followed by 100 μl of dye solution (1 mg ml−1 RITC-APS in DMSO). The dyed emulsion is then gently sedimented by centrifugation and resuspended in 40 ml of deionized water. Synthesis of polystyrene spheres The polystyrene particles are prepared by surfactant-free emulsion polymerization. Typically, 50 ml of styrene monomer (≥99%, Sigma-Aldrich) is first passed through an activated aluminium oxide column to remove inhibitor, and then added to 350 ml of deionized water in a batch reactor equipped with a mechanical stirrer and a reflux condenser. The reactor is then purged with nitrogen and set to a temperature of 70 °C while keeping the styrene emulsified under vigorous stirring. The polymerization starts with the addition of the radical initiator to the reactor (400 mg of potassium persulfate dissolved in 15 ml of deionized water) and continues overnight at 70 °C under mild stirring. After approximately 12 h, the mixture is brought to room temperature and the particles set in 400 ml of deionized water via repeated cycles of centrifugation and resuspension. This procedure yields particles with a diameter of approximately 550 nm. Larger particles are obtained by repeated seeded growth. Each growing step follows the polymerization protocol described above, except that the initial reaction mixture contains an additional 40 ml of seed suspension (10 wt%). Cluster assembly Tetrahedral clusters are assembled by combining oil droplets and polystyrene32. In a typical experiment, two salt solutions (4 ml 500 mM NaCl) are prepared: one is added to 40 ml of the liquid cores (0.1 wt%) emulsion and the other to 40 ml of the polystyrene suspension (10 wt%). The emulsion is then added dropwise to the polystyrene suspension under continuous magnetic stirring. Once the two suspensions are combined, a surfactant solution (500 μl 2 wt% F-108 solution) is added to the mixture to stabilize the newly assembled clusters. Tetrahedral clusters are isolated from the excess polystyrene particles by isopycnic centrifugation. In a typical purification step, 700 parts of the cluster mixture is mixed with 625 parts of 44 wt% glycerol solution and centrifuged at 2,400g for 90 min. A successful centrifugation results in the sedimentation of the clusters at the bottom of the centrifuge tube. The sediment is then resuspended in about 4 ml of deionized water and further purified by density gradient purification16. Finally, the purified clusters are washed and resuspended in 500 μl of deionized water.

Cluster compression and polymerization In a typical deformation experiment, 325 μl THF, 675 μl deionized water, 150 μl 1 wt% dodecyltrimethylammonium bromide (DTAB) solution and 500 μl of cluster solution (the clusters consist of 1.0-μm polystyrene spheres and a 500-nm polymerizable TPM oil droplet, 2 wt%) are mixed together. The amount of THF can be tuned to control the extent of cluster deformation. The deformed clusters are fixed in a mixture of 15 ml deionized water, containing 100 μl 1 wt% F-127 and 40 μl 5 wt% SDS, resulting in a cluster concentration of around 0.06 wt%. The F-127 stabilizes the clusters against aggregation. The SDS controls the wetting between the TPM and the polystyrene spheres. F-127 can also be used to control the wetting, but using SDS suppresses the amount of TPM that migrates to the polystyrene surfaces. The size ratio of the clusters strongly depends on the type and composition of the surfactant solution used to fix them (Extended Data Fig. 1). The last step in the preparation of compressed clusters is the polymerization of the liquid core, which is carried out at 80 °C in the presence of 5 mg AIBN as a radical initiator. Polymerized clusters are washed and resuspended in 1 wt% F-127 solution. DNA functionalization Single-stranded DNA (5′-DBCO-Cy5-T50-TTTACGCGTA-3′) was purchased from Integrated DNA Technologies USA. The DNA strand comes equipped with a dibenzyl cyclooctane (DBCO) group and a fluorescent tag (Cy5, emission maximum 668 nm) and was pre-purified via high-performance liquid chromatography. The DNA is dissolved in standard PBS buffer to a concentration of 100 μM and stored at −4 °C. The compressed clusters (1 ml of suspension at 1 wt% clusters and 1 wt% F-127) are pre-treated with 10 mg of sodium azide (NaN3) and a catalytic amount of potassium iodide (KI). This mixture is kept at 70 °C for 24 h. The activated particles are then washed and stored in water. To graft DNA onto the surface of the compressed clusters, 20 μl of cluster suspension is added to 400 μl PBS solution (PBS, 10 mM, pH = 7.4, 500 mM NaCl) containing 0.1 wt% Triton X-100 and 10 μl of the DNA solution. This mixture is incubated at 55 °C for 2 days before washing the particles in deionized water. When properly stored at 4 °C in PBS buffer, our colloidal systems can be stored for several months without any loss of stability of functionality. Crystallization Colloidal crystals are assembled in a 50/50 mixture of D2O and PBS buffer solution (PBS, 10 mM, pH = 7.4, 280 mM NaCl, Pluronic F-127 1 wt%). The suspension is introduced into a glass capillary (100 μm × 2 mm × 5 cm, VitroCom). The capillary is pretreated with oxygen plasma, exposed to hexamethyldisilazane (HMDS) vapours, flushed with aqueous 1 wt% F127 solution and finally dried with compressed nitrogen. Filled capillaries are sealed using NOA 68 UV glue and mounted on a custom-made temperature-controlled microscope stage. The DNA melting temperature (Tm) is around 40 °C, which varies slightly from batch to batch. Our thermal stage allows us to set a temperature gradient across the capillary (the temperature difference between the two ends is about 1 °C). The sample is also tilted at an angle of about 20° to increase the particle concentration during annealing. The tilting and temperature-gradient directions are perpendicular to each other. The shaped particles with interlock result in particle volume fractions around 0.68, not far below the close-packing limit of 0.74 and well above the 0.34 packing fraction of a diamond lattice of touching spheres. Fixing and drying of the crystals 8-methoxypsoralen (8-MOP) is chosen as the ultraviolet crosslinker to fix the colloidal crystals. Before annealing the sample, 8-MOP is pre-dissolved into the PBS (PBS, 10 mM, pH = 7.4, 280 mM NaCl, Pluronic F-127 1 wt%) solution. After crystallization, the sample is exposed to utlraviolet light for 6 h. Then, both ends of the glass capillary are

carefully opened and the sample is flushed gently on to a clean glass coverslip with PBS solution (PBS, 10 mM, pH = 7.4, 280 mM NaCl, Pluronic F-127 1 wt%). The sample is washed with deionized water and left on the bench and dried overnight.

Photonic bandgap calculation We performed a series of photonic bandstructure calculations using the MIT Photonic Bands software, determining the bands along the high-symmetry points of the Brillouin zone. The direct and inverse cluster diamond lattices both have complete photonic bandgaps between the second and third bands for a range of compression ratios. Photonic band calculations were performed for each of the five lowest energy bands at 223 k-vectors, with the primitive unit cell discretized on a 32 × 32 × 32 grid. The cluster diamond is constructed from the face-centred-cubic Bravais lattice vectors, with a two-cluster basis. The only bandgap that appears among the 50 bands calculated is the one shown, which occurs between the second and third bands. Finite-difference time domain (FDTD) transmission simulations were performed to determine whether the bandgap of the inverse cluster diamond is robust to the disorder caused by polydispersity in the size of each sphere that comprises a cluster. Transmission of a plane-wave source was simulated through 8-unit-cell-thick crystals with polydispersity in particle diameter up to ±5%, exceeding any polydispersity used in experiment. Even the largest disorder studied has only a very minor effect on the transmission, preserving the bandgap. Simulation The tetrahedral particles generated in the laboratory can be described as four partially overlapping spheres arranged with tetrahedral symmetry, with patches located in the middle of the facets. In the experiment, the patches are coated with DNA strands, giving rise to an attractive interaction between patches, while the rest of the particle is repulsive. We model the attractive patches as being the exposed surface of a sphere whose centre is located at the centre of mass of the tetrahedral cluster. Two parameters describe the tetrahedral patchy particles: the compression ratio dcc/(2a) and the size ratio b/a (Fig. 1). Experimentally, it is possible to vary dcc/(2a) substantially, as well as b/a to a lesser extent, which defines a substantial phase space. To refine the domain over which the diamond lattice self-assembles, we ran computer simulations using the HOOMD-blue simulation package36,37. The simulations are performed using a short-range attractive Lennard–Jones potential Up between patches, given by Up(r) = 4ε[(σ/r)2n − (σ/r)n] with n = 24, where r is the distance between the centres of the spheres and σ is the radius of the spheres. This places the minimum in the potential at r = 21/nσ = 1.03σ. The energy scale is set by the well depth ε, which we arbitrarily set to be 10 for the attractive interaction between the DNA-coated patches. To capture the weak attractive depletion interaction between lobes due to the presence of F-127 micelles and a small amount of DNA, we use a Lennard–Jones potential with n = 24 and ε = 3. The interaction between the lobes and DNA-coated patches (which suppresses the depletion interaction) is modelled by a short-range repulsive WCA (Weeks–Chandler–Andersen) potential Uc, given by Uc(r) = 4ε[(σ/r)2n −  (σ/r)n + 1/4] for r ≤ 21/nσ and Uc(r) = 0 for r > 21/nσ, with n = 24 and ε = 10. For these parameters, the melting temperature Tm in energy units is in

the range 1.6–1.7. We explored the phase diagram for a shorter-range potential with n = 48. In a typical simulation run, 216–8,000 particles in a periodic box at 5 vol% are slowly cooled in the vicinity of the aggregation temperature. The final system is analysed using the open-source visualization tool OVITO to discriminate random aggregates from the formation of any crystalline phases51. The results of the simulations are summarized in Extended Data Fig. 3.

Microscopy Bright-field optical images and videos are obtained using a Nikon Eclipse TiE microscope equipped with a CCD camera. Fluorescent images and videos are taken using a Leica SP8 confocal fluorescence microscope. An oil with n = 1.59 (purchased from SPI supplies, CargilleBrand Refractive Index Fluid Series A) was added to dried samples, which were used for the z-stack confocal microscopy. SEM images are taken using Merlin field-emission SEM. All crystals appeared to be isotropic. In some cases, however, the crystals cracked along grain boundaries during drying.

Data availability All data that support the findings are available from the corresponding authors on reasonable request.

Code availability The HOOMD-blue simulation package used to determine the phase diagram of the tetrahedral patchy particles36,37 and the MIT Photonic Bands software used to calculate the photonic bandstructure calculations38 are publicly available. 50. van der Wel, C. et al. Preparation of colloidal organosilica spheres through spontaneous emulsification. Langmuir 33, 8174–8180 (2017). 51. Stukowski, A. Visualization and analysis of atomistic simulation data with OVITO–the open visualization tool. Model. Simul. Mater. Sci. Engin. 18, 015012 (2010). Acknowledgements This research was primarily supported by the US Army Research Office under award number W911NF-17-1-0328. Additional funding was provided by the National Science Foundation under award number DMR-1610788. G.-R.Y. acknowledges support from the NRF (South Korea) under award number 2017M3A7B8065528. We acknowledge the use of shared facilities provided through the Materials Research Science and Engineering Center (MRSEC) programme of the National Science Foundation under award number DMR-1420073. The computational work was supported in part through the NYU IT High Performance Computing resources, services and staff expertise. Author contributions M.H. designed the synthetic protocol, synthesized and crystallized the patchy colloidal clusters, dried the crystals, and performed the optical and electron microscopy. M.H. and Z.G. synthesized the spherical patchy colloidal particles that led to the patchy colloidal cluster idea. É.D. performed the simulations and J.P.G. performed the photonic bandgap calculations that contributed to the design of the photonic crystals. D.J.P. and S.S. conceived the study and supervised the research, with the help of G.-R.Y. The manuscript was written by D.J.P., M.H. and J.P.G. All authors discussed the results and commented on the manuscript. Competing interests The authors declare no competing interests. Additional information Supplementary information is available for this paper at https://doi.org/10.1038/s41586-0202718-6. Correspondence and requests for materials should be addressed to S.S. or D.J.P. Reprints and permissions information is available at http://www.nature.com/reprints.

Article Compression Ratio = 1

a

Compression Ratio = 0.69

b F-127

d

Compression Ratio = 0

c SDS

e

Extended Data Fig. 1 | Controlling compression and size ratios. a–c, The compression ratio of clusters can be changed by adding differing amounts of THF in the deformation step: a, no THF; b, 18 vol% THF; c, 30 vol% THF. d–f, The

Triton X-100

f size ratio can be controlled by using different surfactants in the polymerization step: d, 0.05 wt% F-127; e, 0.05 wt% F-127 + 0.05 wt% SDS; f, 0.05 wt% F-127 + 0.05 wt% Triton X-100. Scale bars, 1 μm.

Extended Data Fig. 2 | Fluorescent microscope image of DNA-coated compressed tetrahedral clusters. DNA strands are fluorescently labelled. The image shows that the single-stranded DNA can be selectively functionalized onto the TPM patches and that the polystyrene surfaces are nearly bare. Note that the polystyrene surfaces need not be completely bare of DNA strands: the areal coverage of DNA just has to be much lower on the polystyrene surfaces than on the TPM patches, so that the DNA melting temperatures for the sphere–patch or sphere–sphere interactions are lower than for the patch–patch interaction. The melting temperatures for the polystyrene–polystyrene and polystyrene–TPM interactions are always 6–20 °C below that of the TPM–TPM interaction. Inset, illustration of DNA-coated compressed clusters: red, DNA-coated TPM patches; white, polystyrene spheres not coated with DNA. Scale bar, 5 μm.

Article a

Extended Data Fig. 3 | Self-assembly of DNA-coated compressed clusters. a, Simulation data (diamond symbols) are obtained using the HOOMD-blue software package; experimental data (circles) are superimposed. A typical simulation run is performed with 216–8,000 particles in a box with periodic boundaries at a volume fraction of 5%. The system is slowly cooled in the vicinity of the aggregation temperature. The final system is analysed to detect and characterize the formation of a potential crystalline phase. The red diamonds represent where the particles crystallize into cubic diamond; the grey diamonds represents where the system condenses into amorphous structures. The attractive DNA interaction is modelled by a Lennard–Jones potential that has its minimum at 1.03b. Using a shorter-range attractive

b

potential between patches moves the region where cubic diamond crystals form to larger size ratios; a longer-range potential does the opposite. The grey line shows the locus of size and compression ratios where all three spherical lobes and the patch on neighbouring particles simultaneously touch in the staggered conformation. The blue circles represent experimental samples that either crystallized in the cubic diamond structure (closed circles; error bars ±0.02 in both directions) or formed random aggregates (open circles; similar uncertainties but with the error bars suppressed). b, A snapshot of the simulated system where the compressed clusters crystallized into cubic diamond crystals (dcc/(2a) = 0.70).

a

b

c

d

e

f

g

h

Extended Data Fig. 4 | Self-assembly of DNA-coated compressed clusters. a–d, Optical microscope bright-field images of two-particle (a) and four-particle (b) aggregates, the 110 plane of a small crystal (c) and the 111 plane of the crystal (d). e–h, Corresponding computer-generated images. Scale bars, 1 μm.

Article a

b

c

d

fc

Δf

Extended Data Fig. 5 | Inverse cubic diamond lattice of clusters. a, Structure for a compression ratio of 0.76. The size ratio is chosen so that patches just touch. The volume fraction of solid material is 0.32. b, Same as in a, but made using a protective oxide layer with a thickness of 10% of the radius of the spherical lobe of a cluster. The volume fraction of solid material is 0.20. c, Band diagram for an inverse diamond lattice of clusters with a compression ratio of 0.76 (the structure shown in a), showing a complete photonic bandgap with a

relative width of 0.12 between the second and third bands (highlighted in blue). The vertical axis is the dimensionless frequency fa/c, where f is frequency, a is lattice constant and c is the speed of light in vacuum. Only the first five bands are shown, but the first 50 were calculated and no other bandgaps were found. d, Minimum index nmin at which a bandgap opens for a range of compression ratios: direct lattice, blue; inverse lattice, red.

Article

Light-driven post-translational installation of reactive protein side chains https://doi.org/10.1038/s41586-020-2733-7 Received: 8 November 2019 Accepted: 15 July 2020 Published online: 23 September 2020

Brian Josephson1,7, Charlie Fehl1,5,7, Patrick G. Isenegger1,7, Simon Nadal1, Tom H. Wright1,6, Adeline W. J. Poh1, Ben J. Bower1, Andrew M. Giltrap1,2, Lifu Chen3, Christopher Batchelor-McAuley3, Grace Roper1, Oluwatobi Arisa1, Jeroen B. I. Sap1, Akane Kawamura1, Andrew J. Baldwin1, Shabaz Mohammed1,2,4, Richard G. Compton3, Veronique Gouverneur1 ✉ & Benjamin G. Davis1,2 ✉

Check for updates

Post-translational modifications (PTMs) greatly expand the structures and functions of proteins in nature1,2. Although synthetic protein functionalization strategies allow mimicry of PTMs3,4, as well as formation of unnatural protein variants with diverse potential functions, including drug carrying5, tracking, imaging6 and partner crosslinking7, the range of functional groups that can be introduced remains limited. Here we describe the visible-light-driven installation of side chains at dehydroalanine residues in proteins through the formation of carbon-centred radicals that allow C–C bond formation in water. Control of the reaction redox allows site-selective modification with good conversions and reduced protein damage. In situ generation of boronic acid catechol ester derivatives generates RH2C• radicals that form the native (β-CH2–γ-CH2) linkage of natural residues and PTMs, whereas in situ potentiation of pyridylsulfonyl derivatives by Fe(ii) generates RF2C• radicals that form equivalent β-CH2–γ-CF2 linkages bearing difluoromethylene labels. These reactions are chemically tolerant and incorporate a wide range of functionalities (more than 50 unique residues/side chains) into diverse protein scaffolds and sites. Initiation can be applied chemoselectively in the presence of sensitive groups in the radical precursors, enabling installation of previously incompatible side chains. The resulting protein function and reactivity are used to install radical precursors for homolytic on-protein radical generation; to study enzyme function with natural, unnatural and CF2-labelled post-translationally modified protein substrates via simultaneous sensing of both chemo- and stereoselectivity; and to create generalized ‘alkylator proteins’ with a spectrum of heterolytic covalent-bond-forming activity (that is, reacting diversely with small molecules at one extreme or selectively with protein targets through good mimicry at the other). Post-translational access to such reactions and chemical groups on proteins could be useful in both revealing and creating protein function.

Methods that use the translational machinery of the cell provide powerful advantages for installing selected modifications into proteins, but can be limited in scope and efficiency8–10. Unnatural amino acid precursors can be degraded or may not be tolerated during biosynthesis; this is especially true for those with reactive side chains11. Post-translational functionalization12–16 offers an alternative strategy that, through its late-stage use, could be broader in scope; in principle, it is limited only by the compatibility of the reaction conditions used with the protein substrate and its context. In one version12–15,17 of post-translational functionalization, a readily generated dehydroalanine (Dha) residue is used in proteins as a singly

occupied molecular orbital (SOMO) acceptor (‘radical acceptor’ or ‘SOMOphile’) that is highly reactive towards several carbon-centred radical species, thereby allowing selective β,γ-C–C bond formation to introduce new side chains in a constitutionally ‘scarless’/‘traceless’ manner (Extended Data Fig. 1). However, incompatibilities of side-chain/carbon radical precursors and the reagents that generate them (for example, single-electron transfer from metals or BH4−)14,15 currently limit reaction scope. Nonetheless, such homolytic single-electron (1e−) chemistry has potential advantages over typical heterolytic two-electron (2e−) reagents. The intrinsic challenges18 of biomolecule modification include: water compatibility; requirement for ‘benign-ness’ (being

1 Chemistry Research Laboratory, Department of Chemistry, University of Oxford, Oxford, UK. 2The Rosalind Franklin Institute, Harwell, UK. 3Physical and Theoretical Chemistry Laboratory, Department of Chemistry, University of Oxford, Oxford, UK. 4Department of Biochemistry, University of Oxford, Oxford, UK. 5Present address: Department of Chemistry, Wayne State University, Detroit, MI, USA. 6Present address: Department of Molecular Biology, Massachusetts General Hospital, Boston, MA, USA. 7These authors contributed equally: Brian Josephson, Charlie Fehl, Patrick G. Isenegger. ✉e-mail: [email protected]; [email protected]

530 | Nature | Vol 585 | 24 September 2020

Dual modes of light-driven alkyl radical generation

Diverse protein scaffolds and sites

R′ = H (catechol), CH2CH2NH2 (dopamine), CH2CH(NH 2)COOH (L-DOPA) HO

R′

O

F

+

Fe(II)

N

B R

O

O

In situ generated BACED Radical generation

R O

pySOOF derivatives

Blue LED light (450 nm) Mild, water-soluble RuII catalysts

R

F

S

H

H

F

or

F

High concentration needed Native residue/PTM precursor

Histone H3 Two variants (Hs, Xl) Sites: K4, K9, K18, K27

Histone H4 Site: K16

AcrA Site: N123

PanC Site: H44

NPβ Site: M61

cabLys3 Site: A104

Radical generation

R Low concentration needed ‘Zero-size’-labelled difluoro analogue precursor

N H Dha

Dha Radical generation

Broad functional group tolerance

X = H or F

O

R

Quenching

X

R′

N3

O Amides

X

Azides O

Aqueous buffer (degassed), RT, pH 6–8, 5–120 min, blue LED light (450 nm) Protein containing dehydroalanine (Dha) at desired site

H N

X

70–95% conversion >50 unique side chains Multiple protein folds, sites and scaffolds

Halogens X = F, Cl, Br, I

O

R′

Ethers

O O

R′

Esters

O S

N

On-protein radical precursor

S(II/IV/VI)

Sulfides/oxides/ ones/ates

NMen Methylamines n = 0, 1, 2, 3

Fig. 1 | Site-selective, light-driven post-translational protein editing. The excitation of mild, water-soluble, protein-compatible RuII photocatalysts with blue light enables dual modes of alkyl radical generation from complementary radical precursors with low oxidative exposure of protein substrates (Eox 50 unique native or difluoro-labelled side chains in proteins; see Extended Data Fig. 8). RT, room temperature; LED, light-emitting diode.

non-destructive to their biological substrates); and low (or non-) reactivity towards a plethora of biogenic acids, amines, alcohols and thiols (ready 2e− reactants) that are present in most biological environments (Extended Data Fig. 1a). By contrast, water and native proteins are less reactive to most19 carbon-centred radicals; therefore, suitably placed SOMOphiles such as Dha can allow more general chemo- and site-selectivity in certain 1e− chemistries (Extended Data Fig. 1b). Other methods for single-electron transfer (and hence carboncentred radical initiation, oxidative or reductive) exist. Catalytic protein methods, in particular, could bring clear advantages20 over previous super-stoichiometric methods (which can drive unwanted side reactions; Extended Data Fig. 1c–e). Furthermore, if regulated by a relatively benign, potentially tissue-penetrating trigger such as light21, such methods could allow additional layers of, for example, temporal, spatial and even kinetic control to complement those of 1e− chemoselectivity. Light-stimulated outer-sphere electron transfer has seen a resurgence in applications to small molecules22,23. However, its use in site-selective, biomolecule modification has been more limited. Leading examples (Extended Data Fig. 1d, e and Supplementary Discussion 1) have largely been restricted to peptides24–26, sometimes requiring mixed organic solvents26 and/or electron-transfer systems that sit towards the extremes of redox ‘windows’, and resulting side reactions have been observed25. Moreover, dependence on certain precursor moieties, such as α-C-carboxyl24 or β-C–H (ref. 25), that cannot be readily re-/pre-positioned, can limit the reaction site and/or lead to lower site-selectivity owing to abundance. These methods have therefore yet to reach their full potential in protein chemistry. Here, we show that a combination of (i) electron transfer at benign, moderate redox potentials using (ii) side-chain carbon radical precursors ‘redox-matched’ with low, even substoichiometric, amounts

of photocatalyst, triggered by (iii) light of appropriate flux, allows the generation and use of both off-protein and on-protein radicals to modify proteins via C–C bond formation (Fig. 1).

Results Photocatalytic carbon-centred radical protein modification. Exploitation of carbon-centred radicals could involve either off- or on-protein radicals with reductive or oxidative initiation. A pre-positioned on-protein SOMOphile, such as Dha, allows the flexibility of off-protein carbon-centred radical generation by either method (Fig. 1, Extended Data Fig. 2). Initial scoping with photocatalysts covering a wide redox spectrum (Extended Data Fig. 2a) under varying aqueous reaction conditions (aerobic/anaerobic, pH, co-solvent, redox mediators, light flux) avoided the use of organic co-solvents or extremes of pH, because these are typically incompatible with many full-length proteins (see Supplementary Discussion 1, Supplementary Tables 1–3 and Extended Data Fig. 3a, b). These experiments revealed: (a) an effectiveness of catechol beyond hydrogen atom transfer27 (as noted previously for organosilicates28) in oxidative activation of alkyl organoboronates and (b) a relative ineffectiveness of alkylhalides as reductive precursors. Both observations suggested avenues for improvement. The potentiation of (and activation of previously unreactive) alkyl boronates by catechol using low-*Eox/*Ered catalysts was surprising, because *Eox-catalyst > Eox-substrate typically determines reactivity; here E represents the oxidation (ox) or reduction (red) potential versus a saturated calomel electrode (SCE) and the asterisk denotes the excited state of the photocatalyst. Primary C–B bonds (≥+1.5 V) were previously inaccessible. However, with catechol, even the challenging substrate PhCH2CH2–BF3K (Eox > +1.6 V) proved to be reactive not only with Cat3 (*Eox = +1.32 V) and Cat4 (Ru(bpz)3; *Eox = +1.45 V), but also with the much Nature | Vol 585 | 24 September 2020 | 531

Article weaker catalysts Cat1 (Ru(bpy)3; *Eox = +0.77 V) and Cat2 (Ru(bpm)3, *Eox = +0.99 V). These milder catalysts gave enhanced conversion: for example, >90% H3-homohomoPhe9 (H3-1h; see Extended Data Fig. 8 for side-chain glossary) from H3-Dha9. The action of catechol was inconsistent with hydrogen atom transfer (HAT)27 alone (Extended Data Fig. 3a and Supplementary Table 4). We tested three mechanistic possibilities (Extended Data Fig. 4 and Supplementary Discussion 2): mediated electron transfer, catalyst modification and substrate modification. (During the course of this work, the beneficial effects of catechols were also independently observed in small-molecule systems; see ref. 29) First, from analogues and potential redox mediators (Extended Data Fig. 4a–c) only aromatic 1,2-vicinal diols potentiated. Second, the Ru complex30 from ligand-exchange catecholo-Ru(bpy)2-Cat6 (Extended Data Fig. 4d) displayed no activity. Third, use of pre-formed boronic acid catechol esters revealed efficient conversion, even without exogenous catechol (Extended Data Fig. 4e, f). Moreover, cyclic voltammetry (Extended Data Fig. 4g–l and Extended Data Fig. 2b inset) revealed a concentration-dependent, shifted boronate Eox that brought these carbon-centred radical precursor substrates within range of the mild catalysts Cat1, Cat2 (see also Supplementary Discussion 3). Together, these data were consistent with in situ boronic acid catechol ester derivatives, termed BACED reagents (BACED, boronic acid catechol ester derivative; Fig. 1, Extended Data Fig. 2c), allowing efficient oxidative side-chain carbon radical generation via turnover of catechol in aqueous medium (even at 2 mol%; Supplementary Table 5) and lowered photocatalyst loadings (for example, from 100 equiv. to 25 mol% of Cat1; Supplementary Table 5) to rare20 substoichiometry. Next, the failure in scoping shown by simple alkylhalides as reductive carbon-centred radical precursors suggested substrate modulation. α-Fluoro substitution alters the stability and reactivity of carbon-centred radicals31 and increases the addition reactivity in water32. H → F variation might therefore give enhanced RF2C•/ RFHC• as near-size33 RH2C• equivalents, with the potential to generate near-‘zero-size’ H → F labels in protein side chains. From various potential RF2C• precursors34, scoping (Supplementary Discussion 4) revealed pySOO–CF2R pyridyl-sulfone-fluoride (pySOOF) reagents with suitable Ered (ref. 35) and enhanced reactivity (in 15 min). However, the initially observed products were consistent with partial, unwanted oxidative (instead of reductive) quenching of the resulting intermediate protein-α-carbon radical to hemiaminal/imine (for example, only 58% conversion to H3-DfeGly9 from H3-Dha using pySOO-CF2H; Extended Data Fig. 2d). Various reductive additives and HAT sources failed or led to other unwanted side reactions (Extended Data Fig. 5) but led us to consider direct electron transfer from metals (Extended Data Figs. 2d, 6); Zn(ii) and Mn(ii) gave no change and Ni(ii) and Cu(ii) inhibited the reaction, but Fe(ii) sources (preferably Fe(ii) sulfate) achieved good conversion (>92% to H3-DfeGly9; Extended Data Fig. 6). Notably, consistent with previous reactivity profiles31, mono-fluoro pySOOF reagent pySOO-CFH2 failed to react, even under optimized conditions (Extended Data Fig. 7), but pySOOF reagents bearing additional carboxylate or acetamide groups (pySOO-CHFCOOH and pySOO-CHFC(O)NH2 generating side chains 2s and 2u, respectively; Extended Data Fig. 8) displayed high reactivities (with only 10–25 equiv. pySOOF; Extended Data Fig. 7). Together, these and other (Extended Data Fig. 7) data were consistent with dual reductive quenching and potentiation by Fe(ii) allowing efficient reductive side-chain carbon-centred radical generation and, again, rare substoichiometry, not only of the photocatalyst but also Fe(ii) (for example, 25 mol% Cat1, 50 mol% FeSO4, 10 equiv. pySOOF, 66% conversion; Supplementary Tables 8–22 and Extended Data Fig. 6). Optimization of BACED and pySOOF reagents. Together, these mechanistic studies revealed mild, efficient complementary pathways using BACED or pySOOF reagents (Fig. 1), potentiated by catechol or 532 | Nature | Vol 585 | 24 September 2020

Fe(ii), respectively. The wide availability of boronic acids (directly as BACED reagents) or pyridylsulfones (allowing ready synthesis of pySOOF reagents; see Supplementary Methods) enabled rapid, broad-scope optimization and application to introduce native and difluorinated amino acid side chains and/or side chains bearing post-translational modifications into proteins (Extended Data Fig. 8). Reaction times were shortened by investigating the light flux in a variable-intensity photoreactor (Extended Data Fig. 3d); use of 50 W power and ~450 nm wavelength reduced the reaction times (from typically 4 h to  6.6 kJ mol−1 (where ΔΔG‡ denotes the difference in the change of Gibbs free energy on moving from the ground state to the transition state; Extended Data Fig. 10), despite the six-bond distance to Nε. To our knowledge, such simultaneous, real-time determinations of substrate- and stereoselectivity in intact proteins have not been previously possible. Finally, the sensitivity of the γ-F2 label could be applied to monitor differential folding and assembly states in a single protein: full step-wise assembly40 of H3-DfeGly9 histone into an octamer (unfolded-H3 monomer → folded-H3 monomer → (H3)2•(H4)2 hetero-tetramer → (H3)2•(H4)2 •(H2A)2•(H2B)2 hetero-octamer) was achieved even at microgram scales (Extended Data Fig. 10). Alkylator proteins trap buried protein–protein interfaces through mimicry. The electrophilic halide side chains included those with side-chain lengths that were well matched to Lys (bromonorleucine (Bnl, 1t), bromohomonorleucine (Bhn, 1u), iodonorleucine (Inl, 1s); Extended Data Fig. 8), which enabled the design of ‘protein alkylators’ with potential context-dependent reactivity based on Lys mimicry. If designed correctly, these would remain inactive under typical conditions in a biological mixture, but would then display enhanced alkylative reactivity in a ‘guided’ manner by virtue of solvent exclusion, effective molarity41,42 and proper mimicry when suitably engaged at a protein–protein interface (PPI). Such a system would require a critical balance in electrophilic reactivity and native shape fidelity (Extended Data Fig. 11a), which has been presciently highlighted as a key goal in protein science18 (see Supplementary Discussion 5). Site-selective insertion of the minimally sized alkylhalide side chains Bnl (1t), Bhn (1u) and Inl (1s) into proteins has not been previously possible. By mimicking the binding of Lys side chains more closely, it might be possible to probe even buried PPIs with reduced artefacts. We tested potential buried43 and transient (substrate•enzyme) PPIs using Bhn (1u) at three sites (4, 9, 27) that are normally occupied by Lys in C-terminally FLAG-HA-tagged histone eH3.1. When incubated with a partner enzyme that processes (and so binds) Lys, Lys-demethylase-KDM4A was observed to crosslink exclusively to Bhn-containing eH3.1-Bhn4, eH3.1-Bhn9 and eH3.1-Bhn27, but not to wild-type (WT) histone eH3.1 (Fig. 3c). The Lys-‘guided’ nature of this crosslinking was consistent with zero crosslinking from incubations with other proteins: neither with Cys-rich serum albumin nor with the known nucleosomal binding partner histone H4; the H3•H4 PPI does not involve key Lys4,9,27,44, whereas the H3•KDM4A PPI does43. This seemingly PPI-selective reaction was further confirmed by MS/MS analysis (Fig. 3c, Extended Data Fig. 11c) of crosslinking to KDM4A-Cys234,Cys30643 located in the buried Zn-binding domain in H3•KDM4A PPI, as well as by real-time fluorescent monitoring of Zn ejection (Fig. 3c, Extended Data Fig. 11b)45 by eH3.1-Bhn9 but not by WT-eH3.1. Moreover, when incubated with human-cell (HeLa) nuclear lysate, eH3.1-Bhn9 showed the ability to enhance capture of interaction partners via eH3.1-adduct formation (Extended Data Fig. 11e). The unusual reactivity of these alkylator proteins was further illustrated by the observation of an inter-protein Williamson-type (-C–O–C-) ether formation that, to our knowledge, is unprecedented16 (Extended Data Fig. 11c, d) and is a reaction with a typical rate that is seemingly too low (kapp  Cat1) to quench/terminate an on-protein α-carbon radical is an exciting additional observation, which suggests future design of gem,gem-α,α-bisalkylated motifs known50 to control stability and conformation. Our discovery of new crosslinks using previously inaccessible unnatural amino acids in proteins (for example, Bhn) suggests that precise mimicry of residues in PPIs may drive new chemistries, and hence selectivities. Whereas current crosslinking methods non-specifically ‘fix’ (for example, formaldehyde, bis-esters), and hence trap, complexes that are longer-lived or more favoured, our results here suggest that the future application of crosslinks could enable the exploitation of reactivity enhanced by effective molarity (see Supplementary Discussion 6) to precisely cross-react minimal, size-matched side chains (for example, Bhn) even in transient, reactive interactions. Importantly, in this scenario it is the relative rate enhancement caused by the environment that is important, rather than the inherent rate of reactivity. We speculate therefore that better crosslinking selectivities may now be designed around new amino acids with unusual chemistries (for example, Williamson ether formation) and counterintuitively lower (not higher) reactivities that work in rare, but more information-rich and relevant, contexts (for example, insertion into precisely matching PPIs that drive effective molarity). Moreover, this crosslinking has enabled complex, activity-based protein inhibition through the modification of conserved active-site residues (for example, covalent inhibition of KDM4A by eH3.1-Bhn9). In this way, one may consider future protein analogues—‘protein covalent inhibitors’—that act akin to emerging, targeted small-molecule covalent inhibitors51, but with enhanced potency and selectivity that exploits PPIs. The observed requirement for continual irradiation in our light-driven approach enables on–off temporal control (Extended Data Figs. 3e, 7). This suggests not only chemical precision in individual proteins but also potential future use of defined spatiotemporal irradiation21,52 to control states of protein ensembles over time and site (for example, tissues). When also paired with in cellulo generation of Dha53, the chemical side-chain versatility that we have observed could drastically improve our ability to probe complex biological systems with atomic precision using the tissue-penetrating trigger of light. To this future end, we have also shown that various biogenic catecholamines (Extended Data Fig. 3f) function in the BACED manifold. This, along with the endogenous presence of Fe(ii) for pySOOF reactivity, suggests promise for in vivo reactivity. Note added in proof: Two relevant papers relating to deboronative radical chain reactions54 and proximity-based covalent attachment of small molecules to proteins55 emerged during the latter stages of the review of this paper. 536 | Nature | Vol 585 | 24 September 2020

Online content Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41586-020-2733-7.

1.

2. 3.

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

12.

13. 14. 15. 16.

17.

18. 19. 20. 21.

22. 23.

24. 25. 26. 27. 28.

29.

30.

31. 32.

Walsh, C. T., Garneau-Tsodikova, S. & Gatto, G. J., Jr. Protein posttranslational modifications: the chemistry of proteome diversifications. Angew. Chem. Int. Ed. 44, 7342–7372 (2005). Deribe, Y. L., Pawson, T. & Dikic, I. Post-translational modifications in signal integration. Nat. Struct. Mol. Biol. 17, 666 (2010). Howard, C. J., Yu, R. R., Gardner, M. L., Shimko, J. C. & Ottesen, J. J. Chemical and biological tools for the preparation of modified histone proteins. Top. Curr. Chem. 363, 193–226 (2015). Yang, A., Cho, K. & Park, H.-S. Chemical biology approaches for studying posttranslational modifications. RNA Biol. 15, 427–440 (2018). Ducry, L. & Stump, B. Antibody−drug conjugates: linking cytotoxic payloads to monoclonal antibodies. Bioconjug. Chem. 21, 5–13 (2010). Hinner, M. J. & Johnsson, K. How to obtain labeled proteins and what to do with them. Curr. Opin. Biotechnol. 21, 766–776 (2010). Leitner, A. et al. Probing native protein structures by chemical cross-linking, mass spectrometry, and bioinformatics. Mol. Cell. Proteomics 9, 1634 (2010). Wang, L., Brock, A., Herberich, B. & Schultz, P. G. Expanding the genetic code of Escherichia coli. Science 292, 498 (2001). Dumas, A., Lercher, L., Spicer, C. D. & Davis, B. G. Designing logical codon reassignment – expanding the chemistry in biology. Chem. Sci. 6, 50–69 (2015). Chin, J. W. Expanding and reprogramming the genetic code. Nature 550, 53–60 (2017). Klemes, Y., Etlinger, J. D. & Goldberg, A. L. Properties of abnormal proteins degraded rapidly in reticulocytes. Intracellular aggregation of the globin molecules prior to hydrolysis. J. Biol. Chem. 256, 8436–8444 (1981). Chalker, J. M. & Davis, B. G. Chemical mutagenesis: selective post-expression interconversion of protein amino acid residues. Curr. Opin. Chem. Biol. 14, 781–789 (2010). Wright, T. H., Vallée, M. R. J. & Davis, B. G. From chemical mutagenesis to post-expression mutagenesis: a 50 year Odyssey. Angew. Chem. Int. Ed. 55, 5896–5903 (2016). Wright, T. H. et al. Posttranslational mutagenesis: a chemical strategy for exploring protein side-chain diversity. Science 354, aag1465 (2016). Yang, A. et al. A chemical biology route to site-specific authentic protein modifications. Science 354, 623–626 (2016). Tamura, T. & Hamachi, I. Chemistry for covalent modification of endogenous/native proteins: from test tubes to complex biological systems. J. Am. Chem. Soc. 141, 2782–2799 (2019). Wright, T. H. & Davis, B. G. Post-translational mutagenesis for installation of natural and unnatural amino acid side chains into recombinant proteins. Nat. Protoc. 12, 2243–2250 (2017). Sletten, E. M. & Bertozzi, C. R. Bioorthogonal chemistry: fishing for selectivity in a sea of functionality. Angew. Chem. Int. Ed. 48, 6974–6998 (2009). Imiołek, M. et al. Selective radical trifluoromethylation of native residues in proteins. J. Am. Chem. Soc. 140, 1568–1571 (2018). Isenegger, P. G. & Davis, B. G. Concepts of catalysis in site-selective protein modifications. J. Am. Chem. Soc. 141, 8005–8013 (2019). Lim, R. K. V. & Lin, Q. Photoinducible bioorthogonal chemistry: a spatiotemporally controllable tool to visualize and perturb proteins in live cells. Acc. Chem. Res. 44, 828–839 (2011). Twilton, J. et al. The merger of transition metal and photocatalysis. Nat. Rev. Chem. 1, 0052 (2017). Prier, C. K., Rankic, D. A. & MacMillan, D. W. C. Visible light photoredox catalysis with transition metal complexes: applications in organic synthesis. Chem. Rev. 113, 5322–5363 (2013). Bloom, S. et al. Decarboxylative alkylation for site-selective bioconjugation of native proteins via oxidation potentials. Nat. Chem. 10, 205 (2018). Yu, Y. et al. Chemoselective peptide modification via photocatalytic tryptophan β-position conjugation. J. Am. Chem. Soc. 140, 6797–6800 (2018). de Bruijn, A. D. & Roelfes, G. Chemical modification of dehydrated amino acids in natural antimicrobial peptides by photoredox catalysis. Chem. Eur. J. 24, 11314–11318 (2018). Povie, G. et al. Catechols as sources of hydrogen atoms in radical deiodination and related reactions. Angew. Chem. Int. Ed. 55, 11221–11225 (2016). Matsui, J. K., Lang, S. B., Heitz, D. R. & Molander, G. A. Photoredox-mediated routes to radicals: the value of catalytic radical generation in synthetic methods development. ACS Catal. 7, 2563–2575 (2017). Robole, Z. M., Rahn, K. L., Lampkin, B. J., Anand, R. K. & VanVeller, B. Tuning the electrochemical redox potentials of catechol with boronic acid derivatives. J. Org. Chem. 84, 2346–2350 (2019). Ghosh, T. et al. Mixed-ligand complexes of ruthenium(ii) containing new photoactive or electroactive ligands: synthesis, spectral characterization and DNA interactions. J. Biol. Inorg. Chem. 10, 496 (2005). Dolbier, W. R. Structure, reactivity, and chemistry of fluoroalkyl radicals. Chem. Rev. 96, 1557–1584 (1996). Zhang, L., Dolbier, W. R., Sheeller, B. & Ingold, K. U. Absolute rate constants of alkene addition reactions of a fluorinated radical in water. J. Am. Chem. Soc. 124, 6362–6366 (2002).

33. O’Hagan, D. Understanding organofluorine chemistry. An introduction to the C–F bond. Chem. Soc. Rev. 37, 308–319 (2008). 34. Lemos, A., Lemaire, C. & Luxen, A. Progress in difluoroalkylation of organic substrates by visible light photoredox catalysis. Adv. Synth. Catal. 361, 1500–1537 (2019). 35. Rong, J. et al. Radical fluoroalkylation of isocyanides with fluorinated sulfones by visible-light photoredox catalysis. Angew. Chem. Int. Ed. 55, 2743–2747 (2016). 36. Berlicki, L., Obojska, A., Forlani, G. & Kafarski, P. Design, synthesis, and activity of analogues of phosphinothricin as inhibitors of glutamine synthetase. J. Med. Chem. 48, 6340–6349 (2005). 37. Griller, D. & Ingold, K. U. Free-radical clocks. Acc. Chem. Res. 13, 317–323 (1980). 38. Chen, Y., Kamlet, A. S., Steinman, J. B. & Liu, D. R. A biomolecule-compatible visible-light-induced azide reduction from a DNA-encoded reaction-discovery system. Nat. Chem. 3, 146–153 (2011). 39. Huang, H. et al. Lysine benzoylation is a histone mark regulated by SIRT2. Nat. Commun. 9, 3374 (2018). 40. Dyer, P. N. et al. Reconstitution of nucleosome core particles from recombinant histones and DNA. Methods Enzymol. 375, 23–44 (2003). 41. Page, M. I. & Jencks, W. P. Entropic contributions to rate accelerations in enzymic and intramolecular reactions and the chelate effect. Proc. Natl Acad. Sci. USA 68, 1678–1683 (1971). 42. Krishnamurthy, V. M., Semetey, V., Bracher, P. J., Shen, N. & Whitesides, G. M. Dependence of effective molarity on linker length for an intramolecular protein−ligand system. J. Am. Chem. Soc. 129, 1312–1320 (2007). 43. Ng, S. S. et al. Crystal structures of histone demethylase JMJD2A reveal basis for substrate specificity. Nature 448, 87–91 (2007). 44. English, C. M., Adkins, M. W., Carson, J. J., Churchill, M. E. & Tyler, J. K. Structural basis for the histone chaperone activity of Asf1. Cell 127, 495–508 (2006).

45. Meeusen, J. W., Tomasiewicz, H., Nowakowski, A. & Petering, D. H. TSQ (6-methoxy8-p-toluenesulfonamido-quinoline), a common fluorescent sensor for cellular zinc, images zinc proteins. Inorg. Chem. 50, 7563–7573 (2011). 46. Freedman, H. H. & Dubois, R. A. An improved Williamson ether synthesis using phase transfer catalysis. Tetrahedr. Lett. 16, 3251–3254 (1975). 47. Mandal, S. et al. A review on the advancement of ether synthesis from organic solvent to water. RSC Adv. 6, 69605–69614 (2016). 48. Levin, M., Stark, M. & Assaraf, Y. G. The JmjN domain as a dimerization interface and a targeted inhibitor of KDM4 demethylase activity. Oncotarget 9, 16861–16882 (2018). 49. Shin, S. & Janknecht, R. Diversity within the JMJD2 histone demethylase family. Biochem. Biophys. Res. Commun. 353, 973–977 (2007). 50. Karle, I. L. & Balaram, P. Structural characteristics of .alpha.-helical peptide molecules containing Aib residues. Biochemistry 29, 6747–6756 (1990). 51. Lonsdale, R. & Ward, R. A. Structure-based design of targeted covalent inhibitors. Chem. Soc. Rev. 47, 3816–3830 (2018). 52. Angerani, S. & Winssinger, N. Visible light photoredox catalysis using ruthenium complexes in chemical biology. Chem. Eur. J. 25, 6661–6672 (2019). 53. Yang, B. et al. Genetically introducing biochemically reactive amino acids dehydroalanine and dehydrobutyrine in proteins. J. Am. Chem. Soc. 141, 7698–7703 (2019). 54. Renaud, P., André-Joyaux, E., Kuzovlev, A. & Tappin, N. D. A general approach to deboronative radical chain reaction with pinacol alkylboronic esters. Angew. Chem. Int. Ed. 59, 13859 (2020). 55. Li, Q. et al. Developing covalent protein drugs via proximity-enabled reactive therapeutics. Cell 182, 85–97 (2020). Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. © The Author(s), under exclusive licence to Springer Nature Limited 2020

Nature | Vol 585 | 24 September 2020 | 537

Article Methods General experiment protocol for BACED and pySOOF reagents For all protein-modification reactions, all aqueous buffers, solvents and protein stocks were degassed for at least 8 h in a glovebox (85%) protein recovery was observed after purification using PD SpinTrap G-25 (GE Healthcare) desalting columns and tracking overall protein absorbance. We note that most of the reaction optimization and BACED/pySOOF reagent screening reactions were performed on the model protein substrate Xenopus laevis histone H3-Dha9 in denaturing buffer (500 mM NH4OAc, 3 M GdnHCl, pH 6.0) at a final concentration of 1 mg ml−1 (66 μM) in volumes of 50–200 μl. Short reaction times (20 min) the temperature could be controlled by submerging the reaction vials in a glass beaker filled with water at the desired temperature. Reporting summary Further information on research design is available in the Nature Research Reporting Summary linked to this paper.

Data availability Key raw MS data and primary numerical data for graphical plots have been deposited in the open-access depository ORA-data (https:// doi.org/10.5287/bodleian:9ewjQ268q) and all raw data are available from the corresponding authors upon request. MS/MS raw data files have been uploaded to the PRIDE repository (accession number PXD019565, https://www.ebi.ac.uk/pride/archive). The following additional databases were used: MaxQuant contaminants database (https://www.maxquant.org) and Uniprot Human Database (https://

www.uniprot.org/proteomes/UP000005640). Source data are provided with this paper.

Acknowledgements This research has received funding from the EPSRC (EP/V011359/1), UK Catalysis Hub (EPSRC Portfolio Grant EP/K014668/1; B.G.D., C.F.), the Swiss National Science Foundation (P2BSP2_178609; P.G.I.), BBSRC (BB/P026311/1; B.G.D., V.G., P.G.I.), Oxford-GSK-Crick Chemical Biology Centre for Doctoral Training Programme (EPSRC, GSK to G.R.) via the EPSRC Systems Approaches to Biomedical Science DTC (EP/R512333/1), Oxford Clarendon Scholarship (to B.J.), Rutherford Foundation (to T.H.W.), UCB (to B.J.B.), Brunei Government Scholarship (to A.W.J.P.) and EU H2020 under Grant Agreement 721902 (to O.A.). We thank S. Hester for experimental support in mass spectrometry, T. Mollner and M. Imiołek for providing small-molecule substrates, S. Faulkner for providing protein, the Chemistry Department workshop for construction of the photoreactor and C. am Ende, W. Stockdale and M. Moomersteeg for discussions. Author contributions C.F., B.J., P.G.I., T.H.W., V.G. and B.G.D. conceived and designed the experiments. T.H.W. designed and performed initial experiments exploring the oxidative initiation pathway. C.F. performed initial experiments exploring reductive pathways. C.F. optimized the initial photochemical boronate reaction. C.F. designed the high-flux visible-light photoreactor. P.G.I. designed and performed all experiments for the use of pySOOF reagents. C.F., B.J., B.J.B., O.A., P.G.I., A.M.G. and A.W.J.P. synthesized and characterized BACED substrates and catalysts; B.J., P.G.I. and B.J.B. expressed and generated protein starting materials. B.J., C.F., P.G.I., A.M.G. and A.W.J.P. explored the scope of BACED substrate side chains and proteins; B.J. optimized additions of BACED reagents to proteins and explored additional protein scope. P.G.I. designed and performed all experiments exploring the pySOOF reagents with iron/Ru(bpy)3 and so optimized the corresponding photochemical reaction. P.G.I. designed, synthesized and characterized pySOOF reagents and explored the scope of substrate side chains and proteins. A.W.J.P. and A.M.G. synthesized additional pySOOF and tested them on proteins. J.B.I.S. synthesized additional pySOOF reagents. B.J. developed the on-protein substitution of side-chain alkyl halides with small molecules. B.J. compared the reductive and oxidative initiation of model substrates, explored side reactions and effects of catalysts, as well as methods for recycling side-chain substrate materials from protein reactions. B.J. and A.M.G. conducted ultraviolet–visible high-pressure LC analysis of reaction products. L.C. and C.B.-M. made the electrochemical measurements on the basis of which they suggested a mechanistic interpretation along with R.G.C.; P.G.I. designed and conducted all of the on-protein radical reactions and applications. B.J. developed and conducted the on-protein MS enzymatic deacylation assays. B.J. and P.G.I. designed and conducted Sirt2 deacetylation 19F NMR (and other) enzyme-tracking reactions; B.J., P.G.I., A.J.B. and B.G.D. analysed the corresponding data. B.J. and P.G.I. conducted and characterized 19F NMR tracking of histone octamer reconstitution. G.R. and B.J. developed and conducted the zinc-ejection assays. B.J. and C.F. designed and performed protein crosslinking experiments; S.N. and S.M. performed LC–MS/MS experiments and crosslinked product analyses. B.J. developed and conducted lysate crosslinking immunoprecipitation experiments and analysis, and S.N. provided HeLa nuclear extracts. C.F., B.J., S.M., S.N., P.G.I., T.H.W., B.J.B., A.W.J.P., L.C., C.B.-M., G.R., A.K., A.J.B. and B.G.D. collected and/or analysed data. C.F., B.J., P.G.I. and B.G.D. wrote the paper. All authors read and commented on the paper. Competing interests A patent is being filed that might afford authors royalties were it to be licensed. Additional information Supplementary information is available for this paper at https://doi.org/10.1038/s41586-0202733-7. Correspondence and requests for materials should be addressed to V.G. or B.G.D. Reprints and permissions information is available at http://www.nature.com/reprints.

Extended Data Fig. 1 | Overview of radical side-chain installation and relevant previous literature. a, Retrosynthetic analysis highlights the chemoselective advantages of using carbon radical reagents paired with the radical acceptor Dha (right) over the typical heterolytic 2e− reagents (left) for the site-specific modification of proteins. b, c, Our previous work on radical

addition to Dha14,17 (b) and the mechanism highlighting unwanted side reactions (red) (c). d, e, Summaries of previous works24,25 using photocatalysts for ‘on-peptide’ radical generation for site-selective peptide modification, highlighting their limitations and potential for undesired side reactions (red).

Article

Extended Data Fig. 2 | Complementary strategies for mild protein-compatible photoredox reactions. a, Oxidative-half potential (Eox) spectrum showing catalyst compatibility with protein-based chemistry for relevant catalysts found in the literature23,24 (bottom and top) and tested in this work (top). b, c, In situ formation of BACED reagents (for side chains 1, yellow highlight) advantageously allows Ruii-catalysed, low-Eox activation (compared to other derivatives) to RCH2• radicals, which then react with Dha in proteins to install side chains. Independent and mixed voltammetric responses of 1 mM catechol and 12 mM phenethylboronic acid on glassy carbon (GC) in PBS, pH 7.10 (inset). See also Extended Data Fig. 4 and Supplementary Discussion 2, 3 for more detailed electrochemical experiments. Intact protein LC–MS (bottom right; chromatogram and mass-to-charge ratio, m/z) shows homohomophenylalanine (1h) installation into histone H3 protein.

d, Ruii-catalysed activation of pySOOF reagents to RCF2• radicals, which then react with Dha in proteins to install ‘zero-size’-labelled side chains. The added Feii drives an unprecedented efficiency (2–5 equiv. of precursor) by suppressing oxidation by Ruii* to imine (and hydrate), suggesting the key role of Feii as a reductant (readily available in biology) that quenches the α-carbon radical adduct generated during the reaction. Intact protein LC–MS shows that difluoroethylglycine (DfeGly, 2a) installation into histone H3 protein is successful with Feii (full conversion; top right, chromatogram and m/z) but not without iron (poor conversion to unwanted side products; bottom centre); see also Extended Data Figs. 5–7 for further details. For the full reaction scope of all side chains (types 1 and 2) edited into proteins, including those allowing previously inaccessible on-protein reactivity, see Extended Data Fig. 8 and Figs. 2, 3.

Extended Data Fig. 3 | See next page for caption.

Article Extended Data Fig. 3 | Investigation and optimization of BACED chemistry. a, A 100% stacked bar chart (n = 1, with single data values represented by the y-axis span of the corresponding bars) showing the results of initial studies of the oxidation of benzyltrifluoroborate with different catalyst strengths and additives to achieve selective single addition. Catechol increased the reactivity for all catalysts, leading to the emergence of reactivity with the weakest Cat1 and almost complete conversion to the double-addition product with Cat3, whereas NaCNBH3 with Cat3 successfully quenched the α-carbon radical promoting single addition. b, Trends of increasing oxidative damage and decreasing reaction control with higher-*Eox catalysts with representative LC–MS ion series and spectra (see Supplementary Methods for details). c, A 100% stacked bar chart (n = 1, with single data values represented by the y-axis span of the corresponding bars) comparing Cat1 and Cat2 reaction conversions to the single-addition product and unwanted α-carbon radical quenching (via either double addition or catechol quenching) for different boronate substrates, arranged with higher-Eox primary boronate substrates on

the left and increasingly stabilized radical precursors on the right. The trends suggest the utility or necessity of using the stronger Cat2 (*Eox = +0.99 V) for the primary boronate substrates and the increased efficiency when using the weaker Cat1 (*Eox = +0.77 V) for the more stabilized radical precursors (secondary and benzyl substrates), probably owing to a slow, controlled release of stable radicals, ensuring efficient addition to Dha instead of selfquenching or overalkylation. d, A light-intensity screen shows that increasing light intensity (450 nm blue LED, 0–50 W) allows high conversion efficiencies to the desired single-addition product with shorter reaction times and lower concentrations of Cat1 (Supplementary Table 5; n = 1; best-fit line overlayed). e, Reaction scheme and LC–MS spectra for the installation of iodonorleucine (Inl). Although Inl installation had poor initial conversion (~50%), the only other species present after the reaction was the starting material (Dha), allowing successive reactions to increase conversion efficiencies to ~75%. f, A screen of catechol derivatives finds that the naturally occurring catecholamines dopamine and l-DOPA could efficiently substitute for the role of catechol.

Extended Data Fig. 4 | See next page for caption.

Article Extended Data Fig. 4 | Mechanistic investigation of the role of catechol in BACED reactions. a, Reaction scheme screening different quinone derivatives for their influence on the oxidation of potassium phenethyltrifluoroborate and subsequent addition to Dha. b, Potential mechanism of catechol derivatives acting as redox mediators to bridge electron transfer between catalyst and substrate. c, LC–MS results of the quinone derivative screening (shown in a and b) ruled out the mechanism in b because only 1,2-diols showed substantial activity, with only catechol avoiding protein degradation. d, A potential mechanism of in situ catalyst modification with catechol, creating catecholoRu(bpy)2 (Cat6)30 was ruled out because it did not promote alkylation with or without the addition of exogenous catechol. e, In situ formation of a reactive boronic acid catechol ester is suggested. f, 1-Propylboronic acid catechol ester and 4-bromobutylboronic acid catechol ester were successfully added without the addition of exogenous catechol. This suggests that the formation of the catechol ester lowers the Eox value of the substrate to a range accessible by Cat1 (*Eox = +0.77 V). g, Voltammetric response of 1 mM catechol in the presence of increasing concentrations of 4-bromobutylboronic acid (black, 0 mM; brown, 3 mM; green, 6 mM; blue, 12 mM; red, 24 mM) at a glassy carbon macroelectrode

in 50 mM phosphate buffer (pH 6) recorded at 100 mV s−1. h, Voltammetric response of catechol only (1 mM; black), 4-bromobutylboronic acid (12 mM; blue), and 4-bromobutylboronic acid in the presence of catechol (12 mM and 1 mM, respectively; red) recorded at 100 mV s−1. i, Simulated voltammetric response for the oxidation of 4-bromobutylboronic acid (12 mM) in the presence of catechol (1 mM) at 100 mV s−1, following the simplified mechanism outlined in e. The simulation highlights the importance of the oxidation of the boronic acid ester being catalytic and leading to the reformation of catechol (see Supplementary Discussion 2, 3). The rate of decomposition of the radical ester has been set at either 1 × 104 s−1 or 0 s−1 (blue and red, respectively). If the oxidation of the boronic acid ester is not catalytic, no peak is predicted to be voltammetrically observable at >1 V versus a saturated calomel reference electrode (SCE). j, Voltammetric response of 4-bromobutylboronic acid (12 mM) in the presence of catechol (1 mM) in 50 mM phosphate buffer (pH 6) as a function of scan rate (25−400 mV s−1). k, l, Voltammetric response of preformed 4-bromobutylboronic acid catechol ester (1 mM) (k) and catechol (1 mM) in 50 mM phosphate buffer (pH 6) (l) as a function of scan rate (25−400 mV s−1).

Extended Data Fig. 5 | See next page for caption.

Article Extended Data Fig. 5 | Initial experiments without iron using various hydride sources, and optimization study with sodium borohydride for pySOOF. a, The formation of oxidative-derived side products H3-imineDfeGly9 and H3-hemiaminal-DfeGly9 was detected in an initial, additive-free, light-driven protein-modification reaction with pySOOF and Ru(bpy)3Cl2. On the basis of this observation a mechanism was postulated, in which the on-protein radical intermediate is oxidized by the excited state of the photocatalyst. To avoid this reaction pathway, the use of hydride donors such as silanes, tertiary amines or borohydrides were assumed to favour the formation of DfeGly-modified histone at site 9. b, Inferior results were observed when 250 equiv. DIPEA, TTMS and catechol (100 equiv.) were used (with 200 equiv. pySOOF and 5 equiv. Ru(bpy)3) (Supplementary Tables 8, 11, 12). For catechol, excellent conversion efficiencies were observed; however, catechol-DfeGly-modified protein was detected as a major side product. Borohydrides such as Na(OMe)3BH (0.25 mg) and NEt4BH4 (0.25 mg) showed promising reactivity, but only the oxidation-derived side products were

formed. Only with sodium borohydride was the desired DfeGly-modified histone formed in moderate conversion (Supplementary Table 8). c, Reduction of Dha to Ala by NaBH4 was identified as a potential limitation for the NaBH4mediated reaction. To test this undesired pathway, a crude reaction mixture was buffer-exchanged to NaPi (100 mM, pH 9, 3 M GdnHCl) and incubated with a large excess of 2-mercaptoethanol. After 1 h at 37 °C, no formation of the corresponding thiol Michael-type protein adduct was detected, proving Ala formation. The 100% stacked bar graphs (n = 1, with single data values represented by the y-axis span of the corresponding bars) summarize the results of the optimization studies of the NaBH4-mediated photochemical reaction with different concentrations of NaBH4, photocatalyst and pySOOF and reaction times (Supplementary Tables 9, 10). Less than 500 equiv. NaBH4 resulted in the formation of undesired oxidative side products. Increasing the reagent concentration, photocatalyst loading or reaction times did not improve the reaction. d, Lower temperature increased the conversion, most probably by slowing down the reduction of Dha by NaBH4.

Extended Data Fig. 6 | See next page for caption.

Article Extended Data Fig. 6 | Optimization study of Fe(ii)-mediated protein modification reaction with pySOOF. a, In the photochemical modification reaction with pySOOF, Fe(ii) likely acts as a reductive quencher for the photoredox cycle and as a single-electron reductant of the on-protein radical intermediate forming the enolate intermediate. However, side products, such as H3-imine-DfeGly9, H3-hemiaminal-DfeGly9 or H3-diDfeGly9, can be generated because of inefficient quenching of the on-protein radical intermediate. b, In an initial experiment with 200 equiv. pySOOF, 280 equiv. FeSO 4, 5 equiv. Ru(bpy)3Cl2 and 66 μM histone H3-Dha9 with a protein concentration of 0.5 mg ml−1, 70% conversion to a mixture of H3-DfeGly9 and H3-diDfeGly (57:43) was observed. The formation of the mono-addition product was favoured at higher protein concentration (1 mg ml−1) and the conversion was increased to 92% (Supplementary Table 13). c, For the Fe(ii)mediated reaction, various metallo- and organophotocatalysts with different

*Ered values (−0.56 to −1.37 V) and radical precursors were tested, and Ru(bpy)3 and pySOOF were identified as the best combination (Supplementary Table 17). d, The 100% stacked bar charts (n = 1, single data values represented by the yaxis span of the corresponding bars) summarize the results of the optimization studies of the FeSO4-mediated photochemical reaction with different reaction times, concentrations of pySOOF and FeSO4, and catalytic amounts of Fe(ii) and photocatalyst, respectively (Supplementary Tables 10–15, 18, 20). Short reaction times and high efficiency with low concentrations of pySOOF were found. However, in cases with high levels of reactivity with  •CF2Me was identified, and no product formation was observed for •CFHCONH2. c, On the basis of the reactivity study of various mono- and difluoro-pySOOF reagents, a suitable radical precursor was designed to allow the efficient generation of •CF2R for the installation of difluorinated- amino acid residues or PTMs. d, With iodo-pySOOF, homolytic bond cleavage between iodine and the carbon centre was induced, and the

pySOOF unit was installed on the Dha-tagged histone (Supplementary Tables 22–23). After further photoredox activation, a captodative-effectstabilized on-protein radical was formed, which then allowed further onprotein chemical reactions by trapping the radical with various acceptors (Supplementary Tables 24–37). Moreover, the homolytic iodo-carbon bond cleavage inspired the design of bromo-difluoro carbonyl-based radical precursors for the installation of CF2Gln- or CF2Glu-derived amino acids. e, Mechanistic studies supported a radical mechanism, as no product formation was observed with 4-hydroxy-TEMPO (TEMPOL). Furthermore, a photocatalytic process was confirmed, as no conversions were detected without a photocatalyst or blue LED irradiation (Supplementary Table 16). On– off experiments showed only product formation during the irradiation period, excluding a radical chain mechanism. Finally, good reactivity (80% conversion) was observed when the sample was prepared under ambient atmosphere; however, levels of oxidative damage (Met oxidation) and oxidative-derived DfeGly-modified histones were also generated.

Extended Data Fig. 8 | Substrate scopes for BACED and pySOOF. a, A comprehensive list of all side chains installed into proteins with the BACED reaction manifold. General conditions: protein 1 mg ml−1; 50 W, 450 nm light; 4 °C to RT; 100–1,500 equiv. BACED precursor reagent; 10 equiv. Cat1 or Cat2; 100 equiv. catechol; 3G5)()()/&"H(5H$## s; IFJKLMNOONFLPQRPRLRSNTJULVQLUQRMWXVQULRNJQJKLVKLFYOQZLUQRMWXVQLONWQLMNO[JQ\LPQMSFX]TQRLXFLPSQL^QPSNURLRQMPXNF_ s >#4$!($"8&7746&$&( ((# s >#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$4$"(+ s >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'"(# s 1>44 $"4#='"$m'$#"($8$=5.7$"98!'.7$47/&6&$7&.7#&(&( H>7$(88$%' ()&()&6& 4$&(#&5#&(& H>#4$!($"8&"/($4($" "#&(&&6&$7&.$7$(/ z#"($8$8!'.7$4&77/&6&$7&.7>&.$#!$7$" &%$6"$"()C&($&7&"#C()# 4($": 0&5#&(&&"#&#(&$7##4$!($"8()&"&7/$!"(#$"7710C& &/$"82&($"&"#&5#&(&)&6."#!$(#(()A&"&2&19/7$" 6&"#4&".&44 #6$&,)((!,DD!&"&2&5.:%DxE&tA:'7: u$%$"&7.7(&&6&$7&.7$"1'!!72"(&/772&"'4$!(2'($"47'#&#&(&&6&$7&.$7$(/(&(2"(:;)$ (&(2"()'7#!6$#()8775$"%$"82&($"=5)&!!7$4&.7,

0

77&5#&(&8%&!) &&6&$7&.7$"1'4G&(&: G&(&.& '#,555:%"6($%&(:42;>z0or#&(&.&*555:&&.$#!$:%+

7$"%87&8#$45& !82#./&"#27/74($"%7&6 "!7&"(%5"$"()B&4(&24"#$($"=5$()(&/5$()!7&"(8 0&"#2$E&($" #1&$82! 8"(%"(/! .$"%&"#27/!$($"#"() &2 )78:"($.#$ @'9&/($44777$" A&7&"(7%/ >"$2&7&"#()%&"$2 l'2&"&4)!&($4$!&"( 7$"$4&7#&(&

"D& z"676#$"() ('#/ s s s

)zAHm "($.#$ '# x&7$#&($"

@'9&/($44777$"

224$&7&"($.#$ '#,…Hk"($.#/=CC€3%$(+



„

Article

Homeostatic mini-intestines through scaffold-guided organoid morphogenesis https://doi.org/10.1038/s41586-020-2724-8 Received: 25 June 2018 Accepted: 24 June 2020 Published online: 16 September 2020 Check for updates

Mikhail Nikolaev1, Olga Mitrofanova1,7, Nicolas Broguiere1,7, Sara Geraldo1, Devanjali Dutta1, Yoji Tabata1, Bilge Elci1, Nathalie Brandenberg1,5, Irina Kolotuev2, Nikolce Gjorevski1,6, Hans Clevers3 & Matthias P. Lutolf1,4 ✉

Epithelial organoids, such as those derived from stem cells of the intestine, have great potential for modelling tissue and disease biology1–4. However, the approaches that are used at present to derive these organoids in three-dimensional matrices5,6 result in stochastically developing tissues with a closed, cystic architecture that restricts lifespan and size, limits experimental manipulation and prohibits homeostasis. Here, by using tissue engineering and the intrinsic self-organization properties of cells, we induce intestinal stem cells to form tube-shaped epithelia with an accessible lumen and a similar spatial arrangement of crypt- and villus-like domains to that in vivo. When connected to an external pumping system, the mini-gut tubes are perfusable; this allows the continuous removal of dead cells to prolong tissue lifespan by several weeks, and also enables the tubes to be colonized with microorganisms for modelling host–microorganism interactions. The mini-intestines include rare, specialized cell types that are seldom found in conventional organoids. They retain key physiological hallmarks of the intestine and have a notable capacity to regenerate. Our concept for extrinsically guiding the self-organization of stem cells into functional organoids-on-a-chip is broadly applicable and will enable the attainment of more physiologically relevant organoid shapes, sizes and functions.

We postulated that the morphogenetic processes that shape cystic intestinal organoids into their characteristic crypt and villus structures could be harnessed to promote in vitro stem cell patterning along predefined spatial boundaries, particularly those that approximate the three-dimensional (3D) topology of the surface of the gut. To this end, and inspired by previous work on micro-engineered intestinal surfaces7,8, we generated a scaffold that would be permeable to gases, nutrients and macromolecules, that would facilitate the efficient adhesion, proliferation and differentiation of intestinal stem cells (ISCs) and that would be stiff enough to serve as a physical barrier restricting the growth of ISCs to predefined shapes. Whereas pure Matrigel was too soft to confine the growth of mouse ISCs (LGR5-eGFP-IRES-creERT2 mouse model; hereafter, LGR5–eGFP+ ISCs), we found that a hybrid matrix composed of a mixture of type-I collagen, which provides a relatively stiff, adhesive substrate, and Matrigel, which contains the key constituents of the native basement membrane, met the necessary requirements. We integrated these hydrogels in a perfusable platform to generate a hybrid microchip system that consists of an elastomeric device with a central chamber for hydrogel loading and subsequent organoid culture, flanked by a pair of (inlet and outlet) reservoirs for cell loading and luminal perfusion, as well as lateral reservoirs that supply medium and growth factors to the basal side of the tissues through the hydrogel

(Fig. 1a, b, Extended Data Fig. 1a). The microchannel, which contains microcavities that mimic the geometry of the native crypts in the mouse small intestine, was laser-ablated within the central gel scaffold (Fig. 1b, Extended Data Fig. 1b, c). These tubular hydrogel scaffolds could be readily colonized with mouse LGR5–eGFP+ ISCs by perfusion from the inlet reservoir. Time-lapse microscopy showed the rapid establishment of a confluent cell sheet that was several times larger than organoids grown in 3D Matrigel (Fig. 1c, Supplementary Video 1). Confocal microscopy revealed a tightly packed, simple epithelium expressing high levels of E-cadherin at the junctions between cells (Extended Data Fig. 1d, Supplementary Video 2). These tissues remained open and free of cells at both ends, enabling the delivery of fluid to the apical side of the epithelium (Fig. 1d) and the removal of non-adherent or dead cells from the lumen (Fig. 1e). Notably, colonization of the tubular scaffolds by primary mouse cholangiocytes (Extended Data Fig. 2a), or by primary human stem and progenitor cells from the small intestine (Extended Data Fig. 2c–e) or trachea (Extended Data Fig. 2f, g), generated coherent, tightly packed and perfusable epithelial tissues. Trachea tubes could be readily cultured at the air–liquid interface (Extended Data Fig. 2f). Collectively, these data show that a scaffold that mimics the basement membrane can be used to reliably build—from primary stem and progenitor cells—openly accessible epithelia with an anatomy similar to that in vivo.

Laboratory of Stem Cell Bioengineering, Institute of Bioengineering, School of Life Sciences (SV), École Polytechnique Fédérale de Lausanne (EPFL), Lausanne, Switzerland. 2Electron Microscopy Facility, Faculty of Biology and Medicine, University of Lausanne, Lausanne, Switzerland. 3Oncode Institute, Hubrecht Institute, Royal Netherlands Academy of Arts and Sciences and University Medical Center, Utrecht, The Netherlands. 4Institute of Chemical Sciences and Engineering, School of Basic Sciences (SB), EPFL, Lausanne, Switzerland. 5Present address: Startlab/SUN bioscience, Epalignes, Switzerland. 6Present address: Roche Pharma Research and Early Development, Basel, Switzerland. 7These authors contributed equally: Olga Mitrofanova, Nicolas Broguiere. ✉e-mail: [email protected] 1

574 | Nature | Vol 585 | 24 September 2020

a

c

External tubing

Organoids

Mini-guts LGR5–eGFP

Matrix-loading port

Day 0

Medium reservoirs

Inlet reservoir

Day 1

Outlet reservoir External g tubing 1,500 μm

400 μm

Side view

1,850 μm Day 3

b

Side view 50 170 150

Day 5

75

1,200

d

Bright-field

40-kDa FITC–dextran

e

9 h without perfusion

After perfusion pulse

Fig. 1 | Establishment of long-term homeostatic culture of tubular mini-guts. a, Schematic of 3D hydrogel-containing microdevice developed for the mini-gut culture. The system consists of a hydrogel chamber in the centre flanked by two external medium reservoirs and two inlet and outlet reservoirs for perfusion through the lumen. b, Dimensions (in μm) of an open microchannel with an in vivo-like anatomical structure, generated by laser ablation. The channel spans the entire length of the central hydrogel compartment. c, Bright-field (left and middle) and LGR5–eGFP fluorescence (right) time-course experiments of epithelium formation in tissue-engineered

mini-guts (middle and right) compared to traditional organoids formed in Matrigel (left). Extended depth of field of bright-field images, calculated for a z-stack of 80 μm; fluorescence confocal images correspond to a maximum intensity projection of a z-stack of around 60 μm. d, Bright-field and fluorescence confocal images of a five-day-old mini-gut tube perfused with fluorescein isothiocyanate (FITC)-tagged dextran (40 kDa), showing the maintenance of epithelium integrity. e, A 10-day-old mini-gut tube with accumulated dead cells that were shed into the lumen over the course of 9 h without perfusion (left), and just after a perfusion pulse (right). Scale bars, 50 μm.

Establishment of tissue homeostasis

Fig. 3d, e, Supplementary Video 4). Thus, our approach establishes— without the need for passaging—a long-living homeostatic organoid culture system in which cell birth and death are balanced.

Conventional epithelial organoids can be propagated nearly indefinitely, but long-term culture in these systems involves continuous passaging that requires breaking the organoids into fragments or dispersed single cells every few days5. We explored the possibility of maintaining tubular intestinal epithelia for several weeks without passaging; that is, by continuously removing dead cells from the rapidly growing epithelia through perfusion. Indeed, high levels of cell shedding and accumulation of dead cells in the closed cavity of conventional organoids led to tissue destruction after about 10 days (Extended Data Fig. 3a). In the absence of perfusion, the epithelial tubes became densely packed with dead cells after 6–10 hours (Fig. 1e, Supplementary Video 3), leading to tissue destruction a few days later (Extended Data Fig. 3c). Notably, when the lumen is perfused with standard organoid culture medium5—or even growth-factor-free medium—every 12 hours, to remove dead cells from the organoid tubes, the lifespan of the tissue could be extended to one month or longer, preserving the overall tissue anatomy and localized niches of LGR5–eGFP+ ISCs (Extended Data

Stereotypical cell-fate patterning We next tested how the induction of differentiation would affect cell fate in ISC-derived epithelial tubes. Fluorescence imaging and immunostaining revealed a biomimetic spatial distribution of cell types (Fig. 2a–f, Supplementary Video 5), similar to the cell-fate patterns along the crypt–villus axis in vivo. We identified crypt-like regions that exclusively contained cells that stained positive for SOX9 (Fig. 2a) and the Paneth cell marker lysozyme (Fig. 2b). Labelling the dividing cells in the tissue with 5-ethynyl-2′-deoxyuridine (EdU) revealed distinct areas of cell proliferation that were to a large extent restricted to the crypt regions (Fig. 2c), whereas offspring cells migrated and replenished the short-lived differentiated cells in the intestinal lumen (Extended Data Fig. 4a)—mirroring the pattern seen in the native small intestine. By contrast, cells that stained positive for markers of enterocytes (Fig. 2d), Nature | Vol 585 | 24 September 2020 | 575

Article a

SOX9 F-actin

d

L-FABP DAPI E-cadherin

b

Lysozyme DAPI F-actin

e

ChgA DAPI E-cadherin

c

EdU DAPI E-cadherin

f

Muc2 DAPI F-actin

h

2

ls

20 10

M

-li

ke

ce

lls

ls el

el

tc Tu f

tc le

th

ob G

ne Pa

e rin oc En

te ro e

nd

ce

ce

lls

lls

es yt ro c

te En

ng di Di

vi

em

ce

ce

lls

lls

0

30

s

0

4

40

id

20

6

50

ut s

40

l

k

Atlas

Villus-top enterocyte markers (sum of log(expression))

Cell-type proportion (%)

Cell-type proportion (%)

60

St

Organoids

8

no

Mini-guts (20 d)

i-g

Mini-guts (10 d) 80

in

j

Organoids

Villus-top enterocytes Paneth cells M-like cells

Enterocytes Goblet cells

rg a

Dividing cells Tuft cells

Stem and progenitor cells Enteroendocrine cells

i

Mini-guts (20 d)

O

Mini-guts (10 d)

M

g

log(expression) 2.5 Wnt3 2.0 M cells 3.5 Villus-top enterocytes 1.5 Immature enterocytes

enteroendocrine cells (Fig. 2e) and goblet cells (Fig. 2f) were found almost exclusively in the central regions of the tubes, which correspond to villus-like areas. For the organoids to be representative of the native intestinal mucosa, they must sustain its secretory and absorptive function. To test this, we removed epithelia from the microchips for downstream histological sectioning and analyses (Extended Data Fig. 4b). Alcian blue staining of acidic mucopolysaccharides identified goblet cells and a thin mucus layer covering the apical side of the epithelium (Extended Data Fig. 4c). Transmission electron microscopy (TEM) analysis confirmed the presence of cells with mucus-containing secretory vesicles, together with a single layer of densely packed enterocytes with their characteristic microvilli forming a brush border on the apical surface (Extended Data Fig. 4d). An analysis of the function of brush border aminopeptidases— the enzymes that are responsible for the final step in the digestion of dietary carbohydrates and proteins—showed that enzymatic activity increased steadily after the induction of differentiation, until a plateau was reached four days later that was sustained in longer-lived tissues (Extended Data Fig. 4e). Together, these data show that a spatially confining hydrogel scaffold promotes the ‘guided’ self-organization of ISCs into a functional intestinal epithelium that exhibits a spatial arrangement of crypt- and villus-like domains similar to that in vivo. 576 | Nature | Vol 585 | 24 September 2020

Fig. 2 | Cell-fate patterning and cellular diversity of tubular mini-guts. a–f, Fluorescence confocal images of representative 7-day-old mini-gut tubes, showing an entire tissue (left) and a higher-magnification view (right), containing: SOX9+ stem and progenitor cells (red, a); lysozyme+ Paneth cells (red, b); proliferating cells following an EdU pulse of 12 h (white, c); L-FABP+ enterocytes (red, d); chromogranin A (ChgA)+ enteroendocrine cells (red, e); and mucin 2 (Muc2)+ goblet cells (red, f). Nuclei are stained with DAPI (blue) and cellular actin filaments are stained with E-cadherin or phalloidin (green). Images correspond to the maximum intensity projection of a z-stack of around 60 μm. Scale bars, 50 μm. Data in a–f are representative of at least two independent experiments. g–i, Unsupervised clustering of the key intestinal epithelial cell types in 10-day-old mini-guts (g), 20-day-old mini-guts (h) and classical 3D organoids (i). j, Cell-type proportions in conventional organoids and mini-guts, compared to in vivo data (‘atlas’)9. Error bars are 95% confidence intervals estimated from theoretical sampling error. k, Expression of villus-top enterocyte marker genes in mini-guts and organoids (sum of a published signature of 42 genes)13. Each point represents a cell. l, Overlay of averaged values of marker genes that define M cells (Zmat3, Mmp15, Myadm, Anxa5 and Marcksl1), immature enterocytes (Fgfbp1, Dmbt1, Pdss1 and Prss32) and villus-top enterocytes (Ada, Ifrd1, Krt20, Pmp22 and Serpinb1a). log(expression) refers to the natural logarithm of count values normalized to 10,000 per cell.

Emergence of rare cell types To shed light on the cellular diversity and the proportions of different cell types that are found in mini-gut tubes, we performed a single-cell RNA sequencing (scRNA-seq) analysis of cells isolated from young (10 days) and older (20 days) mini-guts, as well as pooled Matrigel-derived organoids (Fig. 2g–l). On the basis of previously described cell-type markers in the endogenous intestinal epithelium9, we defined nine main transcriptionally distinct clusters that correspond to the key cell types found in vivo (Fig. 2g–j, Extended Data Fig. 5). Consistent with the maintenance of a crypt-like domain (Fig. 2a, c, Extended Data Fig. 3a, b), we found that mini-guts retained a relatively large proportion of stem and progenitor cells (around 50% after 20 days) compared to conventional organoids (around 15%). Notably, from day 10 to day 20, a progressive shift in the fraction of stem and progenitor cells to enterocytes was apparent (Fig. 2g, h), suggesting that tissue maturation was occurring in the homeostatic mini-gut tubes. The proportions of dividing cells (around 5–10%), Paneth cells (around 1–2%) and Tuft cells (around 0.5–1%) were similar in organoids and mini-gut tubes, and approximated the proportions in endogenous tissues9 (Fig. 2j, Extended Data Fig. 6a–i). Compared to the in vivo condition, the proportion

6 h, infected epithelium DAPI E-cad Crypt-a-Glo

c

Before treatment

2d

3d

6d

7d

10 d

Before treatment

40-kDa FITC–dextran

After 24 h

25 h 29 h

d

32 h

20 h

0h

b

10 h

a

120 h, new oocysts DAPI E-cad Crypt-a-Glo

e

Type-I/type-II meront

24 h, type-I meronts

24 h, type-II meronts

96 h, microgamonts

Sporo-Glo

Sporo-Glo

Sporo-Glo

Early zygote

Shed cell

Macrogamont

Fig. 3 | Perspectives for modelling intestine biology and disease. a, Epithelial wound healing in mini-gut tubes damaged by targeted laser ablation. b, DSS-induced loss of intestinal barrier integrity, as shown by FITC– dextran permeability. c, Time-course experiments of DSS-induced epithelial damage and regeneration. Arrowheads indicate regeneration of the lesion areas (a, c). Scale bars, 50 μm (a–c). Data in a–c are representative of at least three independent experiments. d, Immunofluorescence of C. parvum undergoing its major epicellular stages in the mini-guts. After about 6 h of infection, floating half-empty oocysts were observed; at 24–72 h after infection type-I and type-II meronts were detected; at 72–96 h microgamonts containing

12–16 microgametes were detected; and at 120–144 hours new oocysts were again observed, implying that the parasite was able to complete its full life cycle within the mini-gut lumen. Nuclei of intestinal epithelial cells are stained with DAPI (blue) and cellular actin filaments E-cadherin (green). The different stages of the C. parvum life cycle are stained with Crypt-a-Glo (oocyst outer walls; red) and Sporo-Glo (sporozoites, merozoites and all other intracellular reproductive stages; green). Scale bars, 5 μm. e, Scanning electron microscopy image of distinct stages of the C. parvum life cycle at 72 h after infection. Scale bar, 25 μm. Data in d, e are representative of two independent experiments and at least 10 different mini-gut tube regions were analysed.

of Goblet cells was much lower in both in vitro conditions (around 7% in vivo; 1% in vitro) (Fig. 2j). A few rare and functionally important specialized cell types are difficult to reproduce in intestinal organoid cultures9–11, as they generally require specific manipulations that may disrupt stereotypical organoid patterning along a crypt–villus axis. These cell types include microfold cells (M cells), which are crucial players in mucosal immunity12; a subset of enterocytes found specifically near the tip of intestinal villi that also have immune-modulatory functions13; and enteroendocrine cells, which are the main sources of gut-derived hormones. Cells that express villus-top enterocyte markers13 were found in mini-guts but not in traditional organoids (Fig. 2g–l). They shared some traits with a cluster of rare cells (0.5–2%) positive for marker genes that define immature M cells (Marcksl1 and Anxa5), so-called revival stem cells14 (Clu and Msln) and regenerating cells15 (Ly6a) of the intestinal epithelium (Fig. 2g–l, Extended Data Fig. 7). These M-like cells were not detected in control organoids and expressed a unique set of genes characteristic of M-cell function in vivo (Extended Data Fig. 7a). Their presence in the mini-guts was confirmed by immunostaining for the universal M-cell marker GP214 (Extended Data Fig. 7b). Notably, enteroendocrine cells are relatively abundant (around 5%) in homeostatic mini-guts, resembling the cell fraction found in vivo (Fig. 2j) and capturing the hormone expression profile of key enteroendocrine cell types (Extended Data Fig. 7c). By contrast, these cells are exceedingly rare in conventional organoids (around 0.3%), unless organoid differentiation is promoted through treatment with inhibitors of the Wnt, Notch or EGF pathways11. Collectively, these data show that the diversity of cell types in our homeostatic

mini-guts closely resembles that of the intestinal epithelium in vivo, and includes cell types that are rare or absent in conventional organoids.

Regenerative potential of mini-gut tubes We next tested the extent to which tubular mini-guts could regenerate after an injury induced by three models of epithelial damage. First, we used an ultraviolet laser beam to introduce epithelial lesions at defined locations (Fig. 3a, Supplementary Video 6). Because of the stability of the mini-gut tubes, we could readily track tissue repair by live-cell microscopy; this revealed an invasion of cells from the surrounding tissue that resulted in complete regeneration in less than 32 hours. Next, we treated the tissues with dextran sodium sulfate (DSS), a cytotoxic compound that is frequently used to model ulcerative colitis in mice16. In contrast to conventional organoids, which rapidly collapsed in response to treatment with DSS (Supplementary Video 7), the mini-gut tubes showed a notable ability to regenerate (Fig. 3b, c)—probably owing to the more physiological luminal exposure of the tissues to the toxic polysaccharide. Finally, to mimic the intestinal damage and regeneration that occurs in vivo, we exposed the mini-guts to γ-radiation (Extended Data Fig. 8a, Supplementary Video 8). Exposure to a high dose of radiation (8 Gy) resulted in a loss of stem cells and impaired regeneration (Extended Data Fig. 8a, b). At a lower dose (2 Gy), we observed a rapid depletion of proliferating ISCs and a disruption of the intestinal epithelium, followed by a gradual re-epithelialization until the tissue was completely regenerated with newly established crypts (Extended Data Fig. 8b, c). Together, these Nature | Vol 585 | 24 September 2020 | 577

Article experiments reveal that our bioengineered organoids show notable regenerative potential.

Modelling long-term parasite infection Finally, we investigated whether mini-gut tubes could be used to model long-term infection caused by Cryptosporidium parvum, an obligate parasite that results in life-threatening diarrhoea in immunocompromised adult hosts and in infants17. Research on the pathophysiology of C. parvum has been hindered by the lack of long-term, primary-cell-derived in vitro culture systems. Conventional 3D organoids can be used to model C. parvum infection18, but the luminal inaccessibility and inability of the system to support long-term host–microorganism co-cultures limit the applicability of this system. We infected mini-gut tubes with C. parvum by loading a suspension of oocysts into the inlet reservoir of the microchip (Extended Data Fig. 9a). Live-cell microscopy demonstrated that the tubular organoids support the completion of the life cycle and long-term growth of C. parvum without compromising tissue integrity (Extended Data Fig. 9b, Supplementary Video 9). The identity of each asexual and sexual stage was confirmed using immunofluorescence assays (Fig. 3d) and TEM imaging on mini-gut cross-sections (Fig. 3f, Extended Data Fig. 9c, d). By infecting mini-guts with freshly isolated sporozoites and analysing the luminal content every day, we observed successive rounds of production of newly formed oocysts for at least four weeks (Extended Data Fig. 9e). Gene set enrichment analysis of infected samples showed a significant enrichment of interferon-α response genes, as well as changes in metabolism (Extended Data Fig. 9f). An analysis by cell type showed that the interferon response was not limited to one specific population, but was instead distributed across all cell types, even though the response genes showed some cell-type specificities (Extended Data Fig. 9g; see also Source Data). Altogether, these results show that these bioengineered organoids, similar to primary-stem-cell-derived intestinal monolayers comprising feeder cells19, are ideally suited for mechanistic host–microorganism interaction and long-term infection studies. By combining bioengineering with the self-organization properties of stem cells, here we have generated open, tubular ‘organoids-on-a-chip’ that exhibit exceptional cell-type diversity, tissue architecture and function. The introduction of a microchip-based perfusion system made it possible to efficiently remove shed cells from the lumen and expose it to parasites or medium additives. Previous efforts to model intestinal epithelia through micro-engineering or tissue engineering7,8,19–25 have successfully addressed the problem of lumen accessibility, but it has not yet been possible in these systems to capture the diversity of cell types and the patterning that are found in vivo or in classical organoids. Moreover, existing approaches are based on polarized cell monolayers grown on two-dimensional permeable polymer membranes26,27, which may preclude the modelling of complex 3D multi-tissue interactions. The biomimetic 3D extracellular matrix that surrounds the patterned intestinal epithelium in our system can be readily colonized with non-epithelial cell types (Extended Data Fig. 10a) such as endothelial cells (Extended Data Fig. 10b), immune cells (Extended Data Fig. 10c, d) and myofibroblasts (Extended Data Fig. 10e, f). These supportive cell types were found to communicate with intestinal epithelial cells; for example, macrophages were observed to undergo morphological changes and ingest particles excreted from the epithelium (Extended Data Fig. 10c, Supplementary Video 10). The functional integration of an immune axis in bioengineered, homeostatic organoids opens up new perspectives for disease modelling. We anticipate that by adjusting specific characteristics of the hydrogel scaffolds (for example, their composition, geometry, size, stiffness and signalling inputs), the approach we describe here could be applied to other organoid-forming stem cells—including those derived from other organs, such as the lung, liver or pancreas, and derived from patient biopsies (Extended

578 | Nature | Vol 585 | 24 September 2020

Data Fig. 2). The readily accessible 3D tissue anatomy of our model, which captures the development of stem cells in a highly tractable experimental framework, will answer questions that have so far been difficult to address, and may have substantial potential for drug discovery, diagnostics and regenerative medicine.

Online content Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41586-020-2724-8. 1. 2. 3. 4. 5. 6. 7.

8.

9. 10.

11.

12. 13. 14. 15. 16. 17. 18. 19. 20.

21. 22. 23.

24.

25. 26. 27.

Clevers, H. Modeling development and disease with organoids. Cell 165, 1586–1597 (2016). Sato, T. & Clevers, H. Growing self-organizing mini-guts from a single intestinal stem cell: mechanism and applications. Science 340, 1190–1194 (2013). Dekkers, J. F. et al. A functional CFTR assay using primary cystic fibrosis intestinal organoids. Nat. Med. 19, 939–945 (2013). van de Wetering, M. et al. Prospective derivation of a living organoid biobank of colorectal cancer patients. Cell 161, 933–945 (2015). Sato, T. et al. Single Lgr5 stem cells build crypt-villus structures in vitro without a mesenchymal niche. Nature 459, 262–265 (2009). Gjorevski, N. et al. Designer matrices for intestinal stem cell and organoid culture. Nature 539, 560–564 (2016). Wang, Y. et al. Formation of human colonic crypt array by application of chemical gradients across a shaped epithelial monolayer. Cell. Mol. Gastroenterol. Hepatol. 5, 113–130 (2018). Wang, Y. et al. A microengineered collagen scaffold for generating a polarized crypt-villus architecture of human small intestinal epithelium. Biomaterials 128, 44–55 (2017). Haber, A. L. et al. A single-cell survey of the small intestinal epithelium. Nature 551, 333–339 (2017). de Lau, W. et al. Peyer’s patch M cells derived from Lgr5+ stem cells require SpiB and are induced by RankL in cultured “miniguts”. Mol. Cell. Biol. 32, 3639–3647 (2012). Basak, O. et al. Induced quiescence of Lgr5+ stem cells in intestinal organoids enables differentiation of hormone-producing enteroendocrine cells. Cell Stem Cell 20, 177–190 (2017). Hase, K. et al. Uptake through glycoprotein 2 of FimH+ bacteria by M cells initiates mucosal immune response. Nature 462, 226–230 (2009). Moor, A. E. et al. Spatial reconstruction of single enterocytes uncovers broad zonation along the intestinal villus axis. Cell 175, 1156–1167 (2018). Ayyaz, A. et al. Single-cell transcriptomes of the regenerating intestine reveal a revival stem cell. Nature 569, 121–125 (2019). Yui, S. et al. YAP/TAZ-dependent reprogramming of colonic epithelium links ECM remodeling to tissue regeneration. Cell Stem Cell 22, 35–49 (2018). Kiesler, P., Fuss, I. J. & Strober, W. Experimental models of inflammatory bowel diseases. Cell. Mol. Gastroenterol. Hepatol. 1, 154–170 (2015). Kotloff, K. L. The burden and etiology of diarrheal illness in developing countries. Pediatr. Clin. North Am. 64, 799–814 (2017). Heo, I. et al. Modelling Cryptosporidium infection in human small intestinal and lung organoids. Nat. Microbiol. 3, 814–823 (2018). Wilke, G. et al. A stem-cell-derived platform enables complete Cryptosporidium development in vitro and genetic tractability. Cell Host Microbe 26, 123–134 (2019). Kim, H. J., Huh, D., Hamilton, G. & Ingber, D. E. Human gut-on-a-chip inhabited by microbial flora that experiences intestinal peristalsis-like motions and flow. Lab Chip 12, 2165–2174 (2012). Kasendra, M. et al. Development of a primary human Small Intestine-on-a-Chip using biopsy-derived organoids. Sci. Rep. 8, 2871 (2018). Trietsch, S. J. et al. Membrane-free culture and real-time barrier integrity assessment of perfused intestinal epithelium tubes. Nat. Commun. 8, 262 (2017). Chen, Y., Zhou, W., Roh, T., Estes, M. K. & Kaplan, D. L. In vitro enteroid-derived threedimensional tissue model of human small intestinal epithelium with innate immune responses. PLoS ONE 12, e0187880 (2017). Workman, M. J. et al. Enhanced utilization of induced pluripotent stem cell-derived human intestinal organoids using microengineered chips. Cell. Mol. Gastroenterol. Hepatol. 5, 669–677 (2018). Wang, Y. et al. Long-term culture captures injury-repair cycles of colonic stem cells. Cell 179, 1144–1159 (2019). Bhatia, S. N. & Ingber, D. E. Microfluidic organs-on-chips. Nat. Biotechnol. 32, 760–772 (2014). Wang, Y. et al. Bioengineered systems and designer matrices that recapitulate the intestinal stem cell niche. Cell. Mol. Gastroenterol. Hepatol. 5, 440–453 (2018).

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. © The Author(s), under exclusive licence to Springer Nature Limited 2020

Methods Data reporting No statistical methods were used to predetermine sample size. The experiments were not randomized. The investigators were not blinded to allocation during experiments and outcome assessment unless otherwise stated. Mice Intestinal crypts, bile ducts and intestinal myofibroblasts were extracted from 5–10-week-old heterozygous LGR5-eGFP-IRES-CreERT2 or wild-type C57BL/6J mice ( Jackson Laboratory), following animal experimentation protocols prescribed by EPFL and the Federation for Laboratory Animal Science Association (FELASA), in compliance with local animal welfare laws, guidelines and policies. Isolation of intestinal crypts Mouse intestinal crypts and single LGR5–eGFP ISCs were isolated following previously described procedures. In brief, the proximal part of the intestine was collected, opened longitudinally and washed with ice-cold phosphate-buffered saline (PBS). The luminal side of the intestine was scraped using a glass slide to remove luminal content and villous structures. After a second wash with ice-cold PBS, the intestine was cut into 2–4-mm pieces with scissors. The pieces were transferred to a tube and further washed with cold PBS (5–10 times) with gentle vortexing. Intestinal fragments were incubated in PBS containing 20 mM EDTA, for 20 min on ice. The supernatant was discarded and cold PBS was added to the fragments. Crypts were released by manual shaking of the suspension for 5 min. The supernatant was collected and passed through a 70-μm strainer. The remaining tissue fragments were again resuspended in cold PBS and triturated 5–10 times, and the supernatant was passed through a 70-μm strainer. The previous step was repeated once again. The three crypt-containing fractions were pooled together and centrifuged at 110g for 5 min. The pellet was resuspended in cold Advanced Dulbecco’s modified Eagle medium/Ham’s F-12 (Advanced DMEM/F12) supplemented with 1× Glutamax, 10 mM HEPES and 100 μg ml−1 penicillin–streptomycin (Gibco), and centrifuged at 84g to remove single cells and tissue debris. Crypts were then cultured according to the mouse intestinal organoids culture protocol (see ‘Cell culture’), and after three passages single viable LGR5–eGFPhigh cells were sorted by flow cytometry. Cell culture The following medium formulations were used in the protocol: base medium (BM) was prepared from Advanced DMEM/F12 medium supplemented with 1× Glutamax, 10 mM HEPES and 100 μg ml −1 penicillin–streptomycin (Gibco); BMGF medium was prepared from BM supplemented with 1× B27 supplement, 1× N2 supplement (Gibco) and 1 mM N-acetylcysteine (Sigma-Aldrich); ISC expansion medium (ENRCV) was prepared from BMGF supplemented with growth factors (50 ng ml−1 EGF (E) (Peprotech), 100 ng ml−1 Noggin (N) (EPFL Protein Expression Core Facility), 500 ng ml −1 R-Spondin 1 (R) (EPFL Protein Expression Core Facility)) and small molecules (3 μM CHIR99021 (C) (Stemgent) and 1 mM valproic acid (V) (Sigma-Aldrich)). Isolated crypts or single cells were embedded in Matrigel (Corning; growth-factor-reduced, phenol-red-free formulation) and cast into 25-μl droplets in a 24-well plate. After polymerization of Matrigel (15 min, 37 °C), 500 μl of ENRCV medium was added. For freshly extracted mouse crypts and single-cell culture, 2.5 μM Thiazovivin (Stemgent) was included in ENRCV medium for the first two days to prevent anoikis. Fresh medium was replenished every other day. For passage, organoids were removed from Matrigel and mechanically dissociated into smaller fragments, and then transferred to fresh Matrigel. Passaging was performed every fourth day with a 1:4 split

ratio. Organoids were used for experiments between passage number 5 and 20. A detailed protocol describing organoid culture has been deposited in the Protocol Exchange repository28. Intestinal myofibroblasts were isolated from C57BL/6J mice following a previously published protocol with slight modifications29. Mouse small intestinal tissue was processed and used for intestinal crypt isolation as described above. Remaining intestinal fragments were then washed and incubated in DMEM containing collagenase IV (300 U ml−1; Invitrogen) and dispase (0.08 U ml−1; Roche) for 30 min at 37 °C in a shaking water bath. The supernatant was then collected, centrifuged at 280g for 5 min and resuspended in DMEM supplemented with 10% heat-inactivated fetal bovine serum (HI-FBS), 100 μg ml−1 penicillin–streptomycin, 1× l-glutamine, 1× non-essential amino acids, 1× insulin–transferrin–selenium (Gibco) and 1× Primocin (InvivoGen). Cells were then transferred to T75 culture flasks, and the medium was changed after cell attachment (4–5 h) and every two days. Cells were split 1:2 as needed and used between passage 3 and 8. Collected liver tissues were minced and digested as previously described30. In brief, whole liver was incubated in collagenase XI (0.012%), dispase (0.012%) and FBS (1%) in Advanced DMEM/ F12 with 100 μg ml−1 penicillin–streptomycin, 1× Glutamax and 10 mM HEPES (Gibco), termed BM, for 2–3 h until bile duct fragments were visible. Remaining tissue pieces were let to sediment by gravitation and supernatant containing ductal fragments was collected and centrifuged at 200 rpm for 4 min and washed with PBS. Isolated fragments were resuspended in Matrigel (Corning; growth-factor-reduced, phenol-red-free formulation) and cast into 25-μl droplets in a 24-well plate. Following polymerization of Matrigel, 500 μl of isolation medium, containing BM supplemented with 1× B27 supplement, 1× N2 supplement (Gibco), 1.25 μM N-acetylcysteine (Sigma-Aldrich), 10 nM gastrin (Sigma-Aldrich) and growth factors (50 ng ml−1 EGF (Peprotech), 1 μg ml−1 R-Spondin 1 (EPFL Protein Expression Core Facility), 100 ng ml−1 FGF-10 (Peprotech), 10 mM nicotinamide (Sigma-Aldrich), 50 ng ml−1 HGF (Peprotech), 1 μg ml−1 WNT3A (Time Bioscience), 100 ng ml−1 Noggin (EPFL Protein Expression Core Facility) and 10 μM Y-27632 (Sigma-Aldrich) was added. After four days of culture, the medium was changed to expansion medium (EM) composed of isolation medium without WNT3A, Noggin and Y-27632. Organoids were passaged by mechanical dissociation every 10–14 days in split ratio 1:4 to 1:8. The use of human embryonic stem (ES) cell lines was authorized by the Office Fédéral de la Santé Publique (OFSP) after approval by the cantonal ethical commission (CER-VD). All the experiments reported in this Article were performed under authorization number R-FP-S-2-0014-0000. HES3 MIXL1GFP/+ human ES cells were obtained from A. G. Elefanty (Murdoch Children’s Research Institute) and routinely maintained in mTeSR1 medium (StemCell Technologies) on 6-well plates coated with Matrigel human ES cell-qualified matrix (Corning). The medium was changed every day. For routine culture the cells were passaged every 4–5 days as small clumps using the Gentle Cell Dissociation Reagent (StemCell Technologies). Macrophages were generated from HES3 MIXL1GFP/+ human ES cells as previously described31.In brief, during the first two days of differentiation, human ES cell colonies were specified to the mesoderm by incubation in StemPro Medium (Gibco) with 5 ng ml−1 BMP4, 50 ng ml−1 VEGF (Peprotech) and 2 μM CHIR99021 (Stemgent). In the next step (day 2–4) human haemangioblast-like cell formation was induced by replacing CHIR99021 with 20 ng ml−1 hbFGF (Peprotech) and later (day 4–6) maintained with VEGF and hbFGF only. The cells were then cultured for the next 16 days in StemPro Medium for haematopoietic stem cell differentiation with the following cytokines: differentiation day 6: 10 ng ml−1 VEGF, 10 ng ml−1 hbFGF, 50 ng ml−1 SCF, 30 ng ml−1 DKK-1, 20 ng ml−1 IL-3, 20 ng ml−1 TPO and 20 ng ml−1 FLT3 (Peprotech); differentiation day 8 and 10: 10 ng ml−1 VEGF, 10 ng ml−1 hbFGF, 50 ng ml−1 SCF, 30 ng ml−1 DKK-1, 20 ng ml−1 IL-3, 20 ng ml−1 TPO and 20 ng

Article ml−1 FLT3); differentiation day 12 and 14 (10 ng ml−1 hbFGF, 50 ng ml−1 SCF, 20 ng ml−1 IL-3, 20 ng ml−1 TPO and 20 ng ml−1 FLT3). Replating was performed starting from day 8 until day 22. The medium with cells was collected and centrifuged at 900g for 4 min. The resulting cell pellet was resuspended in fresh medium (1 ml per well) and plated into 6-well plates. From differentiation day 16, for myeloid differentiation the cells were switched to serum-free differentiation (SF-Diff) medium supplemented with 50 ng ml−1 hbFGF, and a full medium change was done every three days up to differentiation day 25, when the floating cells were collected and used for experiments. SF-Diff medium consisted of 50% IMDM, 50% DMEM/F12 with Glutamax, 1× N2 supplement, 1× B27 supplement, 0.5% cell-culture grade bovine serum albumin (BSA) and 100 μg ml penicillin–streptomycin (Gibco). StemPro Medium consisted of StemPro-34 SFM (Gibco), supplemented with 0.5 mM ascorbic acid (Sigma-Aldrich) and 100 μg ml−1 penicillin–streptomycin (Gibco). Human small intestinal organoids cryopreserved at passage 8 were provided by the H. Clevers laboratory (Hubrecht Institute) within the framework of collaboration agreements. Small intestinal organoids were established from duodenal biopsy samples from healthy human donors as previously described32. Endoscopic biopsies were performed at the University Medical Center Utrecht and the Wilhelmina Children’s Hospital. The patients’ informed consent was obtained, and this study was approved by the ethical committee of the University Medical Center Utrecht. Human small intestinal organoids were cultured in human ISC expansion medium composed of 50% L-WRN conditioned medium (1:1 dilution with BM) supplemented with 1× B27 supplement (Gibco), 1 μM N-acetylcysteine (Sigma-Aldrich), 50  ng ml−1 EGF (Peprotech), 500 nM A83-01 (Tocris), 10 nM gastrin (Sigma-Aldrich), 10 mM nicotinamide (Sigma-Aldrich), 10 μM SB202190 (Seleckchem), 10 nM prostaglandin E2 (Tocris). Y-27632 (10 μM; Seleckchem) was used in the first 48 h after single-cell dissociation to prevent detachment-induced cell apoptosis. L-WRN conditioned medium was prepared from L-WRN cells (CRL-3276; ATCC) following a published protocol33. The medium was changed every two days and the expanding organoids were passaged by mechanical dissociation using a fire-polished glass Pasteur pipette every six to eight days. Airway organoids were generated using healthy residual tissue from patients undergoing segmentary tracheal resection at the Centre Hospitalier Universitaire Vaudois (CHUV). The patients’ informed consent was obtained before sampling, and the use of anonymized tissue samples for in vitro organoid culture was approved by the cantonal ethical commission (CER-VD). Tracheal tissue was dissociated using previously published protocols34,35. In brief, tissue was minced and digested in airway organoid medium supplemented with 2 mg ml−1 collagenase (Sigma-Aldrich, C9407) on an orbital shaker at 37 °C for 3 h. Airway organoid medium was prepared from Advanced DMEM/F12 with 1× Glutamax, 10 mM HEPES, 100 μg ml−1 penicillin– streptomycin, 1× B27 supplement (Gibco), 1× Primocin (Invivogen) and 1.25 mM N-acetylcysteine (Sigma-Aldrich) supplemented with 100 ng ml−1 Noggin (EPFL Protein Expression Core Facility), 500 ng ml−1 R-Spondin 1 (EPFL Protein Expression Core Facility), 25 ng ml−1 FGF-7 (Peprotech), 100 ng ml−1 FGF-10 (Peprotech), 500 nM A8301 (Tocris), 5 mM nicotinamide (Sigma-Aldrich), 10 μM SB202190 (Seleckchem) and 5 μM Y-27632 (Seleckhem). The digested tissue suspension was sheared using flamed glass Pasteur pipettes several times. After each shearing step, the suspension was sequentially strained over a 100-μm filter and 2% FBS was added to the strained suspension before centrifugation at 400g. Erythrocytes were lysed in 2 ml red blood cell lysis buffer (Roche) for 5 min at room temperature. After centrifugation at 400g the resulting cell pellets were resuspended in Matrigel and cultured in airway organoid medium. The medium was changed every four days and organoids were passaged by mechanical dissociation using a fire-polished glass Pasteur pipette every two weeks.

Microdevice design and fabrication The microdevice is composed of three main compartments: a hydrogel compartment for organoid culture in the centre, two basal side medium reservoirs flanking the hydrogel compartment, and inlet and outlet medium reservoirs for perfusion of the microchannel (see microdevice schematic structure in Fig. 1, Extended Data Fig. 1). A 1,200-μm-wide and 1,500-μm-long central hydrogel chamber is sandwiched by two open basal side medium reservoirs (4 mm in diameter), separated from the hydrogel chamber by phase-guiding features. The phase-guiding features consist of semi-walls shielding the hydrogel compartment 200 μm from the top combined with a row of pillars spanning the entire height. This design allowed liquid hydrogel loading without spillage to the basal side reservoirs, as well as enabling passive medium diffusion to the basal side of epithelial tissues and/or matrix-embedded cells. From the other sides, the hydrogel chamber was connected to a pair of inlet and outlet reservoirs for medium perfusion and an extra matrix-loading port through which the hydrogel was loaded. Inlet and outlet medium reservoirs, 1.5 mm in diameter, had a dual function: as apical medium reservoirs for perfusion of mini-gut lumens; and to facilitate injection of medium in small quantities for functional tests or bacterial co-culture and so on. Additional smaller ports were designed to allow the connection of perfusion pump tubings to the inlet and outlet reservoirs; in this case inlet and outlet reservoirs also function as air-bubble traps. The microchip platform was fabricated using conventional soft-lithography methods established at the Center of Micronanotechnology (CMi, EPFL). In brief, the device was drawn using a CleWin (Phoenix Software). The designed layout was written with a diode laser with 2,000-nm resolution onto a fused silica plate coated with chrome and positive photoresist (Nanofilm) using an automated system (VPG200, Heidelberg Instruments). Exposed photoresist was removed with a developer (DV10, Süss MicroTec) and the chrome layer underneath was etched with an acid–oxidizer solution of perchloric acid, cerium ammonium nitrate and water. The resulting mask was developed with TechniStrip P1316 (Microchemicals) to remove the residual resist and extensively washed with ultra-pure water. The mould was made from multiple-layered epoxy-based negative photoresist SU8. First, a 200-μm thick layer of SU8 GM1075 (Gerlteltec) photoresist was cast onto a dehydrated silicon wafer using a negative resist coater (LMS200, Sawatec). The wafer was aligned and exposed to ultraviolet (UV) radiation through the first mask (MA6/BA6, Süss MicroTec). After baking at 95 °C, a second 200-μm thick layer of SU8 GM1075 was spin-coated, baked and exposed to UV through the second mask, carefully aligned using dedicated alignment marks. After the post-exposure bake, the wafer was developed with propylene glycol monomethyl ether acetate (Sigma) and baked at 135 °C for 4 h. The wafer was then plasma-activated and silanized with vapored trichloro (1H,1H,2H,2H-perfluorooctyl) silane (Sigma-Aldrich) overnight. This wafer was then used for polydimethylsiloxane (PDMS) moulding (Sylgard 184, Dow Corning). Ten weight-parts of elastomer base were vigorously mixed with 1 part of curing agent and poured onto the mould. After degassing under vacuum, PDMS was baked for 24 h at 80 °C. The resulting PDMS replica was cut and punched with appropriate size biopsy punchers. PDMS chips were soaked in a series of organic solvents to remove unreacted PDMS macromers. The resulting PDMS chips were exposed to oxygen plasma and irreversibly bonded on 35 mm glass bottom dishes (ibidi). Chips were sterilized with UV and kept sterile until further use. Hydrogel loading and microchannel fabrication An extracellular matrix mixture containing 75% (v/v) native bovine dermis type-I collagen solution (5 mg ml−1, Koken, AteloCell) neutralized with 1 M sodium bicarbonate, 10× Dulbecco’s modified Eagle’s medium and Advanced DMEM/F12 (Gibco) to generate 4 mg ml−1 solution and 25% (v/v) Matrigel (Corning, growth-factor-reduced, phenol-red-free

formulation) was injected into the hydrogel compartment of the microdevice through the hydrogel-loading port and incubated at 37 °C for 2 min, after which inlet and outlet and basal side medium reservoirs were filled with Advanced DMEM/F12 medium supplemented with 1× Glutamax, 10 mM HEPES and 100 μg ml−1 penicillin–streptomycin (Gibco). The stiffness of the resulting polymerized hydrogel was consistent from batch to batch and remained at 750 ± 50 Pa. Generation of the microchannels within the hydrogel was performed using a nanosecond laser system (1-ns pulses, 100-Hz frequency, 355 nm; PALM MicroBeam laser microdissection system, Zeiss) equipped with a 10×/0.25 NA objective, at a constant stage speed and a laser power35. A pattern of consecutive parallel lines was created in Wolfram Mathematica and then imported into the PALM MicroBeam system’s interface. Pattern was positioned along the microdevice matrix compartment, covering its entire length, 160 μm from the glass surface. Laser power and etching speed were adjusted to achieve 110–120 μm height microchannel in the hydrogel. After microchannel generation, the devices were perfused with Advanced DMEM/F12 medium, supplemented with 1× Glutamax, 10 mM HEPES and 100 μg ml−1 penicillin–streptomycin (Gibco), and maintained at 37 °C in 5% CO2 humidified air.

Mini-gut development and culture For mini-gut preparation, mouse intestinal organoids (between passage numbers 5 and 20) were dissociated into single cells. Organoids were removed from Matrigel, washed in BM (see medium formulations in ‘Cell culture’) and then dissociated with TrypLE express solution (Gibco) containing 2,000 U ml−1 DNaseI (Roche), 1mM N-acetylcysteine (Sigma-Aldrich) and 10 μM Y27632 (Stemgent) for 8 min at 37 °C. Dissociated cells were washed in BM, supplemented with 10% HI-FBS (Gibco) and passed through a 40-μm cell strainer. After centrifugation at 800g for 4 min, the pellet was resuspended in ENRCV medium containing 2.5 μM thiazovivin (Stemgent) (termed ENRCVT) at a density of about 106 cells per ml. After the removal of medium from the microchannel inlet, outlet and basal side medium reservoirs, 5 μl of cell suspension was introduced into the inlet and cells were allowed to fill the laser-ablated microchannel by gravity-driven flow. Cells were allowed to settle down in crypt-shaped cavities for about 3–6 min. All non-adherent cells were gently washed out from the microchannel by perfusion with ENRCVT. The medium reservoirs (inlet, outlet and basal side reservoirs) were filled with ENRCVT medium and the microdevice was incubated at 37 °C in 5% CO2 humidified air. The medium was changed every day. After the completion of epithelial tube formation (typically 2–3 days) ENRCV medium in the inlet and outlet reservoirs was replaced with differentiation medium (ENR), lacking CHIR99021 and valproic acid, keeping expansion ENRCV medium in the basal side medium reservoirs for another few days. After tissue maturation (typically 4–6 days; depending on the starting number of cells loaded into the tube and the stem cell proliferation rate), the concentration of CHIR99021 and valproic acid (C and V) in the basal side medium reservoirs was gradually decreased over the next three days until complete removal of these growth factors from the medium. Dead cells accumulating in the lumen were removed, and medium changes in the microchannel were performed, either manually or automatically, by perfusing with fresh medium every 12 h. The medium in the basal side medium reservoirs was changed every 24 h. Of note, mini-gut tube growth during the first two weeks was found to be very robust and reproducible, with approximately 9 out of 10 tissues developing properly (that is, acquiring similar morphology, cell-type composition and pattern along a crypt–villus-like axis). Tissue defects and failure of proper mini-gut development, detected in around 10% of cases, may be caused by technical problems related to microdevice preparation (for example, defective microfluidic chips), hydrogel preparation or issues related to organoid culture. The reliable long-term (more than two weeks) culture of mini-guts critically depends on repeated perfusions and medium changes, as well as adequate microdevice handling. Overall, approximately 8 out

of 10 mini-gut tubes could be successfully cultured for one month. A detailed protocol describing mini-gut development and culture has been deposited in the Protocol Exchange repository28. For macrophage mini-gut co-culture experiments, macrophages were resuspended in the ECM mixture (as described, 75% collagen/25% Matrigel (v/v)) and loaded into the hydrogel compartment of the microdevice. After hydrogel polymerization, BM supplemented with 50 ng ml−1 M-CSF (Sigma-Aldrich) was added to the inlet, outlet and basal side medium reservoirs. Laser ablation of the microchannel was performed using the described procedure. Macrophages were ablated in the microchannel region, but remained intact in the surrounding matrix. ISCs were seeded the next day, following the previously described protocol, and mini-guts co-cultured with macrophages were maintained in ENRWM medium (BMGF supplemented with growth factors (50 ng ml−1 EGF (Peprotech), 100 ng ml−1 Noggin (EPFL Protein Expression Core Facility), 500 ng ml−1 R-Spondin 1 (EPFL Protein Expression Core Facility), 50 ng ml−1 WNT3A (Time Bioscience) and 50 ng ml−1 M-CSF (Sigma-Aldrich)). Thiazovivin (2.5 μM) was added during the first 24 h to prevent apoptosis. After complete epithelium formation (2–3 days), ENRWM medium in the inlet and outlet reservoirs was replaced with differentiation medium (ENR). Over the next three days, the concentration of WNT3A in ENRWM medium in the basal side medium reservoirs was gradually decreased until complete removal from the medium. ENR and ENRM medium, in the inlet and outlet and the basal side medium reservoirs, respectively, was replenished every day. M-CSF was used in the basal side medium reservoirs to support the maintenance of macrophages. For myofibroblast mini-gut co-culture experiments, mouse intestinal myofibroblasts (used between passages 3 and 8) were washed with PBS and dissociated with TrypLE Express solution (Gibco) for 5 min at 37 °C. Dissociated cells were passed through a 40-μm cell strainer, centrifuged at 800g for 4 min and resuspended at a density of about 105 cells per ml in DMEM supplemented with 10% HI-FBS, 1× l-glutamine, 1× non-essential amino acids solution, 1× insulin–transferrin–selenium, 100 μg ml−1 penicillin–streptomycin (Gibco) and 1× Primocin (Invivogen), termed MyoDMEM. After the removal of medium from the inlet, outlet and basal side medium reservoirs, 5 μl of cell suspension was introduced into the inlet and cells were allowed to fill in the laser-ablated microchannel by gravity-driven flow. Cells were allowed to settle in crypt-shaped cavities for about 5–10 min and all non-adherent cells were gently washed out from the microchannel and inlet and outlet reservoirs. All medium reservoirs were filled with MyoDMEM medium and the microdevice was incubated at 37 °C in 5% CO2 humidified air. ISCs were co-seeded into mini-guts 3–4 h later, following the previously described protocol, and mini-guts co-cultured with myofibroblasts were maintained in ENRCVT medium supplemented with 5% HI-FBS. For bile duct tubes, mouse ductal organoids were digested with TrypLE Express solution (Gibco) into single cells for 10 min at 37 °C. Digestion was stopped by adding BM containing 10% HI-FBS (Gibco) and cells were passed through a 40-μm cell strainer. Single cells were seeded in the microdevice at a density of 0.5 million cells per ml. Cells were allowed to settle for 10 min and excess cells were gently washed out from the microchannel. The bile duct tubes were cultured in expansion medium (EM) supplemented with Noggin and Y-27632 for one day, then the medium was changed to EM. After three days, the medium was replaced by cholangiocyte differentiation medium with modifications36, BM supplemented with 1× B27 supplement, 1× N2 supplement (Gibco), 1.25 μM N-acetylcysteine (Sigma-Aldrich) and growth factors (50 ng ml−1 EGF (Peprotech), 50 ng ml−1 HGF (Peprotech), 0.1 μM dexamethasone (Sigma-Aldrich/Merck), 10 nM gastrin (Sigma-Aldrich) and 1.25 μM N-acetylcysteine (Sigma-Aldrich). The medium in both basal side medium reservoirs and the perfusion channel was changed every day. For generating human mini-gut and mini-airway tubes, organoids were dissociated into single cells in TrypLE Express solution (Gibco)

Article for 15 min at 37 °C and seeded into the microdevices as described above for mouse ISCs. Non-adherent cells were washed away from the lumen and human mini-guts were subsequently cultured for the first two days in medium containing EGF, Noggin, R-Spondin, gastrin and Y-27632, and then switched to human ISC expansion medium as previously described. The medium for human mini-guts and mini-airway tubes was changed every day. Human mini-airway tubes were cultured in airway organoid medium from day 0 onwards. At day 5, the medium in the inlet reservoirs was removed and an air–liquid interface culture established. Human umbilical vein endothelial cells (HUVECs) (Lonza) were maintained in EGMTM-2 Bulletkit medium (Lonza) and used until passage 12. For 3D culture of endothelial tubes, cells were washed with PBS and then dissociated with TrypLE Express solution (Gibco) for 5 min at 37 °C. Dissociated cells were passed through a 40-μm cell strainer (Falcon), centrifuged at 1,000 rpm for 4 min and resuspended in EGMTM-2 medium at a density of 106 cells per ml. ECM mixture (as described previously, 75% collagen/25% Matrigel (v/v)) was prepared and loaded into the hydrogel compartment of the microdevice. The microchannel layout was adapted to mimic a blood-vessel-like bifurcation. Laser ablation was performed using standard parameters and cell suspension was added to the inlet and outlet reservoirs. After 5–10 min of incubation, inlet and outlet reservoirs were washed with EGMTM-2 and all reservoirs were filled with EGMTM-2 medium. Endothelial tubes were incubated at 37 °C in 5% CO2 humidified air and medium was changed every day during the following two weeks of culture.

Immunofluorescence staining Mini-guts were rinsed with PBS and fixed in 4% paraformaldehyde (PFA; ABCR) for 30 min at room temperature. After rinsing with PBS, samples were permeabilized with 0.2% Triton X-100 (Sigma-Aldrich) in PBS (1 h, room temperature) and blocked in 10% goat serum in PBS (Gibco) containing 0.01% Triton X-100 (termed blocking buffer) for at least 5 h or overnight. Samples were subsequently incubated overnight at 4 °C with primary antibodies diluted in blocking buffer. The following primary antibodies were used: lysozyme (1:50; Thermo Fisher Scientific, PA1-29680), Muc2 (1:50; Santa Cruz, sc-15334), chromogranin A (1:50; Santa Cruz, sc-13090), L-FABP (1:50; Santa Cruz, sc-50380), E-cadherin (1:50, Abcam, ab11512), SOX9 (1:50; Abcam, ab185966), Ki67 (1:50, BD Pharmingen, 550609), GP2 (1:50, MBL, D278-3), EpCAM (1:200, Thermo Fisher Scientific 17-5791-82, APC-conjugated) and ZO-1 (1:50, Thermo Fisher Scientific 33-9100). After washing with blocking buffer for at least 6 h, samples were incubated overnight at 4 °C with secondary antibodies Alexa Fluor 647 goat anti-rabbit, Alexa Fluor 546 goat anti-mouse, Alexa Fluor 488 goat anti-rat, Alexa Fluor 568 donkey anti-mouse, Alexa Fluor 647 donkey anti-rabbit (1:500, Invitrogen), Alexa Fluor 546 phalloidin and Alexa Fluor 488 phalloidin (1:40, Invitrogen) diluted in blocking buffer. Samples were extensively washed for at least 24 h before imaging. Proliferative cells were stained with a Click-iT EdU Alexa Fluor 647 imaging kit (Thermo Fisher Scientific) following the manufacturer’s protocol. Microscopy and image processing Bright-field and fluorescent (eGFP, FITC–dextran) imaging of living mini-gut tubes was performed using a Nikon Eclipse Ti inverted microscope system equipped with a 4×/0.20 NA, 10×/0.30 NA and 20×/0.45 NA air objectives, 395-nm, 470-nm, 555-nm and 632-nm filters, DS-Qi2 and Andor iXon Ultra DU888U (Oxford Instruments) cameras and controlled by NIS-Elements AR 5.11.02 (Nikon Corporation) software. Fixed samples were imaged with a Zeiss LSM 700 Inverted Microscope (Bioimaging and Optics Core Facility), equipped with 10×/0.30 NA and 20×/0.80 NA air objectives, 405-nm, 488-nm, 555-nm and 639-nm lasers and controlled by ZEN 2010 imaging software (Zeiss). Image processing was mainly performed using ImageJ (NIH) using standard contrast and intensity level adjustments. Time-lapse images were acquired on

Nikon Eclipse Ti inverted microscope and processed in ImageJ (NIH) using plug-ins for SIFT linear stack alignment, illumination correction, as well as custom-made scripts developed at EPFL’s Bioimaging and Optics Core Facility. Extended depth of field (EDF) of bright-field images was calculated using built-in NIS-Elements function for a z-stack of 80–100 μm. Animated Supplementary Videos were rendered using Imaris (Bitplane), ImageJ (NIH) and Premiere Pro (Adobe).

Histology Samples were prepared by the Histology Core Facility (EPFL) in accordance with standard procedures. The hydrogel compartment was cut around its perimeter using a razor blade, which allowed us to extract a block of hydrogel containing mini-gut tubes. The samples were fixed in 4% PFA for 12 h at room temperature, washed in 1× PBS and cryoprotected in 1 M (or 30%) sucrose overnight at 4 °C. The samples were equilibrated for 1 h in 7.5% gelatin solution in 1 M sucrose/0.12 M phosphate buffer at 37 °C, placed in a mould filled with gelatin and frozen in cold isopentane (−70 °C). Sectioning was done using a Leica cryostat CM3050S at −30 °C. Section thickness was set at 8 μm. For staining, sections were hydrated in distilled water and immersed in Alcian Blue pH 2.5 solution for 25 min, counterstained with Nuclear Fast Red, dehydrated and mounted with a xylene-based glue. Sections were imaged on a LEICA DM 5500 microscope, DMC 2900 colour camera. Image processing was performed using ImageJ (NIH) using standard contrast and intensity level adjustments. TEM, sample preparation and imaging Microfluidic chips were cut around the hydrogel compartment. Blocks of hydrogel containing epithelial tubes were extracted from the chips and chemically fixed in a mix of 2.5 % glutaraldehyde and 2.0 % PFA in 0.1 M phosphate buffer (pH 7.4) and left for 4 h. Hydrogel blocks were trimmed with a razor blade to ensure that they were not larger than approximately 1 mm along one of the axes. The samples were washed thoroughly with cacodylate buffer (0.1 M, pH 7.4), post-fixed for 40 min in 1.0 % osmium tetroxide with 1.5 % potassium ferrocyanide, and then for 40 min in 1.0 % osmium tetroxide alone. The samples were stained for 30 min in 1% uranyl acetate in water before being dehydrated through increasing concentrations of alcohol and then embedded in Durcupan ACM (Fluka) resin. The samples were then placed in moulds and the resin polymerized at 65 °C for 24 h. Sections (50-nm thickness) were cut with a diamond knife, and collected onto single-slot copper grids with a pioloform support film. Sections were contrasted with lead citrate and uranyl acetate, and images were taken with a transmission electron microscope at 80 kV (Tecnai Spirit, FEI with Eagle CCD camera). Image processing was performed using ImageJ (NIH) using standard contrast and intensity level adjustments. Measurement of aminopeptidase activity For measurements of the specific activity of the apical brush border aminopeptidase, mini-guts were disconnected from the microfluidic syringe pump. The inlet reservoir was manually filled with 1.5 mM l-alanine-4- nitroanilide hydrochloride (A4N; Sigma-Aldrich) solution in ENR medium and allowed to flow through the lumen to the outlet reservoir. After 1 h incubation at 37 °C, the medium was collected from the outlet reservoir. Cleavage product (4-nitroaniline) was quantified using a NanoDrop spectrophotometer (Thermo Fisher Scientific) at 405 nm using ENR medium as a reference. Enzymatic activity was quantified based on the average, measured for n = 3 biologically independent samples. scRNA-seq The cDNA library was constructed using 10X genomics Chromium 3′ reagents v.3, and sequenced using Illumina protocol 15048776 using NextSeq v.2.5 reagents, with a read length of 52 nucleotides and around 70,000 reads per cell. The reads were aligned to mm10 with

Cell Ranger v.3.0.1. Raw count matrices were imported in Seurat 2.3.4 (ref. 37) using RStudio 1.1.463 (www.rstudio.com), and single live cells were selected on the basis of the number of detected genes (approximately 2,500–5,000) and fraction of mitochondrial genes (around 0.05–0.15). The number of cells after filtering was 998 (for organoids), 961 (‘young’ mini-guts), 1,000 (‘old’ mini-guts) and 611 (mini-guts infected with C. parvum). The data were normalized to 10,000 counts per cell, and log1p-transformed using the natural logarithm (referred to as log(expression)). The four datasets were aligned using Seurat canonical correlation analysis on the intersection of the most variable genes, and dimensionality reduction was conducted with UMAP38 in the aligned correlated component space. Louvain clustering yielded 40 clusters that were merged and named on the basis of canonical cell type markers. Cell-cycle scoring was based on published signatures39, and cells in the G2/M phases, which had a gene signature predominantly associated with cell division, were referred to as ‘dividing cells’. Gene sets highlighting villus-top enterocytes13, M cells9 and enteroendocrine cells38 were taken from the respective literature. A similar clustering analysis was performed on the dataset GSE92332 from a previous study9, designated as ‘atlas’, to obtain a reference of the approximate proportions of in vivo cell types. The datasets from mice B2 to B10 were used, whereas B1 was excluded because of strong technical differences to the others. The original cell-type annotations from the atlas were merged to fit our simplified naming: enterocyte immature distal, enterocyte immature proximal, enterocyte mature distal, enterocyte mature proximal, enterocyte progenitor and enterocyte progenitor late were referred to as ‘enterocytes’; stem, transit amplifying (TA) early and TA G1 were referred to as ‘stem and progenitor cells’; TA G2 and enterocyte progenitor early corresponded to dividing cells. The annotations for other cell types (Paneth, goblet, enteroendocrine and Tuft cells) already corresponded to our respective clusters. Visual representations of the data were generated using Seurat internal functions or ggplot240 and cosmetic adjustments were made in Adobe Illustrator. Simultaneous displays of several markers were generated from subtractive colour overlay. Gene set enrichment analysis (GSEA) was performed using the Broad Institute Java stand-alone application v.3.0, with single cells exported to each be considered as a stand-alone RNA-seq dataset associated to both a cell type and a treatment condition. The hallmark gene sets from MSigDB41 were scored across conditions either for all cell types as a bulk, or cell type by cell type, using signal to noise as a metric and the ‘classical’ enrichment statistic, and estimated the family-wise error-rate P values were empirically estimated based on 10,000 random phenotype permutations. The following algorithm was applied to define the best cell-type markers: (1) ribosomal proteins and genes with less than 25% dropout rate were excluded, because genes that are ubiquitously expressed or large families of closely related highly expressed proteins are not useful cell-type-specific biomarkers. (2) The gene expression from cells in each cluster was compared pairwise to that from cells in each other cluster by using the Wilcoxon rank-sum test. Only positive markers were considered. The highest P value and lowest average log ratio from the pairwise comparisons were kept as being the P value and log ratio for each gene in each cluster. (3) The following exceptions were made to the previous rule: (i) the cluster of dividing cells was not used in the pairwise comparisons when looking for markers for the other clusters, because dividing cells might display some markers from any cell type, especially progenitors. (ii) When looking for stem and progenitor cell markers, we further excluded Paneth cells from the pairwise comparisons, because Paneth cells are also expected to be positive for many canonical crypt and stem cell markers. (iii) When looking for enterocyte markers, we excluded ‘top enterocytes’ from the pairwise comparisons, because these cells are expected to express enterocyte markers. (iv) Paneth cells were excluded from the pairwise comparisons when looking for goblet cell markers, as some of the most commonly used goblet cell markers are secretory cell markers that are also expressed in Paneth

cells. (4) To define ‘markers’, we applied a filtering with a double cut-off on significance and log ratio: worst P value in pairwise comparisons less than 0.05, and worst log ratio higher than 1.25 (that is, ratio more than 3.5). The complete list is shown in the Source Data for Extended Data Fig. 5. (5) We defined the ‘best markers’ as the top markers when sorting by descending log ratio. The lists of best markers were used to generate the heat maps, in which the unscaled log(expression) of these genes was plotted against cells grouped by cell type, and sorted by UMAP coordinates within each cluster to facilitate visualization of gradients along the ISC-to-enterocyte axis. A Wilcoxon signed-rank test was used to compute the P values reported for volcano plots (across datasets, as a bulk or by cell type). To align the in vivo atlas to our in vitro datasets, we converted the two Seurat v.2 objects containing the four in vitro datasets and nine in vivo datasets, respectively, to Seurat v.3, and applied Harmony v.1.0 (ref. 42) alignment across modalities and across datasets. The alignment and dimensionality reductions were based on 30 dimensions. The result is shown as a UMAP in the Harmony-aligned space. For more information or complete reproduction, see ‘Data availability’ and ‘Code availability’.

Modelling intestinal epithelial damage and regeneration Mini-guts and intestinal organoids were treated with DSS (MP Biomedicals) to induce epithelial damage. DSS (0.05%) in ENR medium was administered through the inlet reservoir to perfuse the lumen of 7-day-old mini-guts and added to the wells with organoids on day 3 after passaging. After DSS treatment for 24 h, the medium was changed to ENR without DSS, and mini-gut tubes were further cultured for 12 days according to standard protocols. Organoids, collapsed in response to DSS, were carefully collected, washed, re-embedded in fresh Matrigel and cultured in expansion medium (ENRCV). Similar results were obtained in at least three independent experiments with three replicates. For irradiation experiments, mini-guts were exposed to 2-Gy and 8-Gy doses of gamma-radiation using Gammacell 1000 Elite 137Cs source (MDS Nordion) at 0,143 Gy s−1. After irradiation, samples were cultured according to standard protocols and recovery was monitored for up to 11 days. Relative LGR5–eGFP fluorescence was quantified using a specialized ImageJ plug-in for separate crypt regions, normalized to the ROI area and then normalized background intensity was subtracted. Experiments were repeated independently at least twice with three replicates per each condition with similar results. Rectangular gaps in the epithelium were generated using a nanosecond laser system (1-ns pulses, 100-Hz frequency, 355 nm; PALM MicroBeam (Zeiss) laser microdissection system) equipped with a 10×/0.25 NA objective, at a constant stage speed and a laser power. A pattern of consecutive parallel lines was positioned 20 μm above the bottom epithelium of the lumen. Laser power and etching speed were adjusted to induce cellular damage without ablation of the hydrogel. After epithelial tissue damage, samples were connected back to the automated syringe pump and perfused every 3 h, 0.25 μl per min for 20 min. Similar results were obtained in at least three independent experiments with three replicates. Infection of mini-guts with C. parvum C. parvum oocysts (Iowa strain, University of Arizona) were stored in sterile PBS with 100 μg ml−1 penicillin–streptomycin (Gibco) at 4 °C and used within two months. For mini-gut infection, around 1 × 106 oocysts were incubated in 10% (v/v) Clorox bleach in PBS on ice and washed three times with 1 ml BM medium and centrifugation (3 min, 8,000g, 4 °C). The oocysts were resuspended in 200 μl ENR medium supplemented with 0.5% (w/v) sodium taurocholate (Sigma-Aldrich). For mini-gut infection, a 5-μl suspension of oocysts was added directly to the inlet reservoir of the chip and left without perfusion for 6 h. Samples were perfused manually, in which 10 μl of medium was collected from the outlet reservoir and fresh ENR medium was added to the

Article inlet reservoir. In the following days, mini-guts were cultured as previously described, and perfusion was done manually twice a day. Media collected from infected mini-guts at different time points were pooled together and used for the detection of mouse inflammatory cytokines using a Multi-Analyte ELISArray kit (Qiagen, MEM-004A), following the manufacturer’s protocol. The plotted values were normalized to the ENR medium used for mini-gut perfusion. Samples were imaged every day before perfusion on a Nikon Eclipse Ti-E microscope equipped with DS-Qi2 camera and 10×, 20× and 63× objectives. For immunofluorescence analysis, mini-guts were processed as described. After staining with DAPI and secondary antibodies, samples were washed three times, and fluorescein-labelled antibodies Crypt-a-Glo or Sporo-Glo (Waterborne) were added to label oocysts or intracellular stages of the parasite, respectively, and incubated at room temperature for 1 h. After washing once with 0.1% Tween-20 in PBS, the samples were imaged on s Leica SP8 confocal microscope (Bioimaging and Optics Core Facility), using LAS X software (Leica) and processed using standard contrast and intensity level adjustments in ImageJ (NIH). In another set of experiments aimed at assessing whether newborn oocysts could be obtained, mini-guts were infected with the sporozoites. As previously described, about 1 × 106 oocysts were incubated in 10 % (v/v) Clorox bleach in PBS on ice and washed three times with 1 ml BM medium by centrifugation (3 min, 8,000g, 4 °C). Oocysts were resuspended in 1 ml of freshly prepared excystation medium with 1.5% (w/v) sodium taurocholate (Sigma-Aldrich) in BM and incubated for 1.5 h at 37 °C. Samples were checked microscopically for the extent of excystation and then incubated for an additional 30 min to reach approximately 60–80% excystation. After incubation 9 ml of BM was added to remove remaining oocysts and shells, the suspension was filtered through a Swinnex-25 47-mm (Millipore) apparatus with a 3-μm pore-size polycarbonate filter (Costar/Nucleopore). Another 5 ml of DMEM was added to wash the filter and then the filtered sporozoite suspension was centrifuged for 20 min at 3,400g to pellet sporozoites. The sporozoites were resuspended in 200 μl of ENR and 5 μl was added directly to the inlet reservoirs of the chips and left without perfusion for 12 h. In the following days, mini-guts were perfused once a day and all medium was collected from the lumen for immunostaining to Crypt-a-Glo and DAPI. Samples were fixed in 4% PFA and centrifuged for 3 min at 8,000g. The pellet was washed twice in PBS and centrifuged for 3 min at 8,000g. The last time pellet was resuspended in 50 μl of Crypt-a-Glo containing 1 μg ml−1 DAPI. After 2 h of incubation at room temperature, samples were washed twice in PBS and centrifuged for 3 min at 8,000g. Finally, pellets were resuspended in 10 μl of PBS, transferred to glass-bottomed 24-well plates (MatTek) and left overnight at 4 °C before imaging to allow all oocysts to homogeneously sediment on the glass-bottom surface. Imaging was done on Zeiss LSM 700 Inverted Microscope (Bioimaging and Optics Core Facility), equipped with 10×/0.30 NA and 20×/0.80 NA air objectives and controlled by ZEN 2010 imaging software (Zeiss). Quantification was performed using the standard toolkit from ImageJ. Of note, presumably owing to an incomplete filtration process, some remaining oocyst and broken shells were observed at days 1–3. The experiment was designed in such a way that all medium from the lumen was collected during perfusion, leaving a minimal number of unattached oocysts and sporozoites in the mini-guts.

Electron microscopy of the C.-parvum-infected samples For TEM, samples were fixed at day 1, 2, 3 and 5 and were processed as previously described43. In brief, the samples were fixed in 1% PFA and 2.5% glutaraldehyde in 0.1 M PB buffer for 2 h at room temperature, followed by 1 h incubation in 2% (w/v) osmium tetroxide and 1.5% (w/v) K4[Fe(CN)6] in 100 mM PB buffer. Samples were incubated for 1 h in 1% (w/v) tannic acid in 100 mM PB buffer, then 30 min in 2% (w/v) aqueous solution of osmium tetroxide followed by 1% (w/v) uranyl acetate for 2 h at room temperature. At the end of gradual dehydration cycles,

samples were flat-embedded in Epon-Araldite mix43. Polymerized flat blocks were trimmed using a 90° diamond trim tool, and the arrays of 100-nm sections were obtained using a 35° ATC diamond knife (Diatome) mounted on a Leica UC6 microtome. Sections were transferred to wafers using a modified array tomography procedure44. For the ultrastructure analysis, wafers were mounted on aluminium stubs and analysed using a FEI Helios Nanolab 650 scanning electron microscope (Thermo Fisher Scientific). The imaging settings were as follows: MD or ICD detectors, accelerating voltage: 2 kV, current 0.8 nA, dwell time 6 μs. Images were collected manually or using the AT module of the MAPS program (Thermo Fisher Scientific).

Reporting summary Further information on research design is available in the Nature Research Reporting Summary linked to this paper.

Data availability scRNA-seq data have been deposited to the Gene Expression Omnibus (GEO) public repository with the accession code GSE148366. Additional supporting data related to gene-expression analyses of mini-gut tubes infected with C. parvum have been deposited to https://figshare.com/ projects/mini-guts/80819. Source data are provided with this paper.

Code availability The code used for scRNA-seq data analysis is available at https://github. com/nbroguiere/miniguts. 28. Nikolaev, M. et al. Bioengineering microfluidic organoids-on-a-chip. Protoc. Exch. https:// doi.org/10.21203/rs.3.pex-903/v1 (2020). 29. Koliaraki, V. & Kollias, G. Isolation of intestinal mesenchymal cells from adult mice. Bio-protocol 6, e1940 (2016). 30. Huch, M. et al. In vitro expansion of single Lgr5+ liver stem cells induced by Wnt-driven regeneration. Nature 494, 247–250 (2013). 31. Takata, K. et al. Induced-pluripotent-stem-cell-derived primitive macrophages provide a platform for modeling tissue-resident macrophage differentiation and function. Immunity 47, 183–198 (2017). 32. Blokzijl, F. et al. Tissue-specific mutation accumulation in human adult stem cells during life. Nature 538, 260–264 (2016). 33. Miyoshi, H. & Stappenbeck, T. S. In vitro expansion and genetic modification of gastrointestinal stem cells in spheroid culture. Nat. Protocols 8, 2471–2482 (2013). 34. Sachs, N. et al. Long-term expanding human airway organoids for disease modeling. EMBO J. 38, e100300 (2019). 35. Brandenberg, N. & Lutolf, M. P. In situ patterning of microfluidic networks in 3D cell-laden hydrogels. Adv. Mater. 28, 7450–7456 (2016). 36. Chen, C. et al. Bioengineered bile ducts recapitulate key cholangiocyte functions. Biofabrication 10, 034103 (2018). 37. Butler, A., Hoffman, P., Smibert, P., Papalexi, E. & Satija, R. Integrating single-cell transcriptomic data across different conditions, technologies, and species. Nat. Biotechnol. 36, 411–420 (2018). 38. Becht, E. et al. Dimensionality reduction for visualizing single-cell data using UMAP. Nat. Biotechnol. 37, 38–44 (2019). 39. Kowalczyk, M. S. et al. Single-cell RNA-seq reveals changes in cell cycle and differentiation programs upon aging of hematopoietic stem cells. Genome Res. 25, 1860–1872 (2015). 40. Wickham, H. ggplot2: Elegant Graphics for Data Analysis (Springer, 2016). 41. Subramanian, A. et al. Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc. Natl Acad. Sci. USA 102, 15545–15550 (2005). 42. Korsunsky, I. et al. Fast, sensitive and accurate integration of single-cell data with Harmony. Nat. Methods 16, 1289–1296 (2019). 43. Kolotuev, I. Positional correlative anatomy of invertebrate model organisms increases efficiency of TEM data production. Microsc. Microanal. 20, 1392–1403 (2014). 44. Burel, A. et al. A targeted 3D EM and correlative microscopy method using SEM array tomography. Development 145, dev160879 (2018). 45. Mabbott, N. A., Donaldson, D. S., Ohno, H., Williams, I. R. & Mahajan, A. Microfold (M) cells: important immunosurveillance posts in the intestinal epithelium. Mucosal Immunol. 6, 666–677 (2013). 46. Nakato, G. et al. New approach for M-cell-specific molecules screening by comprehensive transcriptome analysis. DNA Res. 16, 227–235 (2009). 47. Hartl, M. & Schneider, R. A unique family of neuronal signaling proteins implicated in oncogenesis and tumor suppression. Front. Oncol. 9, 289 (2019). 48. Funda, D. P. et al. CD14 is expressed and released as soluble CD14 by human intestinal epithelial cells in vitro: lipopolysaccharide activation of epithelial cells revisited. Infect. Immun. 69, 3772–3781 (2001).

49. Nakamura, Y., Kimura, S. & Hase, K. M cell-dependent antigen uptake on follicleassociated epithelium for mucosal immune surveillance. Inflamm. Regen. 38, 15 (2018). 50. Hase, K. et al. Distinct gene expression profiles characterize cellular phenotypes of follicle-associated epithelium and M cells. DNA Res. 12, 127–137 (2005). 51. Dillon, A. & Lo, D. D. M cells: intelligent engineering of mucosal immune surveillance. Front. Immunol. 10, 1499 (2019). 52. Lim, J. S. et al. Caveolae-mediated entry of Salmonella typhimurium in a human M-cell model. Biochem. Biophys. Res. Commun. 390, 1322–1327 (2009). 53. Terahara, K. et al. Comprehensive gene expression profiling of Peyer’s patch M cells, villous M-like cells, and intestinal epithelial cells. J. Immunol. 180, 7840–7846 (2008). 54. Hase, K. et al. The membrane-bound chemokine CXCL16 expressed on follicle-associated epithelium and M cells mediates lympho-epithelial interaction in GALT. J. Immunol. 176, 43–51 (2006). 55. Kanaya, T. & Ohno, H. The mechanisms of M-cell differentiation. Biosci. Microbiota Food Health 33, 91–97 (2014). 56. Roulis, M. et al. Paracrine orchestration of intestinal tumorigenesis by a mesenchymal niche. Nature 580, 524–529 (2020).

Acknowledgements We thank M. Juhas for the generation of stem-cell-derived macrophages and help with co-culture experiments; R. Guiet and O. Burri for programming image-processing plug-ins; J. Dorsaz, J. Pernollet and all engineers of the Center of Micronanotechnology (CMi, EPFL) for support in microfabrication; S. Rezakhani, G. Sorrentino and K. Schoonjans for help with cholangiocyte isolation; D. Schaefer and M. Riggs for providing oocysts; R. O’Connor for expertise in analysing C. parvum epicellular stages; F. Gorostidi and S. Kishore for providing trachea tissue samples; and A. Manfrin, S. Höhnel, G. Rossi, M. Knobloch and A. Persat for inputs on the manuscript. We acknowledge support from the following EPFL core facilities: CMi, Histology, BIOP, CryoEM and CECF. This work was funded by the Swiss National Science Foundation (SNSF) research grant 310030_179447; the National Center of Competence in Research (NCCR) ‘Bio-Inspired Materials’ (https://www.

bioinspired-materials.ch/); the EU Horizon 2020 research programme INTENS (http://www. intens.info/); the Personalized Health and Related Technologies (PHRT) Initiative from the ETH Board; and EPFL. S.G. was supported in part by a fellowship from the Novartis Foundation for Medical-Biological Research. N.G. was supported in part by an EMBO Long-Term Postdoctoral Fellowship. Author contributions M.N. and M.P.L. conceived the study, designed experiments, analysed data and wrote the manuscript. N. Broguiere analysed all scRNA-seq data. S.G. developed an early version of the organoid culture technology and contributed to study design. M.N. and Y.T. designed the microdevice. M.N and O.M. conducted experiments on modelling damage and regeneration. O.M. performed human mini-gut and airway tube experiments. M.N. and D.D. performed and analysed C. parvum infection experiments. I.K. performed electron microscopy imaging of the C.-parvum-infected samples. B.E. performed bile duct tube experiments. N. Brandenberg and N.G. conducted preliminary experiments on matrix microstructuring and contributed to study design. H.C. proposed and designed C. parvum infection experiments and provided feedback on the manuscript. Competing interests The EPFL has filed for patent protection (EP16199677.2, PCT/EP2017/079651, US20190367872A1) on the scaffold-guided organoid technology described herein, and M.P.L., M.N., S.G., Y.T., N. Brandenberg and N.G. are named as inventors on those patents. M.P.L. and N. Brandenberg are shareholders in SUN bioscience SA, which is commercializing those patents. H.C. is an inventor on several patents related to organoid technology. The other authors declare no competing interests. Additional information Supplementary information is available for this paper at https://doi.org/10.1038/s41586-0202724-8. Correspondence and requests for materials should be addressed to M.P.L. Peer review information Nature thanks Dominic Grun, Thomas F. Meyer, Honorine Ward and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Reprints and permissions information is available at http://www.nature.com/reprints.

Article

Extended Data Fig. 1 | Bioengineering intestinal stem cell epithelia with a tubular, in-vivo-like architecture. a, Photograph of the fully assembled microchip system. b, Schematic cross-sectional view of laser ablation using a nanosecond-pulsed laser. c, Hybrid collagen I/Matrigel scaffold in the central chamber before and after microchannel ablation. Scale bars, 200 μm.

d, Fluorescence confocal images of a representative three-day old epithelial tube. Cells are labelled with DAPI (blue, nuclei) and E-cadherin (green). Images correspond to the maximal intensity projection of a z-stack of 100 μm. Scale bars, 50 μm. Data are representative of at least two independent experiments.

Extended Data Fig. 2 | Establishment of shape-controlled organoid culture from a variety of epithelial stem and progenitor cells. a, Development of bile duct tube composed of mouse cholangiocytes. b, Fluorescence confocal images of representative 3-day-old bile duct tubes, showing an entire tissue (top) stained for EpCAM (orange) and a higher magnification view (bottom) stained for actin filaments (green) and tight junction protein ZO-1 (red). Nuclei stained with DAPI (blue). Scale bars, 50 μm. c, Development of tubular mini-guts composed of human ISCs. d, Fluorescent confocal image showing formation of tightly packed single-layered epithelium in 15-day-old human mini-gut. e, Fluorescent confocal image showing proliferating Ki67+ cells

predominantly localized to the crypts in 10-day-old human mini-gut. Nuclei and actin filaments stained with DAPI (blue) and Phalloidin (green), respectively. Images correspond to the maximal intensity projection of a z-stack of 80 μm. f, Formation of a human mini-airway epithelial tube and establishment of air–liquid interface (ALI) culture from day 5 onwards. Scale bars, 100 μm. g, Fluorescence confocal image of a representative 7-day-old human mini-airway epithelial tube, showing an entire tissue (top) and a higher magnification view (bottom) stained for nuclei (DAPI, blue), E-cadherin (green) and Ki67+ proliferating cells (pink). Scale bars, 50 μm. All data are representative of at least two independent experiments.

Article

Extended Data Fig. 3 | Establishment of long-term culture and in vitro tissue homeostasis. a, Bright-field (middle) and LGR5–eGFP fluorescence (right) images of mini-gut progression on days 7 and 10, compared to organoids (left). b, Long-term propagation of mini-guts (up to 30 days) shown in bright-field (left) and LGR5–eGFP fluorescence (right). Data are represent ative of at least four independent experiments. EDF of bright-field images, calculated for a z-stack of 80 μm; fluorescence confocal images correspond to a maximal intensity projection of a z-stack of around 60 μm. Scale bars, 50 μm. c, Bright-field and LGR5–eGFP fluorescence of mini-gut deterioration due to the massive accumulation of dead cells within the lumen in the absence of

perfusion. Scale bars, 100 μm. d, Tubular mini-guts maintain epithelial integrity and morphology in different cell culture media used for lumina perfusion. No difference in epithelium morphology and stability was detected when tissues were apically exposed to organoid culture medium (ENR) or minimal media lacking growth factors (BMGF, BM). Similar results were obtained in at least two independent experiments with n = 2 samples per each condition. Scale bars, 50 μm. e, Frequency map showing the localization of LGR5–eGFP-expressing ISCs in 7-10-day-old tissues (left) and 30-day-old tissues (right). Average of the maximal intensity projection of a z-stack of around 60 μm for n = 20 tissues (7–10-day-old) and n = 8 tissues (30-day-old).

Extended Data Fig. 4 | Mini-gut tubes undergo rapid cell turnover and comprise key functional intestinal cell types. a, Epithelial tissue turnover assessed through EdU pulse-chase experiments. Ten-day-old mini-guts were treated with EdU for 12 h basally and apically, followed by a chase period of four days. At 0 h after EdU removal, the majority of EdU+ cells resided within the crypts and adjacent regions. A 24-h EdU pulse-chase revealed distinct regions of cell proliferation that were to a large extent restricted to the crypts. Two days after the EdU pulse, numerous EdU+ cells were found in the lumen, suggesting the occurrence of intestinal epithelial cell migration from the crypts to the villus-like domains. Labelled cells were virtually absent four days after the EdU pulse, suggesting that tubular mini-guts underwent full turnover of the epithelium. Data are representative of one EdU labelling experiment with n = 2 replicates per condition. Scale bars, 50 μm. b, Micrograph of a mini-gut tube

removed from the microchip for downstream histological sectioning and analysis. Scale bar, 200 μm. c, Histological cross-sections of 7-day-old mini-gut tubes stained with Alcian Blue showing acidic polysaccharides of the mucus layer (blue) counterstained with Nuclear Fast Red. The entire perpendicular section (left) and a higher magnification view of the goblet cells (right) are shown. Similar results were obtained for 10 sections from two biologically independent samples. Scale bars, 20 μm. d, Transmission electron microscopy cross-sectional views of 7-day-old mini-gut tube. Goblet cell (left; scale bar, 2 μm) and enterocyte brush border (right; scale bar, 0.3 μm). Data are representative of two samples. e, Gradual increase in aminopeptidase activity after induction of differentiation in mini-gut tubes. Mean ± s.d. from n = 3 biologically independent experiments.

Article

Extended Data Fig. 5 | Canonical markers from the various intestinal cell types are accurately reproduced in vitro. Heat map of unbiased top markers (Methods) associated with the various cell type clusters found in vitro. Canonical signature genes of stem cells (Olmf4, Lgr5, Axin2) form a continuous gradient towards enterocytes (Fabp1, Apoa1) and villus-top enterocytes (Ada, Krt20). Paneth cells express an array of antibacterial markers (defensins,

lysozyme), enteroendocrine cells are marked by chromogranins (Chga, Chgb) and subpopulation-specific hormones. Tuft cells selectively express the landmark gene Dclk1, as well as eicosanoid biosynthesis pathway enzymes (Alox5ap, Hpgds, Ltc4s) and receptors implicated in taste transduction (Trpm5, Gng13).

Extended Data Fig. 6 | Cell types identified in vitro closely resemble their in vivo counterparts. a–f, h, Overlay of canonical cell type markers expression (a, d), cell-cycle phase (c, f) and corresponding attributed cell types (b, e, h) found in vitro (a–c) and in vivo (d–f). g, i, Combined aligned in vitro and in vivo

datasets showing good match between the cell types identified in the separate analyses. The in vivo versus in vitro datasets (g, i) are generated with different protocols, and gene expression values are therefore not directly comparable between the two.

Article

Extended Data Fig. 7 | Identification of rare cell types in the mini-guts. a, Dot plot highlighting genes relevant to the identification of the small cluster designated as microfold-like (M-like) cells that share similarities with M cells residing in follicle associated epithelia (FAE) in vivo. M-like cells express the canonical immature M-cell markers Anxa5 and Marcksl145, involved in gram-negative bacteria binding/endocytic transport45,46 and regulation of cytoskeleton/adhesion, respectively47. Other genes related to bacterial sampling are also expressed, including Prnp 46, Cd14 48 and Aif1l49. M-like cells also selectively express additional phagocytosis-related markers such as Myadm and Cyba. Notably, transcripts marking mucus secretion (sum of Muc1, Muc2, Muc3, Muc3a, Muc4 and Muc13, here referred to as ‘Mucins’) and IgA transcytosis (Pigr) are missing, which is another trait of FAE47. Several other genes related to cytoskeleton and adhesion are also strongly upregulated in this population, including the FAE/M-cell markers Actn112,50 and Itgb149. Additional similarities to transcripts marking M-cells include the tight junction marker Cldn4 (Claudin 4) involved in antigen sampling/endocytosis45,51, the caveolae marker Cav1 52 and the cytokine Cxcl1653 that mediates lympho-epithelial interaction in gut associated lymphoid tissue 54, as well as several upregulated NFκB target genes 51. Several other known FAE and M-cell markers are missing in these M-like cells, including Spib, the master controller of M-cell differentiation acting downstream of RANKL signalling 55. This suggested that M-like cells in mini-gut tubes are only partially analogous to M-cells. We noted that our M-like cell population also shared many

transcriptional similarities with two recently described, rare cell populations in the intestine, namely ‘revival stem cells’ (RSCs)14,56 and regenerative fetal-like stem cell15,56. In particular, M-like cells in mini-guts were found to selective express the RSC markers Clu and Msln14,56, previously reported as FAE/M-cell markers9,46,53, and Ly6a (Sca1), that also defines regenerative fetal-like epithelial cells15,56. A characteristic feature of both RSCs and fetal-like stem cells is the activation of the YAP pathway, mediated by focal adhesions, inflammation or prostaglandin E2. Both YAP target genes and prostaglandin-related genes were found to be strongly and selectively expressed in our M-like cell population as well. b, Fluorescence confocal images of representative 15 days-old mini-gut tube, showing an entire tissue (left column) and a higher magnification view (right column) containing GP2+ (red) M-like cells. Data are representative of two replicates. Scale bar, 100 μm. c, Expression of the key enteroendocrine genes in the mini-guts tubes. Neurog3, a marker of immature enteroendocrine cells, forms a gradient towards Chga, Chgb and Neurod1, marking mature enteroendocrine cells. Furthermore, a subpopulation of the enterochromaffin cells defined by hormone substance P (Tac1) and Tph1, encoding the rate-limiting enzyme in serotonin synthesis, can be detected. A subpopulation of cholecystokinin producing I-cells (Cck) was found, co-expressing proglucagon products (Gcg), and varying levels of peptide YY (Pyy), ghrelin (Ghrl) and gastrin (Gast). Enteroendocrine cells were also found to highly express Wnt3, which may partially contribute to the observed higher number of stem cells in mini-guts.

Extended Data Fig. 8 | Capacity of mini-gut tubes to regenerate after radiation-induced damage. a, Mini-gut tubes fail to regenerate and rapidly deteriorate upon exposure to 8 Gy radiation dose. Overlaid bright-field and LGR5–eGFP fluorescence time-course images of epithelial damage in mini-gut tubes induced by exposure to 8 Gy radiation dose are shown. Scale bars, 60 μm. b, Rapid replenishment of LGR5–eGFP+ stem cells initially eliminated by exposure to 2 Gy radiation dose. A graph showing normalized fluorescence intensity of eGFP measured in the crypts before and after exposure to 2 Gy and

8 Gy radiation dose. Mean ± s.d. for n = 4 samples. c, Time-course of mini-gut tubes regeneration upon 2 Gy radiation dose-induced damage shown in bright-field and LGR5–eGFP fluorescence. EDF of bright-field images, calculated for a z-stack of 80 μm; fluorescence confocal images correspond to a maximal intensity projection of a z-stack of around 60 μm. Scale bars, 60 μm. All data are representative of at least two independent experiments with n = 3 replicates per each condition.

Article

Extended Data Fig. 9 | See next page for caption.

Extended Data Fig. 9 | Modelling C. parvum infection in mini-gut tubes. a, Schematic representation of the C. parvum life cycle and how it can be assessed in mini-gut tube cultures. b, Bright-field live imaging of C. parvum infection in mini-guts with major epicellular stages. After about 24 h of infection, floating half-empty oocysts, broken shells and freshly excysted sporozoites were observed; on the following day, 6–8-merozoite-containing type I meronts and 4-merozoite-containing type II meronts could be detected; 3 days post infection microgamonts containing 12–16 microgametes were detected and starting from day 5 oocysts containing 1–4 sporozoites were again observed. Identity of the observed epicellular stages was confirmed by specific immunostaining (Fig. 3d). Scale bars, 3 μm. c, d, Scanning electron microscopy of distinct stages of C. parvum life cycle with c, different epicellular stages observed in a single cross-section, including microgamont, early zygote and developing oocyst in the mini-gut tubes 96 h after infection. Scale bar, 20 μm. d, All major epicellular stages of the C. parvum were observed in other samples, including trophozoite (24 h post infection), type II meront (48 h post

infection) and macrogamont, early zygote and developing oocyst (72–96 h post infection). Scale bars, 1 μm. Data are representative of independent observations from one experiment. e, Quantification of oocysts produced in mini-gut tubes over four weeks. Mean ± s.d. for n = 3 replicates analysed. Data are representative of independent observations from one experiment. f, Hallmark pathways from the molecular signature database (MSigDB) enriched in C.-parvum-infected mini-guts compared to control tissues, as estimated from GSEA. The epithelium responds to the infection through interferon-α, with a family wise error rate (FWER). P value 3G5)()()/&"H(5H$## r; IFJKLMNOONFLPQRPRLRSNTJULVQLUQRMWXVQULRNJQJKLVKLFYOQZLUQRMWXVQLONWQLMNO[JQ\LPQMSFX]TQRLXFLPSQL^QPSNURLRQMPXNF_ r >#4$ !($"8&7746&$&( ((# r >#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$4$"(+ r >8 >3G6&$&($"*:%:(&"#&##6$&($"+& 4$&(#($2&( 8'"4(&$"(/*:%:4"8$#"4$"(6&7+ 7)/!()$(($"%=()(((&($($4*:%:a=P=W+5$()4"8$#"4$"(6&7=884($E=#% 88#2&"#b6&7'"(# r "($.#$ '# ~&7$#&($"

C()#

"D& }"676#$"() ('#/ r r

r

)}AHm A*0w=;)28$)=&(…C>qHw€€wo+,2"47"&7&"($.#/!#'4#$"2'D47"@ww:A'$8$#./&88$"$(/ 4)2&(%&!)/88754/(2(/ &"($GxxHf0}oooH‚:A'$8$# ./&88$"$(/4)2&(%&!)/88754/(2(/ &"($HGqqHf0}HowƒH7 7!7(&4"('!7(5$()'(7$!'#47!7(: r >"'2$ 4&76&7'8"'2.8477!4"(&%*5$() (&($($4+$!6$##:

C()#7%/

1&2!7!!&&($"



}"('2"( 18(5&

&+)&&4($E&($"80f22.&"./8754/(2(/:0f &2!7 5#$7'(#&(p:q)&2&(4$(=5&)#(5$45$()Af1 &"#$"4'.&(#5$()!$2&/2'2"47"&7&"($.#$ 8w)&(y€ˆ:8($"4'.&($"=47755&)#(5$4$"Af1&"#&"&7/E#"()8754/(2(: .+0f!8"4$"6&$"& &/:0f 5 (&$"#5$()()4"4"(&($" 877;&4#/$"Af18 w)&(y€ˆ:1„f0k"}5& B4$(#./&.7'7&&"##(4(#./&qypDyp8$7(:77;&4%&($"% 8() (&($"%477!!'7&($"*Aw+#8$"#&77()0fD!0f=()"!!!$&(4"(75'"5$()" "&  "($.#$ 7&.77#477: 77!!'7&($"&.'"#&"4 &>+"&)7/& 77!!'7&($" 5)$%)()&"‚p 8()((&76"(4##$ $"& " 774&:



Š

Article

Hydroxychloroquine use against SARS-CoV-2 infection in non-human primates https://doi.org/10.1038/s41586-020-2558-4 Received: 30 April 2020 Accepted: 10 July 2020 Published online: 22 July 2020 Check for updates

Pauline Maisonnasse1,11, Jérémie Guedj2,11, Vanessa Contreras1,11, Sylvie Behillil3,4,11, Caroline Solas5,11, Romain Marlin1,11, Thibaut Naninck1, Andres Pizzorno6, Julien Lemaitre1, Antonio Gonçalves2, Nidhal Kahlaoui1, Olivier Terrier6, Raphael Ho Tsong Fang1, Vincent Enouf3,4,7, Nathalie Dereuddre-Bosquet1, Angela Brisebarre3,4, Franck Touret8, Catherine Chapon1, Bruno Hoen9, Bruno Lina6,10, Manuel Rosa Calatrava6, Sylvie van der Werf3,4, Xavier de Lamballerie8 & Roger Le Grand1 ✉

Coronavirus disease 2019 (COVID-19) has rapidly become a global pandemic and no antiviral drug or vaccine is yet available for the treatment of this disease1–3. Several clinical studies are ongoing to evaluate the efficacy of repurposed drugs that have demonstrated antiviral efficacy in vitro. Among these candidates, hydroxychloroquine (HCQ) has been given to thousands of individuals infected with severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2)—the virus that causes COVID-19—worldwide but there is no definitive evidence that HCQ is effective for treating COVID-194–7. Here we evaluated the antiviral activity of HCQ both in vitro and in SARS-CoV-2-infected macaques. HCQ showed antiviral activity in African green monkey kidney cells (Vero E6) but not in a model of reconstituted human airway epithelium. In macaques, we tested different treatment strategies in comparison to a placebo treatment, before and after peak viral load, alone or in combination with azithromycin (AZTH). Neither HCQ nor the combination of HCQ and AZTH showed a significant effect on viral load in any of the analysed tissues. When the drug was used as a pre-exposure prophylaxis treatment, HCQ did not confer protection against infection with SARS-CoV-2. Our findings do not support the use of HCQ, either alone or in combination with AZTH, as an antiviral drug for the treatment of COVID-19 in humans.

Infection with SARS-CoV-2 is characterized by initial mild disease associated with respiratory symptoms at the peak of viral replication1,8. In some patients, a late severe immunological syndrome occurs 6–14 days after the onset of symptoms that may require intensive care and is responsible for most of the fatalities1–3. HCQ has well-documented in vitro activity against various viruses4 and has emerged as an active compound against SARS-CoV-2 in different screening programmes, including a library of 1,520 Food and Drug Administration (FDA)-approved compounds5. In Vero E6 cells, HCQ has a 50% maximal effective concentration (EC50)5,9,10 that varies between 0.7 and 4 μM. It may inhibit viral transport in endosomes by alkalinizing the intra-organelle compartment10,11 and affect glycosylation, as reported for other viruses12. The drug may also act as an immunomodulatory agent13,14. In patients with lupus, HCQ decreases the level of inflammatory cytokines11,15,16, which may be relevant for the treatment of COVID-192. Furthermore, it has been proposed that AZTH, which displays in vitro

antiviral activity against SARS-COV-25,17, could potentiate the efficacy of HCQ6. On the basis of these properties, HCQ has been considered for the treatment of COVID-19, alone or in combination with AZTH6,7. We and others have set up non-human primate (NHP) models of SARS-CoV-2 infection18–20. Here we used cynomolgus macaques (Macaca fascicularis) to test different treatment strategies with HCQ, alone or in combination with AZTH, before or after the peak of viral replication. We also tested HCQ administration as pre-exposure prophylaxis treatment against SARS-CoV-2 infection.

In vitro efficacy of HCQ against SARS-CoV-2 infection We first evaluated the in vitro antiviral activity of HCQ against a SARS-CoV-2 strain isolated from one of the first patients with COVID19 in France. Post-infection treatment of Vero E6 cells with HCQ resulted in a dose-dependent antiviral effect, with 50% inhibitory concentration

Center for Immunology of Viral, Auto-immune, Hematological and Bacterial diseases (IMVA-HB/IDMIT), Université Paris-Saclay, Inserm, CEA, Fontenay-aux-Roses, France. 2Université de Paris, IAME, Inserm, Paris, France. 3Unité de Génétique Moléculaire des Virus à ARN, GMVR, Institut Pasteur, UMR CNRS 3569, Université de Paris, Paris, France. 4Centre National de Référence des Virus des infections respiratoires (dont la grippe), Institut Pasteur, Paris, France. 5Laboratoire de Pharmacocinétique et Toxicologie, Aix-Marseille Université, APHM, Unité des Virus Emergents (UVE) IRD 190, INSERM 1207, Hôpital La Timone, Marseille, France. 6CIRI, Centre International de Recherche en Infectiologie, (Team VirPath), Université de Lyon, Inserm, U1111, Université Claude Bernard Lyon 1, CNRS, UMR5308, ENS de Lyon, Lyon, France. 7Plate-forme de microbiologie mutualisée (P2M), Pasteur International Bioresources Network (PIBnet), Institut Pasteur, Paris, France. 8Unité des Virus Emergents (UVE), Aix-Marseille Université, IRD 190, INSERM 1207, IHU Méditerranée Infection, Marseille, France. 9Emerging Diseases Epidemiology Unit, Institut Pasteur, Paris, France. 1

10 Laboratoire de Virologie, Centre National de Référence des Virus des infections respiratoires (dont la grippe), Institut des Agents Infectieux, Groupement Hospitalier Nord, Hospices Civils de Lyon, Lyon, France. 11These authors contributed equally: Pauline Maisonnasse, Jérémie Guedj, Vanessa Contreras, Sylvie Behillil, Caroline Solas, Romain Marlin. ✉e-mail: [email protected]

584 | Nature | Vol 585 | 24 September 2020

Control n = 8 Hi D1 n = 5 n=4

Hi D1 n = 5 +AZTH Lo D5 n = 4 PrEP

n=5

0

4

8

12

16

20

Time (d.p.i.)

–8 –4

0

4 8 12 16 20 –8 –4 Time (d.p.i.)

0

0

MF14 MF15 MF16 MF17

4 8 12 16 20 –8 –4 Time (d.p.i.)

0

Hi D1 + AZTH MF23 MF24 MF25 MF26

Fig. 1 | Study design and viral loads in the respiratory tract of SARS-CoV2-infected cynomolgus macaques treated with HCQ and AZTH. a, Study design. The red dotted line indicates infection with 106 PFU of SARS-CoV-2 by the combined intranasal and intratracheal routes. Coloured areas indicate HCQ treatment periods. Each group received either a high (Hi) or a low (Lo) dose of HCQ according to the regimens described in the Methods. The treatment started 1 d.p.i. (D1) or 5 d.p.i. (D5), or 7 days before viral challenge for the pre-exposure prophylaxis (PrEP) group. One group received AZTH in

(IC50) values of 2.2 μM (0.7 μg ml−1) and 4.4 μM (1.4 μg ml−1) at 48 and 72 h after infection, respectively, which is within the range of previously reported values21 (Extended Data Fig. 1a). We next studied infection in a model of reconstituted human airway epithelium (MucilAir, Epithelix) developed from primary nasal or bronchial cells differentiated and cultured in an air–liquid interphase22. In contrast to previous observations for remdesivir23, the antiviral activity of HCQ in Vero E6 cells did not translate to the human airway epithelium model; doses of 1 μM or 10 μM HCQ did not significantly reduce SARS-CoV-2 apical viral titres at 48 h after infection (Extended Data Fig. 1b). HCQ also did not protect the integrity of epithelial tissue during infection, as the trans-epithelial electrical resistance values were comparable with the values of untreated cells and significantly lower than those of the mock-infected controls.

Infection of macaques with SARS-CoV-2 Cynomolgus macaques were infected on day 0 with a total dose of 106 plaque-forming units (PFU) of a primary SARS-CoV-2 isolate (BetaCoV/ France/IDF/0372/2020; passaged twice in Vero E6 cells) by combined intranasal and intratracheal routes. Control NHPs (n = 8) had high viral loads in nasopharyngeal and tracheal samples (swabs), as estimated by quantitative PCR with reverse transcription (RT–qPCR), as early as 1 day after infection (d.p.i.). In tracheal samples, the viral load peaked at 2 d.p.i. (Fig. 1b and Extended Data Fig. 2a), with a median peak value of 7.9 log10 copies per ml. After 2 d.p.i., the viral loads progressively decreased and most NHPs had undetectable viral loads by 10 d.p.i. Similar profiles were observed for nasopharyngeal shedding (Extended Data Fig. 2b), whereas low viral loads were detected for more than 3 weeks in rectal samples and bronchoalveolar lavages (Extended Data Fig. 2c, d). NHPs exhibited mild clinical signs, including coughing or sneezing without dyspnoea, as has been reported for most patients with COVID-19 during the early infection period. The NHPs also developed early lymphocytopenia at 2 d.p.i. (Extended Data Fig. 5). No major changes were observed in heart rate, respiratory rate and oximetry analyses. Typical focal ground glass opacities associated with pleural thickening24,25 were observed in computed tomography (CT) scans with variable degrees of severity (Fig. 2 and Extended Data Fig. 3). Lesions were detectable as early as 2 d.p.i. and persisted up to 13 d.p.i. in some NHPs. None of the control NHPs developed a severe disease similar to what is observed in the late stages of the severe forms of the disease in humans.

0

4 8 12 16 20 Time (d.p.i.)

d

PrEP MF27 MF28 MF29 MF30 MF31

MF18 MF19 MF20 MF21 MF22

4 8 12 16 20 –8 –4 Time (d.p.i.)

c

Lo D1 MF9 MF10 MF11 MF12 MF13

Lo D5

10 9 8 7 6 5 4 3 –8 –4

Hi D1

MF5 MF6 MF7 MF8

4 8 12 16 20 –8 –4 Time (d.p.i.)

0

Median viral load in throat (log10[copies per ml])

–4

Control MF1 MF2 MF3 MF4

10 9 8 7 6 5 4 3

4 8 12 16 20 Time (d.p.i.)

10 9 8 7 6 5 4 3 0

4 8 12 16 20 Time (d.p.i.)

0

4 8 12 16 20 Time (d.p.i.)

Median viral load in BAL (log10[copies per ml])

–8

Lo D1

b

SARS-CoV-2 infection

Viral load (log10[copies per ml])

a

10 9 8 7 6 5 4 3

combination with a high dose of HCQ. The control group received vehicle (water) as placebo. b–d, Viral loads were analysed by PCR in throat swabs (b, c) and bronchoalveolar lavages (BAL) (d). The limit of detection was estimated to be 2.3 log10 copies of SARS-CoV-2 RNA per ml and the limit of quantification was estimated to be 3.9 log10 copies per ml (dotted horizontal line). b, Shaded zones indicate treatment periods and each symbol and line combination represents one NHP. Dotted vertical lines indicate day of treatment initiation. c, d, Data are represented as medians of each group as described in a.

Treatment with HCQ To assess the anti-viral efficacy of HCQ, macaques received HCQ daily by gavage for 10 or more days. A treatment regimen of 90 mg kg−1 on 1 d.p.i. (loading dose) followed by a daily maintenance dose of 45 mg kg−1 was found to generate a clinically relevant plasma drug exposure in a group of uninfected NHPs (Extended Data Fig. 4b). In parallel, we also tested a lower treatment regimen, with a loading dose of 30 mg kg−1 and a maintenance dose of 15 mg kg−1. Overall, 9 NHPs were infected on day 0 and treated using the high treatment regimen (Hi D1, n = 5) or the low treatment regimen (Lo D1, n = 4), both starting at 1 d.p.i. We also examined the effect of a late low-dose treatment starting at 5 d.p.i.—when viral RNA levels are 3–4 log lower compared with peak values—to evaluate the benefit of HCQ in accelerating the clearance of the virus (Lo D5, n = 4). We focused on RT–qPCR analyses to assess the in vivo antiviral efficacy of HCQ because it provides a quantitative analysis, has a higher sensitivity and is less prone to variability than culture-based assays. In addition, RT–qPCR analysis is the only method that enables a comparison with results reported in human patients. Furthermore, virus titration in culture assays can be affected by many factors in addition to the number of viral particles, including any residual HCQ in the samples and host factors such as cytokines. All treated NHPs had tracheal viral RNA load kinetics that were similar to those of untreated NHPs, with median peak viral loads of 7.1 and 7.5 log10 copies per ml for the Hi D1 and Lo D1 groups, respectively, compared with 7.9 log10 copies per ml in the control group. Similarly, the areas under the curve (AUCs) of the viral load were similar between all groups, with values of 36.9 and 39.7 log10 copies × day per ml, for the Hi D1 and Lo D1 groups, respectively, compared with 40.3 log10 copies × day per ml in control NHPs (P = 0.62 and P = 0.37, respectively). Similar results were obtained for the nasopharyngeal swabs, and there were no differences in the levels of viral replication in bronchoalveolar lavages (Fig. 1d and Extended Data Fig. 2). In NHPs treated from 1 d.p.i. or 5 d.p.i., HCQ did not accelerate the time to viral clearance, and the median times to the first unquantifiable viral load were 4.5, 7.0, 7.0 and 7.0 days in the control, Lo D1, Hi D1 and Lo D5 groups, respectively. Next, we evaluated the combination therapy of HCQ and AZTH, which was administered from 1 d.p.i., in which HCQ was given as a high dose as described above, and AZTH was given at a loading dose of 36 mg kg−1 followed by a daily dose of 18 mg kg−1 to mimic human Nature | Vol 585 | 24 September 2020 | 585

Article Animals below the limit of quantification (%)

0.5 0.4 0.3 0.2 0.1 0

100

50 25 0

3

6

10

10 8 6 4 2 0

e 0

0

Fig. 2 | Time course of lung lesions by CT analysis of SARS-CoV-2-infected cynomolgus macaques treated with HCQ. Lung lesions were assessed by chest CT before infection with SARS-CoV-2 and at 2, 5 and 11 or 13 d.p.i. a, Heat map of the overall CT score. Scores include lesion types (ground-glass opacity, crazy-paving pattern, consolidation or pleural thickening (scored from 0 to 3)) and lesion volume (scored from 0 to 4) summed for each lobe. Scores are consensus values from two independent evaluators. Dotted lines indicate treatment initiation. ‘X’ corresponds to missing data, macaques were not scanned at these time points. b, Representative images of lung lesions in two NHPs at baseline, 2, 5 and 13 d.p.i. Red arrows indicate typical lesions. Numbers at the bottom left of each image represent the CT score associated with the NHP and time point. Scores are the average over all scans made for the macaque at that time point.

exposure (Hi D1 + AZTH, n = 5). No effects of treatment were observed on either the viral RNA load in the different analysed tissues (Fig. 1d and Extended Data Fig. 2) or clinical scores. Clinical signs were comparable to control NHPs, with some NHPs exhibiting high CT scores in the Hi D1 + AZTH group (Fig. 2). In parallel, we also treated NHPs with a high dose of HCQ, starting 7 days before viral challenge as pre-exposure prophylaxis treatment (n = 5). Again, the kinetics of viral RNA loads were similar to those of the control group and no differences in the reduction in the AUC, peak viral load or time to first unquantifiable viral load were observed (Fig. 1 and Extended Data Fig. 2).

Relation between HCQ concentration and virus kinetics In the NHPs of Hi D1, Hi D1 + AZTH and pre-exposure prophylaxis groups, the plasma exposures were comparable to those observed in routine clinical practice 3–5 days after HCQ initiation using a dose of 200 mg three times daily (Fig. 3a). Drug trough concentrations were lower in both the Lo D1 and Lo D5 groups. When we assessed whether the higher drug exposure could generate more-rapid virus clearance, neither the time to attain the viral load limit of quantification nor the peak viral load were significantly associated with plasma HCQ concentrations (Fig. 3b–d). Finally, in an additional group of uninfected macaques, we characterized the HCQ pharmacokinetics in blood and plasma as the accumulation of HCQ in the lungs 6 days after the initiation of treatment 586 | Nature | Vol 585 | 24 September 2020

P > 0.1

9 8 7 6

55 50

20 15 10 5 0 0.1 0.2 0.3 0.4 0.5 HCQ plasma concentration (μg ml–1)

2

4 6 8 Time (d.p.i.)

10

PrEP P > 0.1

40 35 30 25

f

0

45

0 0.1 0.2 0.3 0.4 0.5 HCQ plasma concentration (μg ml–1)

25

0

d AUC viral load (log10[copies × day per ml])

8

Peak viral load (log10[copies per ml])

5

Hi D1

Lo D1

HCQ lung concentration (μg g–1)

Day 2

Control

c

P > 0.1

75

tro H l iD 1 H i D Lo 1 + D1 AZ TH Lo D5 Pr Pa EP tie nt s

0

b

C on

0

a HCQ plasma concentration (μg ml–1)

MF17 Lo D1

Baseline

MF2 Control

Day 5

MF1 MF2 MF3 MF4 MF5 MF6 MF7 MF8 MF9 MF10 MF11 MF12 MF13 MF14 MF15 MF16 MF17 MF18 MF19 MF20 MF21 MF22 MF27 MF28 MF29 MF30 MF31 MF23 MF24 MF25 MF26

b

Time (d.p.i.) 2 5 11/13

HCQ lung concentration (μg g–1)

Lo D5

PrEP

Hi D1 + AZTH

Lo D1

Hi D1

Control

0

Day 13

a

0 0.1 0.2 0.3 0.4 0.5 HCQ plasma concentration (μg ml–1)

25 20 15 10 5 0

0 0.5 1.0 1.5 2.0 2.5 HCQ blood concentration (μg ml–1)

Fig. 3 | Pharmacokinetic and viral kinetic parameters in cynomolgus macaques. a, Individual mean plasma trough concentrations of HCQ in NHPs during treatment. HCQ plasma trough concentrations determined within the context of routine therapeutic drug monitoring using the same method are provided for hospitalized patients (n = 25) who received 200 mg three times daily. The box bounds represent the 25th and 75th percentiles, the middle line shows the mean, and whiskers show the minimum and maximum values. b, Time to the first measurement below the limit of quantification in NHPs with a mean plasma trough concentration 0.1 μg ml−1 (grey). The time to viral suppression was compared between the two groups using a log-rank test (n = 31 macaques in total). c, Peak viral load according to mean HCQ plasma trough concentration. d, Area under the curve (AUC) or viral kinetic curve between 1 and 9 d.p.i. c, d, A Spearman correlation test was performed to assess the association between drug concentration and viral kinetic parameters (n = 31 macaques in total). e, HCQ lung and plasma concentrations in uninfected NHPs (n = 6). f, HCQ lung and blood concentrations in uninfected NHPs (n = 6).

(Fig. 3e, f and Extended Data Fig. 4). The blood concentrations in the high-dose HCQ group were higher than 1.4 μg ml−1, showing that the drug concentrations in the blood remained above the drug EC50 values that we identified in Vero E6 cells during in vitro efficacy assessment of HCQ against SARS-CoV-2 infection. The mean blood-to-plasma ratio was 6.8 (Extended Data Fig. 4), close to the value of 7.2, which was reported in healthy volunteers during various treatment intervals and durations26. Consistent with predictions made in physiological pharmacokinetic models, these levels of drug exposure in the plasma and blood produced higher exposure concentrations in lung tissues, with a lung-to-plasma ratio ranging from 27 to 177 (Fig. 3f), allowing lung tissues to achieve concentrations that were mostly above the drug EC50 values found in Vero E6 cells in all NHPs during the treatment period.

Eotaxin

Control

Hi D1

Lo D1 Hi D1 +AZTH Lo D5

MCP-1

IL-1RA *

IFNα

MF1 MF2 MF3 MF4 MF5 MF6 MF7 MF8 MF9 MF10 MF11 MF12 MF13 MF14 MF15 MF16 MF17 MF18 MF19 MF20 MF21 MF22 MF23 MF24 MF25 MF26

Time (d.p.i.) Concentration (pg ml–1) 0

IL-2

In conclusion, our evaluation of HCQ in the NHP model does not support its use as an antiviral agent for the treatment of COVID-19 in humans.

IL-15

Online content Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41586-020-2558-4. 1. 2. 3. 4. 0 2 5 7 9

200

0 2 5 7 9

400 0 200 400 600 0

0 2 5 7 9

40

80 0

0 2 5 7 9

2,000 4,000 0

0 2 5 7 9

30

60

0 2 5 7 9

0

20

40

Fig. 4 | Cytokines and chemokines in the plasma of SARS-CoV-2-infected cynomolgus macaques treated with HCQ. a, Heat map of plasma concentrations of eotaxin (also known as CCL-11), MCP-1 (also known as CCL-2), IFNα, IL-1RA, IL-2 and IL-15 at 0, 2, 5, 7 and 9 d.p.i. The asterisk indicates a significant difference in IL-1RA concentrations at 2 d.p.i. between the control group and the Hi D1 and Hi D1 + AZTH groups (P = 0.0287 and P = 0.0451, respectively). Further analyses of IL-1RA are shown in Extended Data Fig. 6c. Statistical significance was determined using a two-sided Mann–Whitney U-test without correction for multiple testing.

5. 6.

7. 8. 9. 10. 11. 12.

Pathogenesis and host response to HCQ treatment High alanine aminotransferase and creatinine kinase levels were observed in NHPs treated with the high HCQ and particularly the HCQ + AZTH regimen compared with control NHPs (Extended Data Figs. 7, 8). HCQ treatment did not prevent lymphocytopenia (Extended Data Fig. 5) nor pulmonary lesions, as shown by CT scan analysis (Fig. 2 and Extended Data Fig. 3). Similar lesion scores were observed in control and treated NHPs. All NHPs exhibited an increase in the concentrations of type-I IFNα, IL-1RA, CCL2 and CCL11 in plasma at 2 d.p.i. (Fig. 4 and Extended Data Fig. 6). In addition, IL-15 peaked early during infection, which suggests that innate lymphoid cells have a role in the control of initial viral replication in both drug-treated and untreated NHPs. When compared with control NHPs, TNF was significantly increased and IL-1RA was significantly reduced at 2 d.p.i. (Fig. 4 and Extended Data Fig. 6) in the groups that received the high dose of HCQ alone (P = 0.032 and P = 0.028, respectively) or with AZTH (P = 0.037 and P = 0.045, respectively).

Conclusions Our study shows that cynomolgus macaques are a relevant model for the analysis of the early stages of SARS-Cov-2 infection in hum ans1,3,18–20,27,28. We found no antiviral activity nor clinical efficacy of HCQ treatment, regardless of the timing of treatment initiation, either before infection, early after infection (before the peak of the viral load) or late after infection (after the peak of the viral load). This was in spite of high HCQ concentrations in the blood and lungs, and plasma exposures that were similar to those observed in patients with COVID-19 treated who were with HCQ. Thus, treatment with HCQ is unlikely to have antiviral activity in respiratory compartments. Our results illustrate the frequent discrepancy between results from in vitro assays and in vivo experiments, as reported for other viral infections such as influenza, dengue or chikungunya virus, for which clinical trials did not demonstrate efficacy of chloroquine or HCQ for the treatment of these infections4,29.

13.

14. 15.

16.

17. 18. 19. 20. 21.

22.

23. 24. 25. 26.

27. 28. 29.

He, X. et al. Temporal dynamics in viral shedding and transmissibility of COVID-19. Nat. Med. 26, 672–675 (2020). Huang, C. et al. Clinical features of patients infected with 2019 novel coronavirus in Wuhan, China. Lancet 395, 497–506 (2020). Chen, G. et al. Clinical and immunological features of severe and moderate coronavirus disease 2019. J. Clin. Invest. 130, 2620–2629 (2020). Touret, F. & de Lamballerie, X. Of chloroquine and COVID-19. Antiviral Res. 177, 104762 (2020). Touret, F. et al. In vitro screening of a FDA approved chemical library reveals potential inhibitors of SARS-CoV-2 replication. Sci. Rep. 10, 13093 (2020). Gautret, P. et al. Clinical and microbiological effect of a combination of hydroxychloroquine and azithromycin in 80 COVID-19 patients with at least a six-day follow up: a pilot observational study. Travel Med. Infect. Dis. 34, 101663 (2020). Magagnoli, J. et al. Outcomes of hydroxychloroquine usage in United States veterans hospitalized with Covid-19. Med 1, 1–14 (2020). Li, Q. et al. Early transmission dynamics in Wuhan, China, of novel coronavirus-infected pneumonia. N. Engl. J. Med. 382, 1199–1207 (2020). Wang, M. et al. Remdesivir and chloroquine effectively inhibit the recently emerged novel coronavirus (2019-nCoV) in vitro. Cell Res. 30, 269–271 (2020). Liu, J. et al. Hydroxychloroquine, a less toxic derivative of chloroquine, is effective in inhibiting SARS-CoV-2 infection in vitro. Cell Discov. 6, 16 (2020). Ponticelli, C. & Moroni, G. Hydroxychloroquine in systemic lupus erythematosus (SLE). Expert Opin. Drug Saf. 16, 411–419 (2017). Savarino, A., Di Trani, L., Donatelli, I., Cauda, R. & Cassone, A. New insights into the antiviral effects of chloroquine. Lancet Infect. Dis. 6, 67–69 (2006). Rainsford, K. D., Parke, A. L., Clifford-Rashotte, M. & Kean, W. F. Therapy and pharmacological properties of hydroxychloroquine and chloroquine in treatment of systemic lupus erythematosus, rheumatoid arthritis and related diseases. Inflammopharmacology 23, 231–269 (2015). Schrezenmeier, E. & Dörner, T. Mechanisms of action of hydroxychloroquine and chloroquine: implications for rheumatology. Nat. Rev. Rheumatol. 16, 155–166 (2020). Monzavi, S. M. et al. Efficacy analysis of hydroxychloroquine therapy in systemic lupus erythematosus: a study on disease activity and immunological biomarkers. Inflammopharmacology 26, 1175–1182 (2018). Fanouriakis, A. et al. 2019 Update of the Joint European League Against Rheumatism and European Renal Association-European Dialysis and Transplant Association (EULAR/ ERA-EDTA) recommendations for the management of lupus nephritis. Ann. Rheum. Dis. 79, 713–723 (2020). Zhang, B. et al. Macrolide derivatives reduce proinflammatory macrophage activation and macrophage-mediated neurotoxicity. CNS Neurosci. Ther. 25, 591–600 (2019). Rockx, B. et al. Comparative pathogenesis of COVID-19, MERS, and SARS in a nonhuman primate model. Science 368, 1012–1015 (2020). Williamson, B. N. et al. Clinical benefit of remdesivir in rhesus macaques infected with SARS-CoV-2. Nature https://doi.org/10.1038/s41586-020-2423-5 (2020). Shi, R. et al. A human neutralizing antibody targets the receptor-binding site of SARS-CoV-2. Nature 584, 120–124 (2020). Yao, X. et al. In vitro antiviral activity and projection of optimized dosing design of hydroxychloroquine for the treatment of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). Clin. Infect. Dis. 71, 732–739 (2020). Nicolas de Lamballerie, C. et al. Characterization of cellular transcriptomic signatures induced by different respiratory viruses in human reconstituted airway epithelia. Sci. Rep. 9, 11493 (2019). Pizzorno, A. et al. Characterization and treatment of SARS-CoV-2 in nasal and bronchial human airway epithelia. Cell. Rep. Med 1, 100059 (2020). Pan, F. et al. Time course of lung changes at chest CT during recovery from coronavirus disease 2019 (COVID-19). Radiology 295, 715–721 (2020). Shi, H. et al. Radiological findings from 81 patients with COVID-19 pneumonia in Wuhan, China: a descriptive study. Lancet Infect. Dis. 20, 425–434 (2020). Tett, S. E., Cutler, D. J., Day, R. O. & Brown, K. F. A dose-ranging study of the pharmacokinetics of hydroxy-chloroquine following intravenous administration to healthy volunteers. Br. J. Clin. Pharmacol. 26, 303–313 (1988). Liu, Y. et al. Viral dynamics in mild and severe cases of COVID-19. Lancet Infect. Dis. 20, 656–657 (2020). Chen, N. et al. Epidemiological and clinical characteristics of 99 cases of 2019 novel coronavirus pneumonia in Wuhan, China: a descriptive study. Lancet 395, 507–513 (2020). Roques, P. et al. Paradoxical effect of chloroquine treatment in enhancing chikungunya virus infection. Viruses 10, 268 (2018).

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. © The Author(s), under exclusive licence to Springer Nature Limited 2020

Nature | Vol 585 | 24 September 2020 | 587

Article Methods Ethics and biosafety statement Cynomolgus macaques (M. fascicularis), aged 37–40 months and originating from Mauritian AAALAC-certified breeding centres, were used in this study. All macaques were housed in IDMIT infrastructure facilities (CEA, Fontenay-aux-Roses), under BSL-2 and BSL-3 containment when necessary (animal facility authorization D92-032-02, Prefecture des Hauts de Seine, France) and in compliance with European Directive 2010/63/EU, the French regulations and the Standards for Human Care and Use of Laboratory Animals of the Office for Laboratory Animal Welfare (OLAW, assurance number A5826-01, United States). The protocols were approved by the institutional ethical committee ‘Comité d’Ethique en Expérimentation Animale du Commissariat à l’Energie Atomique et aux Energies Alternatives’ (CEtEA 44) under statement number A20-011. The study was authorized by the ‘Research, Innovation and Education Ministry’ under registration number APAFIS#244342020030216532863v1. HCQ and AZTH Hydroxychloroquine sulfate (HCQ) was manufactured for Sanofi by the Chinoin Pharmaceutical and Chemical Works under good manufacturing practice conditions and provided as the base powder. Batch number DU017 was solubilized extemporaneously in water at 5, 10 or 15 mg ml−1 depending on the group and the dose. Tablets of AZTH (250 mg) (Sandoz, batch number KH5525) were crushed and suspended extemporaneously at 12 mg ml−1 of AZTH in water. Macaques and study design To evaluate the efficacy of HCQ and HCQ + AZTH treatments, the macaques were randomly assigned in sex-balanced experimental groups. No statistical methods were used to predetermine sample size. Challenged macaques were exposed to a total dose of 106 PFU of SARS-CoV-2 through a combination of intranasal and intratracheal routes (day 0), using atropine (0.04 mg kg−1) as premedication and ketamine (5 mg kg−1) with medetomidine (0.042 mg kg−1) as anaesthesia. The regimen comprising a high dose of HCQ in group ‘Hi D1’ (n = 5) consisted of a loading dose of 90 mg kg−1 at 1 d.p.i. and a daily maintenance dose of 45 mg kg−1, for a total of 10 days. The ‘Hi D1 + AZTH’ regimen (n = 5) consisted of the same HCQ regimen as for the Hi D1 group combined with one loading dose of 36 mg kg−1 of AZTH at 1 d.p.i., followed by a daily maintenance dose of 18 mg kg−1 AZTH for 10 days. The low-dose (Lo) regimen consisted of a HCQ loading dose of 30 mg kg−1 and a daily maintenance dose of 15 mg kg−1 for 12 days. The low-dose treatment of the ‘Lo D1’ group (n = 4) was initiated at 1 d.p.i. and the low-dose treatment of the ‘Lo D5’ group (n = 4) was initiated at 5 d.p.i. The PrEP regimen (n = 5) consisted of a loading dose of 30 mg kg−1 HCQ 7 days before challenge, followed by a daily dose of 15 mg kg−1 for 4 days and 45 mg kg−1 for 3 days before virus challenge and then 45 mg kg−1 until 6 d.p.i. Treatments were delivered by gavage. Placebo-treated macaques received water, which was the vehicle for HCQ. Macaques were observed daily and clinical examinations were performed at baseline, daily for one week and then twice weekly on macaques that were anaesthetized using ketamine (5 mg kg−1) and metedomidine (0.042 mg kg−1). Body weight, rectal temperature, respiration, heart rates and oxygen saturation were recorded and blood, as well as nasopharyngeal, tracheal and rectal swabs, were collected. Bronchoalveolar lavages were performed using 50 ml sterile saline on 6, 14, 21 and 28 d.p.i. Chest CT scans were performed at baseline and on 2, 5 and 11 or 13 d.p.i. in macaques that were anaesthetized using tiletamine (4 mg kg−1) and zolazepam (4 mg kg−1). Blood cell counts, haemoglobin and haematocrit were determined from EDTA-treated blood samples using a HMX A/L analyser (Beckman Coulter). Biochemistry parameters including alanine aminotransferase (ALT), aspartate aminotransferase (AST), albumin, haptoglobin, creatinine, creatine

kinase, lactate dehydrogenase (LDH) and total protein, were analysed with standard kits (Siemens) and C-reactive protein with a canine kit (Randox) in lithium heparin plasma, inactivated with Triton X-100, using an ADVIA1800 analyser (Siemens). The pharmacokinetics of HCQ was assessed using the same administration procedure in six uninfected macaques, randomly assigned as pairs into three experimental groups as described in Extended Data Fig. 4. The pharmacokinetic low (PK Lo) group received a low loading dose (30 mg kg−1) at day 0 and a low daily maintenance dose (15 mg kg−1) for 5 days. The pharmacokinetic high (PK Hi) and ‘PK Hi + AZTH’ groups received a high loading HCQ dose (90 mg kg−1) on day 0 and a high daily maintenance dose (45 mg kg−1) for 6 days, without or with AZTH (loading dose of 36 mg kg−1 and maintenance of 18 mg kg−1), respectively. Blood samples were taken at 0, 2, 4 and 6 h after treatment on day 0, and before treatment on the following days. For the PK Hi and PK Hi + AZTH groups, blood samples were also collected at 0, 2, 4 and 6 h after treatment after treatment on day 5. Macaques were humanly euthanized 24 h after the administration of the last dose using 18.2 mg kg−1 of pentobarbital sodium intravenously under tiletamine (4 mg kg−1) and zolazepam (4 mg kg−1) anaesthesia. Samples of lung were collected at necropsy for HCQ quantification.

Determination of HCQ concentrations Quantification of HCQ in plasma, blood and lung tissues was performed by a sensitive and selective validated high-performance liquid chromatography coupled with tandem mass spectrometry method (Quattro Premier XE LC-MS/MS, Waters) as previously described30, with lower limits of quantification of 0.015 μg ml−1 for plasma and 0.05 μg ml−1 for blood and lung tissue. Blood samples were centrifuged within 1 h to collect plasma samples. Lung biopsies collected after euthanasia were thoroughly rinsed with cold 0.9% NaCl to remove blood contamination and blotted with filter paper. Then, each lung biopsy was weighed and homogenized with 1 ml of 0.9% NaCl using a Mixer mill MM200 (Retsch). Cellular debris was removed by centrifugation, and the supernatant was stored at −80 °C. HCQ was extracted by a simple protein precipitation method, using methanol for plasma and ice-cold acetonitrile for blood and tissue homogenates. In brief, 100 μl of sample matrix was spiked with 10 μl of internal standard working solution (HCQ-d5, Alsachim), vortexed for 2 min followed by centrifugation for 10 min at 4 °C. The supernatant was evaporated for blood- and tissue-homogenate samples. Dry residues or plasma supernatants were then transferred to 96-well plates and 5 μl was injected. To assess the selectivity and specificity of the method and matrix effect, blank plasma, blood and tissues from control macaques were processed and compared with that of HCQ and index-selectivity-spiked plasma, blood or tissue homogenate samples. Furthermore, each baseline sample (H0) of treated macaques was processed in duplicate, including one spiked with HCQ prepared equivalent to quality control samples. Concentrations in blood (μg ml−1), plasma (μg ml−1) and lung (μg g−1) were determined for each uninfected macaque and in plasma only for infected macaques. Drug accumulation in the lung was assessed by calculating a lung-to-blood and a lung-to-plasma concentration ratio. No signs of haemolysis were observed, either visually (when only plasma samples were available) or after verification of the consistency between the two matrixes (when both plasma and blood samples were available). HCQ plasma trough concentrations determined within the context of routine therapeutic drug monitoring using the same method, 3–5 days after initiation of HCQ at 200 mg three times daily were provided for comparison. Viruses and cells For the in  vivo studies, SARS-CoV-2 virus (hCoV-19/France/ lDF0372/2020 strain) was isolated by the National Reference Center

for Respiratory Viruses (Institut Pasteur) as previously described31. Virus stocks used in  vivo were produced by two passages on mycoplasma-free Vero  E6 cells in Dulbecco’s modified Eagle’s medium (DMEM) without FBS, supplemented with 1% penicillin (10,000 U ml−1) and streptomycin (10,000 μg ml−1) and 1 μg ml−1 TPCK-trypsin at 37 °C in a humidified CO2 incubator and titrated on Vero E6 cells. For the in vitro studies, the viral strain hCoV-19/France/IDF0571/2020 was provided by X. Lescure and Y. Yazdanpanah from the Bichat Hospital, where the isolate was obtained from a patient returning from Jichang (China) and passaged three times. For the virus used in the in vivo experiments, whole-genome sequencing was performed as previously described31 with no modifications observed compared with the initial specimen27. For sequencing of the virus used in vitro, viral RNA extraction was done using the QiAmp viral RNA Kit (Qiagen). The complete viral genome sequence was obtained using Illumina MiSeq sequencing technology. Sequences were deposited after assembly on the GISAID EpiCoV platform under accession numbers EPI_ISL_406596 for hCoV-19/France/lDF0372/2020 and EPI_ISL_411218 for hCoV-19/ France/IDF0571/2020.

Viral replication kinetics and antiviral treatment in Vero E6 cells Vero E6 cells were seeded 24 h in advance in multi-well 6 plates, washed twice with PBS and then infected with SARS-CoV-2 at the indicated multiplicities of infection (MOI). For HCQ treatment, the inoculum of infected Vero E6 cells was removed 1 h after infection (h.p.i.) and cells were immediately treated with solutions in DMEM of HCQ. Supernatants were collected at 48 and 72 h.p.i. and stored at −80 °C for RNA extraction and viral quantification. Viral quantification in Vero E6 cells Viral stocks and collected samples were titrated by tissue-culture infectious dose 50% (TCID50 ml−1) in Vero E6 cells, using the Reed and Muench statistical method. Relative quantification of the viral genome was performed by one-step real-time quantitative reverse transcriptase and polymerase chain reaction (RT–qPCR) from viral RNA extracted using the QiAmp viral RNA Kit (Qiagen) in the case of supernatants or apical washings. Primer and probe sequences were selected from those designed by the School of Public Health/University of Hong Kong (L. Poon, D. Chu and M. Peiris) and synthetized by Eurogentec23. Real-time one-step RT–qPCR was performed using the EXPRESS One-Step Superscript qRT–PCR Kit (Invitrogen, 1178101K). Thermal cycling was performed in a StepOnePlus Real-Time PCR System (Applied Biosystems) in MicroAmp Fast Optical 96-well reaction plates (Applied Biosystems, 4346907), as previously described23. Viral infection and treatment in reconstituted human airway epithelia MucilAir human airway epithelia (HAE) reconstituted from human primary cells obtained from nasal (pool of donors) or bronchial (single donors) biopsies were provided by Epithelix and maintained in air–liquid interphase with specific culture medium in Costar Transwell inserts (Corning) according to the manufacturer’s instructions. For infection experiments, apical poles were gently washed twice with warm OptiMEM medium (Gibco, ThermoFisher Scientific) and then infected directly with a 150-μl dilution of virus in OptiMEM medium, at a MOI of 0.1. For mock infection, the same procedure was performed using OptiMEM as inoculum. Samples collected from apical washes or basolateral medium at different time points were separated into two tubes: one for TCID50 viral titration and one RT–qPCR. HAE cells were collected in RLT buffer (Qiagen) and total RNA was extracted using the RNeasy Mini Kit (Qiagen) for subsequent RT–qPCR and Nanostring assays. Treatments with HCQ were applied through basolateral poles. All treatments were initiated on day 0 (1 h after viral infection) and

continued once daily. Samples were collected at 48 h.p.i. Variations in trans-epithelial electrical resistance (ΔTEER) were measured using a dedicated volt–ohm meter (EVOM2, Epithelial Volt/Ohm Meter for TEER) and expressed as Ω cm−2.

Virus quantification in NHP samples Upper respiratory (nasopharyngeal and tracheal) and rectal specimens were collected with swabs (Universal transport medium, Copan; or Viral Transport Medium, CDC, DSR-052-01). Tracheal swabs were performed by insertion of the swab above the tip of the epiglottis into the upper trachea at approximately 1.5 cm of the epiglottis. All specimens were stored between 2 °C and 8 °C until analysis with a plasmid standard concentration range containing an rdrp gene fragment including the RdRp-IP4 RT–PCR target sequence. The protocol describing the procedure for the detection of SARS-CoV-2 is available on the WHO website (https://www.who.int/docs/default-source/ coronaviruse/real-time-rt-pcr-assays-for-the-detection-of-sars-cov -2-institut-pasteur-paris.pdf?sfvrsn=3662fcb6_2). Plasma cytokine analysis Cytokines were quantified in EDTA-treated plasma using NHP ProcartaPlex immunoassay (ThermoFisher Scientific) for IFNα, IL-1RA, IL-1β, CCL-2 (also known as MCP-1), CCL-11 (also known as eotaxin), CXCL-11 (also known as ITAC), CXCL-1 (also known as BLC), granzyme B and PDGF-BB, using NHP Milliplex (Millipore) for CD40L, G-CSF, GM-CSF, IFNγ, IL-2, IL-4, IL-5, IL-6, CXCL-8 (also known as IL-8), IL-10, IL-13, IL-15, IL-17A, CCL-3 (also known as MIP-1α), CCL-4 (also known as MIP-1β), TNF, VEGF and a Bioplex 200 analyser (Bio-Rad) according to manufacturer’s instructions. Chest CT and image analysis Acquisition was done using a CT system (Vereos-Ingenuity, Philips) in BSL-3 containment facilities on anaesthetized macaques placed in a supine position and monitored for heart rate, oxygen saturation and body temperature. An intravenous bolus of iodine contrast agent (Vizipaque, 320mg ml−1, GE Heathcare, 3 ml kg−1) was injected (Medrad CT Stellant injector, Bayer) in the saphenous vein 20 s before the initiation of CT scan acquisition. The CT detector collimation was 64 × 0.6 mm, the tube voltage was 120 kV and intensity of about 120 mA. Automatic dose optimization tools (Dose Right, Z-DOM, 3D-DOM; Philips Healthcare) regulated the intensity. CT images were reconstructed with a slice thickness of 1.25 mm and an interval of 0.25 mm. Images were analysed using INTELLISPACE PORTAL 8 software (Philips healthcare). All images had the same window level of −300 and window width of 1,600. Lesions were defined as ground glass opacitiy, crazy-paving pattern, consolidation or pleural thickening as previously described24,25. Lesions and scoring were assessed independently in each lung lobe by two individuals, and the final results were made by consensus. The overall CT score includes lesion type (scored from 0 to 3) and lesion volume (scored from 0 to 4) summed for each lobe as described in Extended Data Fig. 3. Statistical analysis The following viral kinetic parameters were calculated in each experimental group as medians (and minimum–maximum): viral load peak, area under the curve of the log10 viral load, time to first unquantifiable viral load. Each viral kinetic parameter was compared with untreated macaques using Wilcoxon rank-sum or log-rank tests (Microsoft Excel 2016, GraphPad Prism version 7). To evaluate a potential effect of drug exposure on viral dynamics, we further evaluated the correlation of the viral kinetic parameters with the plasma concentrations of HCQ, taking the mean trough concentrations observed in each infected macaque between 1 and up to 15 days after treatment as a marker of drug exposure during treatment period (Spearman test, without adjusting for multiple testing).

Article Ethics committee All information on the ethics committee is available at https://cache. media.enseignementsup-recherche.gouv.fr/file/utilisation_des_animaux_fins_scientifiques/22/1/comiteethiqueea17_juin2013_257221.pdf. Reporting summary Further information on research design is available in the Nature Research Reporting Summary linked to this paper.

Data availability The data that support the findings of this study are included in the paper and Supplementary Information. 30. Chhonker, Y. S., Sleightholm, R. L., Li, J., Oupický, D. & Murry, D. J. Simultaneous quantitation of hydroxychloroquine and its metabolites in mouse blood and tissues using LC-ESI-MS/MS: an application for pharmacokinetic studies. J. Chromatogr. B Analyt. Technol. Biomed. Life Sci. 1072, 320–327 (2018). 31. Lescure, F. X. et al. Clinical and virological data of the first cases of COVID-19 in Europe: a case series. Lancet Infect. Dis. 20, 697–706 (2020). Acknowledgements We thank B. Delache, E. Burban, J. Demilly, N. Dhooge, S. Langlois, P. Le Calvez, M. Potier, F. Relouzat, J. M. Robert and C. Dodan for help with animal studies; B. Fert and C. Mayet for help with in vivo imaging studies; Q. Pascal for performing the necropsies; J. Morin for help with the cytokine measurements and preparation of reagents; K. Barthelemy, M. Basso, N. Doudka and M. Giocanti for help with HCQ concentration measurements; B. Lacarelle and R. Guilhaumou for help with analysing internal drug concentration data; J. Bertrand for help with analysing pharmacokinetic data; D. Desjardins for help with the AZTH pharmacokinetic study; C. Aubenque, M. Barendji, L. Bossevot, N. Dimant, J. Dinh, A. S. Gallouet, M. Leonec, I. Mangeot and K. Storck for help with sample processing; M. Albert, M. Barbet and F. Donati for help with the production, titration and sequencing of the virus stocks used in vivo and processing of samples for RT–PCR; A. S. Gallouet, S. Keyser, E. Marcos-Lopez, B. Targat and B. Vaslin for help with the experimental studies in the context of COVID-19-induced constraints; F. Ducancel and Y. Gorin for help with the logistics and safety management; E. Sultan for providing guidance on HCQ dose selection and discussion on pharmacokinetics/pharmacodynamics results and commenting on the paper; Sanofi for providing the HCQ batch used in these experiments; the Fondation Bettencourt Schueller and the Region Ile-de-France for the contribution to the implementation of imaging facilities; and the Domaine d’Intérêt Majeur (DIM) ‘One Health’ for its support. This study received financial support from REACTing, the National Research Agency (ANR; AM-CoV-Path) and the European

Union’s Horizon 2020 (H2020) research and innovation program Fight-nCov (101003555), European Union IMI2 program CARE (101005077) and the European Infrastructure TRANSVAC2 (730964). The virus stock was obtained through the EVAg platform (https://www. european-virus-archive.com/), funded by H2020 (653316). The Infectious Disease Models and Innovative Therapies (IDMIT) research infrastructure is supported by the ‘Programme Investissements d’Avenir’, managed by the ANR under reference ANR-11-INBS-0008. Author contributions A.B. performed RT–PCR viral quantification and analysed the data. A.G. contributed to statistical analysis. A.P. performed in vitro evaluation of HCQ (Vero E6 and HAE), and contributed to data analysis and the preparation of the manuscript. B.H. contributed to study design, data analysis and the writing of the paper. B.L. coordinated the in vitro evaluation of HCQ (Vero E6 and HAE), analysed the data and contributed to the writing of the paper. C.C. coordinated the imaging facility. J.G. contributed to data analysis, the pharmacokinetics/ pharmacodynamics study and the writing of the paper. J.L. contributed to clinical follow-up of macaques, data analysis and the writing of the paper. M.R.C. designed the in vitro evaluation of HCQ (Vero E6 and HAE), supervised and coordinated the work, analysed the data and contributed to the writing of the paper. N.D.-B. contributed to the animal work and cytokine measurements, analysed the data and coordinated IDMIT core activities. N.E. developed the RT–qPCR assay and analysed the data. N.K. performed CT scans and acquisition parameter design, and contributed to data analysis. O.T. performed in vitro evaluation of HCQ (Vero E6 and HAE), contributed to data analysis and manuscript preparation. P.M. contributed to project conception and design of the study, contributed to animal work, the coordination of the experiments, data analysis and the writing of the paper. R.H.T.F. coordinated the animal core facility, and contributed to study design and data analysis. R.L.G. conceived the project, designed the study, coordinated the work, analysed the data and wrote the article. R.M. contributed to the design of the study, animal work, data analysis and contributed the writing of the paper. S.B. performed RT–qPCR viral quantification and analysed the data. C.S. supervised and coordinated the HCQ pharmacokinetics analysis, provided clinical data on plasma HCQ levels and contributed to the writing of the paper. F.T. contributed to in vitro antiviral evaluation. S.v.d.W. conceived the project, designed the study, provided the viral challenge stock, coordinated the viral load quantification, analysed the data and wrote the paper. T.N. performed CT scans and quantification, contributed to the quantification design, generated CT figures and wrote the paper. V.C. contributed to data analysis, statistical analyses, figures design and the writing of the paper. V.E. developed the RT–qPCR viral quantification assay. X.d.L. contributed to study design, pharmacokinetics/ pharmacodynamics analysis and the writing of the paper. Competing interests J.G. has worked as consultant for Roche. Additional information Supplementary information is available for this paper at https://doi.org/10.1038/s41586-0202558-4. Correspondence and requests for materials should be addressed to R.L.G. Peer review information Nature thanks Robin Ferner and Debby van Riel for their contribution to the peer review of this work. Reprints and permissions information is available at http://www.nature.com/reprints.

a

MOI 0.01 - 48 hpi

100

50

0

0.1

100

50

0

1 10 HCQ concentration (µM)

b

0.1

IC50 (µM)

CC50 (µM)

SI

MOI 0.01 48 hpi

2.189

215

98.218

MOI 0.01 72 hpi

4.394

161

36.640

1 10 HCQ concentration (µM)

Treatment post-infection 2

1

Nasal Epithelium MOI 0.1 48hpi

0 600

TEER (Ohmms/cm2)

HCQ

Nasal Bronchial

3 Apical relative viral production

MOI 0.01 - 72 hpi

150 Relative viral production

Relative viral production

150

400 Bronchial Epithelium MOI 0.1 48hpi

200

Apical Relative viral production

Infectious titer (log10 TCID50/mL)

Untreated

1 +/- 0.53

6.8

HCQ - 1 µM

1.92 +/- 0.46

7.3

HCQ - 10 µM

1.70 +/- 0.03

6.97

Untreated

1 +/- 0.09

7.63

HCQ - 1 µM

1.90 +/- 0.11

7.3

HCQ - 10 µM

1.77 +/-0.85

7.63

10 µM -H CQ

1µ M In fe cte d

-H CQ

In fe cte d

-U nt re at ed

In fe cte d

M O CK

-U nt re at ed

0

Extended Data Fig. 1 | In vitro evaluation of the antiviral activity of HCQ against SARS-CoV-2. a, Dose–response curves of HCQ at 48 and 72 h.p.i. in Vero E6 cells. Vero E6 cells were seeded 24 h in advance in multi-well six-well plates, washed twice with PBS and then infected with SARS-CoV-2 (BetaCoV/ France/IDF0571/2020 SARS-CoV-2 strain) at the indicated MOI. The inoculum of infected Vero E6 cells was removed 1 h.p.i. and cells were immediately treated with different concentrations of HCQ. Supernatants were collected at 48 and 72 h.p.i. and stored at −80 °C for RNA extraction and viral titration by RT–qPCR. Results were expressed in relative viral production compared with the untreated control. The table summarizes the IC50, cytotoxic concentration 50% (CC50) and selectivity index (SI) for each condition. b, Apical relative viral production and trans-epithelial resistance (TEER in Ohms cm−2) between the apical and basal poles in nasal and bronchial HAE at 48 h.p.i. MucilAir HAE

reconstituted from human primary cells obtained from nasal or bronchial biopsies were provided by Epithelix and maintained in air–liquid interphase. For infection experiments, apical poles were gently washed twice with warm OptiMEM medium (Gibco, ThermoFisher Scientific) and then infected directly with nasal swab samples or a 150-μl dilution of virus in OptiMEM medium, at a MOI of 0.1. For mock infection, the same procedure was performed using OptiMEM as inoculum. Samples collected from apical washes were separated into two tubes: one for TCID50 viral titration and one for RT–qPCR. Results are expressed in relative viral production compared with the infected untreated control. The table summarizes the relative viral production values (mean ± s.d.) and the infectious titres (log10[TCID50 ml−1) of three biological replicates tested in duplicate.

Article a CTRL MF1 MF2 MF3 MF4

MF5 MF6 MF7 MF8

10 9 8 7 6 5 4 3 -8 -4 0

MF9 MF10 MF11 MF12 MF13

MF14 MF15 MF16 MF17

MF18 MF19 MF20 MF21 MF22

4

8 12 16 20 24 28

4

4

Viral load (Log10 copies / mL)

4

4

4

8 12 16 20 24 28

4

10 9 8 7 6 5 4 3

4

4

8 12 16 20 24 28

8 12 16 20 24 28

4

8 12 16 20 24 28

4

8 12 16 20 24 28

4

8 12 16 20 24 28

8 12 16 20 24 28

-8 -4 0

4

8 12 16 20 24 28

-8 -4 0

4

8 12 16 20 24 28

-8 -4 0

4

8 12 16 20 24 28

-8 -4 0

4

8 12 16 20 24 28

-8 -4 0

4

8 12 16 20 24 28

10 9 8 7 6 5 4 3 -8 -4 0

4

8 12 16 20 24 28

10 9 8 7 6 5 4 3 -8 -4 0

4

10 9 8 7 6 5 4 3 -8 -4 0

8 12 16 20 24 28

10 9 8 7 6 5 4 3 -8 -4 0

4

10 9 8 7 6 5 4 3 -8 -4 0

-8 -4 0

10 9 8 7 6 5 4 3 -8 -4 0

8 12 16 20 24 28

BAL

10 9 8 7 6 5 4 3

8 12 16 20 24 28

10 9 8 7 6 5 4 3 -8 -4 0

8 12 16 20 24 28

4

d

10 9 8 7 6 5 4 3 -8 -4 0

8 12 16 20 24 28

10 9 8 7 6 5 4 3 -8 -4 0

-8 -4 0

10 9 8 7 6 5 4 3 -8 -4 0

8 12 16 20 24 28

Rectal

10 9 8 7 6 5 4 3

8 12 16 20 24 28

10 9 8 7 6 5 4 3

10 9 8 7 6 5 4 3

MF23 MF24 MF25 MF26

4

c

10 9 8 7 6 5 4 3 -8 -4 0

8 12 16 20 24 28

10 9 8 7 6 5 4 3 -8 -4 0

Lo D5

-8 -4 0

Nasopharyngeal

10 9 8 7 6 5 4 3 -8 -4 0

Hi D1 + AZTH

10 9 8 7 6 5 4 3

8 12 16 20 24 28

10 9 8 7 6 5 4 3

Lo D1

b

10 9 8 7 6 5 4 3 -8 -4 0

MF27 MF28 MF29 MF30 MF31

4

10 9 8 7 6 5 4 3

Hi D1

PrEP

Tracheal

10 9 8 7 6 5 4 3 -8 -4 0

4

8 12 16 20 24 28

Days post infection Extended Data Fig. 2 | Viral loads of SARS-CoV-2-infected cynomolgus macaques treated with HCQ. a–d, Viral loads measured by RT–qPCR in throat swabs (a), nasal swabs (b), rectal swabs (c) and bronchoalveolar lavages (d). The limit of detection was estimated at 2.3 log10 copies of SARS-CoV-2 RNA per ml and the limit of quantification was estimated at 3.9 log10 copies per ml (dotted

horizontal line). Shaded zones indicate treatment periods. Baseline was adjusted to day 0 on the graphs. CTRL, control; D1, treatment started on day 1; D5, treatment started on day 5; Hi, high HCQ dose; Lo, low HCQ dose; PrEP, pre-exposure prophylaxis.

a

d

CTRL Day 2

Day 5

Baseline

Day 2

Baseline

Day 2

Day 5

Day 11/13

Day 5

Day 11/13

MF19

MF1 MF2

MF20

MF3

MF21

MF4

MF22

MF5

e

b

MF23

ND

MF24

ND

Hi D1 Day 2

Day 5

Day 11/13

MF10

MF26

MF9

Baseline

Lo D5

MF25

MF7

ND

MF8

MF6

Hi D1 + AZTH

Day 11/13 MF18

Baseline

f

PrEP Day 2

Day 5

MF29

MF13

MF28

MF12

MF27

MF11

Baseline

c

Day 5

Day 11/13

MF17

MF16

MF15

MF31

MF14

Day 2

MF30

Lo D1 Baseline

g

h

Lesion type None GGO Pleural thickenning Crazy-paving pattern Consolidation

Operator Date

Time Animal point

Score 0 1 1 3 3

Localisation

Lesion extension None < 25% of the lobe 25 < x < 50% of the lobe 50 < x < 75% of the lobe 75 < x < 100% of the lobe

GGO

Pleural thickenning

Score 0 1 2 3 4

Pleural effusion None Slight Mild Severe

Lesion Crazy-paving Consolidation extension pattern

Pleural effusion

Score 0 1 2 3

Total score

Cranial right lobe Middle right lobe Caudal right lobe Cranial left lobe culmen Cranial left lobe lingula Caudal left lobe Accessory lobe Pleural space Total

Extended Data Fig. 3 | Representative transversal slices of lung CT scans from SARS-CoV-2-infected cynomolgus macaques treated with HCQ. Imaging was performed at baseline and day 2, 5 and 11 or 13 post-exposure to SARS-CoV-2. a–f, Images are presented for each macaque according to their

group, with a window level of −300 and a window width of 1,600. g, h, CT scoring scales (g) and example of the analysis worksheet (h). Total CT score is the sum of all lung lobe scores including lesion type, extension contributions and pleural effusion severity. ND, not determined.

Article b

a

0

1

2

3

4

5

6

PK Lo (n=2)

PK Hi (n=2)

PK Hi + AZTH

HCQ blood conc. (µg/mL)

9.0

HCQ plasma conc. (µg/mL)

(n=2)

c

R=0.84, p < 0.001

7.5 6.0 4.5 3.0 1.5 0.0 0.0

0.5

1.0

1.5

2.0

2.5

Plasma 10

3.0

HCQ plasma conc. (µg/mL) Extended Data Fig. 4 | Plasma and blood HCQ concentrations of six uninfected NHPs. a, Pharmacokinetic (PK) study design. Three groups of two cynomolgus macaques received either a high (Hi) dose regimen of HCQ composed of a loading dose of 90 mg kg−1 and a daily maintenance dose of 45 mg kg−1 or a low (Lo) dose regimen composed of a 30 mg kg−1 loading dose

Total blood 10

1

1

0.1

0.1

0.01

0.01

0.001

0.001

0

24 48 72 96 120 144

10

0

24 48 72 96 120 144

0

24 48 72 96 120 144

0

24 48 72 96 120 144

10

1

1

0.1

0.1

0.01

0.01

0.001

0.001

0

24 48 72 96 120 144

10

10

1

1

0.1

0.1

0.01

0.01

0.001

0.001

0

24 48 72 96 120 144

Time since treatment initiation (hours) and a daily maintenance dose of 15 mg kg−1. One group received AZTH in combination with HCQ with a loading dose of 36 mg kg−1 followed by an 18 mg kg−1 daily maintenance dose. b, Individual plasma and blood HCQ concentrations (conc.) up to 5–6 days after the initiation of treatment. c, Correlation between plasma and blood HCQ concentrations.

a

WBC

MF5

MF2

MF6

MF3

MF7

MF4

MF8

10

5 -8 -4 0

4

8 12 16 20 24 28

20

Hi D1

MF11 MF13

0

MF16

0

-8 -4 0

4

8 12 16 20 24 28

0

-8 -4 0

4

8 12 16 20 24 28

MF18 MF20 MF22

0

0

-8 -4 0

4

8 12 16 20 24 28

10

4

8 12 16 20 24 28

MF25

-8 -4 0

4

8 12 16 20 24 28

0

x 103 cells / µL

8 12 16 20 24 28

-8 -4 0

4

8 12 16 20 24 28

0

0

-8 -4 0

4

8 12 16 20 24 28

8 12 16 20 24 28

0.0

-8 -4 0

4

8 12 16 20 24 28

-8 -4 0

4

8 12 16 20 24 28

0

-8 -4 0

4

8 12 16 20 24 28

-8 -4 0

4

8 12 16 20 24 28

0.5 -8 -4 0

4

8 12 16 20 24 28

0.0 4.0 3.0 1.5

15

1.0 5

5 8 12 16 20 24 28

4

4.0 3.0 1.5

10

4

-8 -4 0

0.5

5

-8 -4 0

0.0

1.0

10

0

8 12 16 20 24 28

4.0 3.0 1.5

5

10

0

4

10

15

MF27

-8 -4 0

5

20

PrEP

0

15

5

MF26

4

1.0

10

10

MF24

-8 -4 0

0.5

5

-8 -4 0

0.0 4.0 3.0 1.5

15

15

MF23

8 12 16 20 24 28

10

20

Lo D5

4

5

0

8 12 16 20 24 28

1.0

5

MF21

-8 -4 0

15

15 10

MF19

0

5

8 12 16 20 24 28

4

0.5

10

4

-8 -4 0

4.0 3.0 1.5

5

-8 -4 0

0.0

1.0

10

20

Hi D1 + AZTH

8 12 16 20 24 28

5

5

MF17

4

10

10

MF15

-8 -4 0

15

15

MF14

0.5

0

5

20

Lo D1

MF31

8 12 16 20 24 28

5

MF12

MF30

4

10

10

MF10

MF29

-8 -4 0

15

MF9

Monocytes

1.0 5

0

d 4.0 3.0 1.5

5

10

0

Neutrophils

15

15

MF1

MF28

c

Lymphocytes

10

20

CTRL

b

-8 -4 0

4

8 12 16 20 24 28

0

0.5 -8 -4 0

4

8 12 16 20 24 28

0.0

Days post infection Extended Data Fig. 5 | Complete blood count of SARS-CoV-2-infected cynomolgus macaques treated with HCQ. a–d, Absolute numbers of white blood cells (WBC) (a), lymphocytes (b), neutrophils (c) and monocytes (d) in

SARS-CoV-2-infected macaques. Baseline was adjusted to day 0 on the graphs, and to the first day of treatment (day 7 pre-exposure) for the PrEP group. Shaded zones indicate treatment periods.

Article

Extended Data Fig. 6 | Cytokines and chemokines in the plasma of SARS-CoV-2-exposed cynomolgus macaques treated with HCQ. a, b, Heat maps of the plasmatic concentrations of 20 cytokines. Each column represents one cytokine or chemokine; the colour scale (in pg ml−1) is shown at the bottom. The asterisk indicates a significant difference in the concentration of TNF at 2 d.p.i. between the control group and the Hi D1 and/or Hi D1 + AZTH groups, as

shown in d. ‘X’ indicates that no measurement was determined for this time point. c, d, Concentrations of IL-1RA and TNF at 2 d.p.i. Each plot represents one macaque. Symbols represent mean of duplicate measurements for individual macaques of two replicates for IL-1RA and a single replicate for TNF. Statistical significance was determined using a two-sided Mann–Whitney U-test without correction and P values are indicated on the graphs (c, d).

a

IU / mL

600 400 300

MF1

MF5

MF2

MF6

600 400

MF3

MF7

300

MF4

MF8

200

100

100 0

2

4

6

d

MF24 MF25

Lo D1

600 400

MF9 MF10

MF14 MF15

300

MF11

MF16 MF17

200

MF13

100

0

2

4

e MF23

300

0

8 10 12 14 16 18 20 Lo D5

600 400

c

Hi D1

MF12

200

0

IU / mL

b

CTRL

6

8 10 12 14 16 18 20

2

4

6

600 400

MF20

300

MF34 MF35 MF32

MF21

MF26

200

200

100

100

MF22

8 10 12 14 16 18 20

Days post infection

MF19

MF18

300

0

f

Hi D1 + AZTH

600 400

0

MF33

200

MF36

0

0

2

4

6

8 10 12 14 16 18 20

0

MF37

100 0

2

4

Days post infection Extended Data Fig. 7 | Plasma ALT levels of cynomolgus macaques treated with HCQ. In addition to the six SARS-CoV-2-infected groups, three groups of two macaques were treated but not infected to follow HCQ pharmacokinetics.

6

8 10 12 14 16 18 20

0

0

2

4

PK Hi PK Lo PK Hi + AZTH

6

Days post treatment a–f, ALT levels in the plasma are shown for all infected macaques (a–e) and the uninfected, HCQ-treated macaques (f). Shaded zones indicate treatment periods.

Article a

CTRL

b

Hi D1

c

d

Lo D1

Hi D1 + AZTH

e

Lo D5

f

PK MF34 MF35

MF1

MF5

MF2

MF6

MF3

MF7

MF4

MF8

MF9 MF10 MF11 MF12 MF13

MF18

MF14

MF19

MF15

MF20

MF16

MF21

MF17

MF22

MF23

MF32

MF24

MF33

MF25 MF26

MF37

PK Lo PK Hi + AZTH

300

IU/L

ASAT

400

MF36

PK Hi

200 100 0

0 2 4 6 8 10 12 14 16 18 20

0 2 4 6 8 10 12 14 16 18 20

0 2 4 6 8 10 12 14 16 18 20

0 2 4 6 8 10 12 14 16 18 20

0 2 4 6 8 10 12 14 16 18 20

0 2 4 6

0 2 4 6 8 10 12 14 16 18 20

0 2 4 6 8 10 12 14 16 18 20

0 2 4 6 8 10 12 14 16 18 20

0 2 4 6 8 10 12 14 16 18 20

0 2 4 6 8 10 12 14 16 18 20

0 2 4 6

0 2 4 6 8 10 12 14 16 18 20

0 2 4 6 8 10 12 14 16 18 20

0 2 4 6 8 10 12 14 16 18 20

0 2 4 6

50

g/L

Albumin

60

40 30 20

5,000

IU/L

Creatin K

15,000 800 600 400 200 0

0 2 4 6 8 10 12 14 16 18 20

0 2 4 6 8 10 12 14 16 18 20

mg/L

Creatinine

15 10 5 0

0 2 4 6 8 10 12 14 16 18 20

0 2 4 6 8 10 12 14 16 18 20

0 2 4 6 8 10 12 14 16 18 20

0 2 4 6 8 10 12 14 16 18 20

0 2 4 6 8 10 12 14 16 18 20

0 2 4 6

0 2 4 6 8 10 12 14 16 18 20

0 2 4 6 8 10 12 14 16 18 20

0 2 4 6 8 10 12 14 16 18 20

0 2 4 6 8 10 12 14 16 18 20

0 2 4 6 8 10 12 14 16 18 20

0 2 4 6

3

g/L

Haptoglobin

4

2 1 0 2,000

IU/L

LDH

800 600 400 200 0

0 2 4 6 8 10 12 14 16 18 20

0 2 4 6 8 10 12 14 16 18 20

0 2 4 6 8 10 12 14 16 18 20

0 2 4 6 8 10 12 14 16 18 20

0 2 4 6 8 10 12 14 16 18 20

0 2 4 6

0 2 4 6 8 10 12 14 16 18 20

0 2 4 6 8 10 12 14 16 18 20

0 2 4 6 8 10 12 14 16 18 20

0 2 4 6 8 10 12 14 16 18 20

0 2 4 6 8 10 12 14 16 18 20

0 2 4 6

0 2 4 6 8 10 12 14 16 18 20

0 2 4 6 8 10 12 14 16 18 20

0 2 4 6 8 10 12 14 16 18 20

0 2 4 6 8 10 12 14 16 18 20

0 2 4 6 8 10 12 14 16 18 20

0 2 4 6

100

mg/L

CRP

80 60 40 20 0

80

g/L

Total protein

100

60 40 20 0

Days post infection Extended Data Fig. 8 | Biochemistry analysis of cynomolgus macaques treated with HCQ. In addition to the six SARS-CoV-2-infected groups, three groups of two macaques were treated but not infected to follow HCQ pharmacokinetics. a–f, AST, albumin, creatine kinase, creatinine, haptoglobin,

Days post treatment

LDH, C-reactive protein (CRP) and total proteins levels were analysed in the plasma of all infected (a–e) and uninfected, treated (f) groups. Shaded zones indicate treatment periods.

Last updated by author(s): Jun 3, 2020

Reporting Summary Nature Research wishes to improve the reproducibility of the work that we publish. This form provides structure for consistency and transparency in reporting. For further information on Nature Research policies, see our Editorial Policies and the Editorial Policy Checklist.

nature research | reporting summary

Corresponding author(s): Roger LE GRAND

Statistics For all statistical analyses, confirm that the following items are present in the figure legend, table legend, main text, or Methods section. n/a Confirmed The exact sample size (n) for each experimental group/condition, given as a discrete number and unit of measurement A statement on whether measurements were taken from distinct samples or whether the same sample was measured repeatedly The statistical test(s) used AND whether they are one- or two-sided Only common tests should be described solely by name; describe more complex techniques in the Methods section.

A description of all covariates tested A description of any assumptions or corrections, such as tests of normality and adjustment for multiple comparisons A full description of the statistical parameters including central tendency (e.g. means) or other basic estimates (e.g. regression coefficient) AND variation (e.g. standard deviation) or associated estimates of uncertainty (e.g. confidence intervals) For null hypothesis testing, the test statistic (e.g. F, t, r) with confidence intervals, effect sizes, degrees of freedom and P value noted Give P values as exact values whenever suitable.

For Bayesian analysis, information on the choice of priors and Markov chain Monte Carlo settings For hierarchical and complex designs, identification of the appropriate level for tests and full reporting of outcomes Estimates of effect sizes (e.g. Cohen's d, Pearson's r), indicating how they were calculated Our web collection on statistics for biologists contains articles on many of the points above.

Software and code Policy information about availability of computer code Data collection

Data was collected using classical Excel 2016 Files and stored in a custom database called BatLab. All the data is available upon request to the corresponding author

Data analysis

Data was analyzed using GraphPad Prism 7,INTELLISPACE PORTAL 8 software

For manuscripts utilizing custom algorithms or software that are central to the research but not yet described in published literature, software must be made available to editors and reviewers. We strongly encourage code deposition in a community repository (e.g. GitHub). See the Nature Research guidelines for submitting code & software for further information.

Data Policy information about availability of data All manuscripts must include a data availability statement. This statement should provide the following information, where applicable:

The data that support the findings of this study are available from the corresponding author upon reasonable request. GISAID EpiCov platform accession numbers are: EPI_ISL_406596 for hCoV-19/France/lDF0372/2020 and EPI_ISL_411218 for hCoV-19/France/IDF0571/2020

April 2020

- Accession codes, unique identifiers, or web links for publicly available datasets - A list of figures that have associated raw data - A description of any restrictions on data availability

1

Please select the one below that is the best fit for your research. If you are not sure, read the appropriate sections before making your selection. Life sciences

Behavioural & social sciences

Ecological, evolutionary & environmental sciences

For a reference copy of the document with all sections, see nature.com/documents/nr-reporting-summary-flat.pdf

Life sciences study design All studies must disclose on these points even when the disclosure is negative. Sample size

No sample size calculation was performed, Samples size was determined as the minimal number of animals allowing non-parametric statistical analysis while complying with the 3Rs rule on reducing, replacing and refining the use of animals for scientific purpose.

Data exclusions

No data has been excluded from analysis

Replication

Duplicates were performed for all measurments (viral loads, drug concentration)

Randomization

Animal were randomly allocated to experimental groups. There was no other group comparison studies requiring randomization in the reported work

Blinding

Animals care, clinical examination and sampling was not blinded because constrains associated to BSL3 containment. However, viral load and drug concentrations were determined blinded by the groups at Institut Pasteur and Aix-Marseille University, respectively

nature research | reporting summary

Field-specific reporting

Reporting for specific materials, systems and methods We require information from authors about some types of materials, experimental systems and methods used in many studies. Here, indicate whether each material, system or method listed is relevant to your study. If you are not sure if a list item applies to your research, read the appropriate section before selecting a response.

Materials & experimental systems

Methods

n/a Involved in the study

n/a Involved in the study

Antibodies

ChIP-seq

Eukaryotic cell lines

Flow cytometry

Palaeontology and archaeology

MRI-based neuroimaging

Animals and other organisms Human research participants Clinical data Dual use research of concern

Eukaryotic cell lines Policy information about cell lines Cell line source(s)

African green monkey kidney VERO C1008 cell line [Vero 76, clone E6, Vero E6] (ATCC® CRL-1586™)

Authentication

This cell line has been directly obtained from ATCC but not authenticated in-house

Mycoplasma contamination

Confirmed negative test for mycoplasma contamination

Commonly misidentified lines

No commonly misidentified cell lines were used in the study

(See ICLAC register)

Policy information about studies involving animals; ARRIVE guidelines recommended for reporting animal research Laboratory animals

The study have included 35 cynomolgus macaques (Macaca fascicularis), both of male and female gender, aged 37-40 months

Wild animals

No wild animals were used in the studyl. Al used animals were purposed bred and imported from AAALAC certified breeding centers

Field-collected samples

No data was collected from the field

April 2020

Animals and other organisms

2

The protocols were approved by the institutional ethical committee “Comité d'Ethique en Expérimentation Animale du Commissariat à l’Energie Atomique et aux Energies Alternatives” (CEtEA #44) under statement number A20-011. The study was authorized by the “Research, Innovation and Education Ministry” under registration number APAFIS#24434-2020030216532863v1.

Note that full information on the approval of the study protocol must also be provided in the manuscript.

nature research | reporting summary

Ethics oversight

April 2020

3

Article

Chloroquine does not inhibit infection of human lung cells with SARS-CoV-2 https://doi.org/10.1038/s41586-020-2575-3 Received: 8 May 2020 Accepted: 16 July 2020 Published online: 22 July 2020 Check for updates

Markus Hoffmann1,2 ✉, Kirstin Mösbauer3,4, Heike Hofmann-Winkler1, Artur Kaul1, Hannah Kleine-Weber1,2, Nadine Krüger1, Nils C. Gassen5, Marcel A. Müller3,4,6, Christian Drosten3,4 & Stefan Pöhlmann1,2 ✉

The coronavirus disease 2019 (COVID-19) pandemic, which is caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), has been associated with more than 780,000 deaths worldwide (as of 20 August 2020). To develop antiviral interventions quickly, drugs used for the treatment of unrelated diseases are currently being repurposed to treat COVID-19. Chloroquine is an anti-malaria drug that is used for the treatment of COVID-19 as it inhibits the spread of SARS-CoV-2 in the African green monkey kidney-derived cell line Vero1–3. Here we show that engineered expression of TMPRSS2, a cellular protease that activates SARS-CoV-2 for entry into lung cells4, renders SARS-CoV-2 infection of Vero cells insensitive to chloroquine. Moreover, we report that chloroquine does not block infection with SARS-CoV-2 in the TMPRSS2-expressing human lung cell line Calu-3. These results indicate that chloroquine targets a pathway for viral activation that is not active in lung cells and is unlikely to protect against the spread of SARS-CoV-2 in and between patients.

Chloroquine and hydroxychloroquine are used for the treatment of malaria and have been widely used to treat patients with COVID-19. Both of these drugs are currently under investigation in more than 80 registered clinical trials for the treatment of COVID-19 worldwide2,3. Chloroquine and hydroxychloroquine inhibit the ability of SARS-CoV-2 to infect Vero cells1,5,6, providing a rational for using these drugs for the treatment of COVID-19. However, it is unknown whether these drugs inhibit the infection of lung cells and it is poorly understood how they inhibit infection with SARS-CoV-2. Chloroquine and hydroxychloroquine increase the endosomal pH of cells and inhibit viruses that depend on low pH for cell entry7. We investigated whether these drugs could also block the cell entry by SARS-CoV-2 and whether entry inhibition accounted for the prevention of infection with SARS-CoV-2. Moreover, we investigated whether entry inhibition is cell-type-dependent, as the virus can use pH-dependent and pH-independent pathways for entry into cells. The spike (S) protein of SARS-CoV-2, which mediates viral entry, is activated by the endosomal-pH-dependent cysteine protease cathepsin L (CTSL) in some cell lines4. By contrast, entry into airway epithelial cells, which express low levels of CTSL8, depends on the pH-independent, plasma-membrane-resident serine protease TMPRSS24. Notably, the use of CTSL by coronaviruses is restricted to cell lines8–10, whereas TMPRSS2 activity is essential for the spread and pathogenesis of the virus in the infected host11,12. We compared the inhibition by chloroquine and hydroxychloroquine of S-mediated entry into Vero (kidney), TMPRSS2-expressing Vero and Calu-3 (lung) cells. Calu-3 cells, as with the airway epithelium, express low amounts of CTSL8 and SARS-CoV-2 entry into these cells is dependent on TMPRSS24. By contrast, entry of SARS-CoV-2

into Vero cells is CTSL-dependent, and both CTSL and TMPRSS2 support entry into TMPRSS2-expressing Vero cells4. As a control, we used camostat mesylate, which inhibits TMPRSS2-dependent entry into cells4. Treatment with camostat mesylate did not interfere with cell viability, whereas chloroquine and hydroxychloroquine slightly reduced the viability of Vero, TMPRSS2-expressing Vero and Calu-3 cells when applied at the highest concentration (Fig. 1a). Inhibition of S-driven entry by camostat mesylate was observed only in TMPRSS2+ cell lines, as expected (Fig. 1a and Table 1). Moreover, chloroquine and hydroxychloroquine inhibited S-driven entry into TMPRSS2− Vero cells with high efficiency whereas the inhibition of entry into TMPRSS2+ Calu-3 and TMPRSS2+ Vero cells was inefficient and absent, respectively (Fig. 1a and Table 1). Therefore, chloroquine and hydroxychloroquine can block S-driven entry, but this inhibition is cell-line-dependent and efficient inhibition is not observed in TMPRSS2+ lung cells. We next investigated whether the cell-type-dependent differences in entry inhibition translated into differential inhibition of authentic SARS-CoV-2. Indeed, chloroquine efficiently blocked SARS-CoV-2 infection of Vero kidney cells, as expected1, but did not efficiently inhibit SARS-CoV-2 infection of Calu-3 lung cells (Fig. 1b, c). A subtle reduction in SARS-CoV-2 infection was seen in the presence of 100 μM chloroquine, consistent with the modest inhibition of cellular entry of S-bearing pseudotypes under those conditions (Fig. 1a), but this effect was not statistically significant. In summary, chloroquine did not efficiently block the infection of Calu-3 cells with S-bearing pseudotypes and authentic SARS-CoV-2, indicating that—in these cells—chloroquine does not appreciably interfere with viral entry or the subsequent steps of the viral replication cycle.

1 Infection Biology Unit, German Primate Center - Leibniz Institute for Primate Research, Göttingen, Germany. 2Faculty of Biology and Psychology, Georg-August-University Göttingen, Göttingen, Germany. 3Institute of Virology, Charité-Universitätsmedizin Berlin, corporate member of Freie Universität Berlin, Humboldt-Universität zu Berlin, Berlin Institute of Health, Berlin, Germany. 4German Centre for Infection Research, associated partner Charité, Berlin, Germany. 5Department of Psychiatry and Psychotherapy, University of Bonn, Bonn, Germany. 6Martsinovsky Institute of Medical Parasitology, Tropical and Vector Borne Diseases, Sechenov University, Moscow, Russia. ✉e-mail: [email protected]; [email protected]

588 | Nature | Vol 585 | 24 September 2020

Vero

*

75

*** ***

50

*** ***

25 0 0.1 1 10 Concentration (μM)

***

***

25

150

Vero

100

0.1 1 10 Concentration (μM)

25

50

150

50 25 0

0.01

0.1 1 10 Concentration (μM)

Extracellular virus

13

NS

9

*

7

***

***

NS

NS

NS

11 NS

NS

NS

9 7

3 2

3

0

100

100

0

1

10

100

Chloroquine (μM)

Fig. 1 | Chloroquine does not block infection of human lung cells with SARS-CoV-2. a, Vero, TMPRSS2-expressing Vero and Calu-3 cells were preincubated for 2 h with the respective inhibitors (0 μM, 0.01 μM, 0.1 μM, 1 μM, 10 μM or 100 μM) and then inoculated with replication-defective vesicular stomatitis virus reporter particles bearing the S protein. Top, the transduction efficiency of the virus was assessed. Bottom, cells were not inoculated with virus particles but cell viability after drug treatment was instead assessed at the same time as transduction was quantified. Transduction efficiency was quantified by measuring virus-encoded luciferase activity in cell lysates. Cell viability was measured using the CellTiter-Glo assay. Data are mean ± s.e.m. of three biological replicates, each of which consisted of quadruplicate samples. Data were normalized as the relative entry efficiency or cell viability of inhibitor-treated cells compared with those of untreated cells (set to 100%). The calculated 50% inhibitory concentration (IC50) values are summarized in Table 1. b, Untreated or chloroquine-preincubated Vero and Calu-3 cells were inoculated with SARS-CoV-2 Munich isolate (patient isolate 929, BetaCoV/Munich/BavPat1/2020|EPI_ISL_406862) at a multiplicity of infection (MOI) of 0.001. After inoculation for 24 h, viral RNA was isolated from the culture supernatant (extracellular virus) (dark blue) and the infected cells (intracellular virus) (light blue), and SARS-CoV-2 genome equivalents (GE) were determined by quantitative PCR with reverse transcription. Data are mean ± s.e.m. of three biological replicates, each of which consisted of single

Research to confirm our results in primary respiratory epithelium is ongoing. Moreover, virus production in Calu-3 cells relative to Vero E6 cells was more robust in the present study compared with a previously published study13, potentially due to the use of the Calu-3 subclone 2B4 in the previous but not the present study. Nevertheless, our results suggest that chloroquine and hydroxychloroquine will exert no antiviral activity in human lung tissue and will not be effective against COVID-19, in keeping with the results of recent clinical trials14,15. Moreover, our results highlight the fact that cell lines that mimic important aspects of

Calu-3

6

***

4

3

10

0.1 1 10 Concentration (μM)

7

NS

5

1

1

Vero

6

5

Chloroquine (μM)

50

0.01

7

5 0

75

c

13

11

* ***

100

100

Calu-3

15

***

Calu-3

0 0.1 1 10 Concentration (μM)

Intracellular virus

NS

*** 100

25 0.01

100

Vero

15

*

75

0

***

0.1 1 10 Concentration (μM)

125

100

25

*** 0.01

TMPRSS2-expressing Vero

log10(PFU per ml)

75

***

50

100

Cell viability (%)

Cell viability (%)

*

** ***

75

0

125

100

b

***

50

0.01

log10(GE per ml)

Cell viability (%)

75

100

125

log10(GE per ml)

100

Calu-3

125

0 0.01

150

150

TMPRSS2-expressing Vero

125

log10(PFU per ml)

100

Entry efficiency (%)

Entry efficiency (%)

150

125

Hydroxychloroquine

Chloroquine

Camostat mesylate 150

Entry efficiency (%)

a

NS

5

NS

NS

4 3 2 1

*** 0

1

0

10 100

Chloroquine (μM)

0

1

10 100

Chloroquine (μM)

samples. c, The experiment was conducted as described in b, but the number of infectious SARS-CoV-2 particles in culture supernatants was determined by plaque titration using Vero E6 cells. PFU, plaque-forming units. Statistical significance was analysed by two-way analysis of variance (ANOVA) with Dunnett’s post hoc test. NS, not significant (P > 0.05); *P ≤ 0.05; **P ≤ 0.01; ***P ≤ 0.001. P values (from left to right) are as follows. a, Entry efficiency (camostat mesylate/chloroquine/hydroxychloroquine), Vero (0.9999/0.8587/0.9997, 0.9842/0.9846/0.3904, 0.6860/0.0991/0.0223, 0.9968/0.0001/0.0001, 0.9997/0.0001/0.0001), TMPRSS2-expressing Vero (0.9999/0.9968/0.9795, 0.1251/0.9962/0.9998, 0.0004/0.9997/ 0.9999, 0.0001/0.9967/0.9982, 0.0001/0.9981/0.9986; Calu-3 (0.9900/ 0.9999/0.9986, 0.0003/0.9999/0.9983, 0.0001/0.9988/0.9929, 0.0001/ 0.1291/0.9938, 0.0001/0.0005/0.0045); cell viability (camostat mesylate/ chloroquine/hydroxychloroquine), Vero (0.9273/0.9999/0.9999, 0.9999/ 0.8710/0.9642, 0.9999/0.9996/0.9999, 0.9999/0.8958/0.4818, 0.9998/ 0.0838/0.0161), TMPRSS2-expressing Vero (0.9998/0.9999/0.9959, 0.9811/0.9985/0.9362, 0.9998/0.9985/0.9997, 0.9997/0.8835/0.9998, 0.9999/0.0315/0.1422), Calu-3 (0.9986/0.9999/0.9999, 0.9999/0.9997/0.9999, 0.9986/0.9999/0.8134, 0.9924/0.9275/0.7125, 0.9983/0.0492/0.0002). b, (extracellular/intracellular), Vero (0.6844/0.6989, 0.0121/0.0002, 0.0002/ 0.0001), Calu-3 (0.9434/0.8800, 0.9999/0.8830, 0.0517/0.3924). c, (extracellular/ intracellular), Vero (0.9561, 0.0001, 0.0001), Calu-3 (0.1184, 0.9997, 0.0987).

Table 1 | Half-maximal inhibitory concentrations of the tested drugs IC50 (μM) Vero

TMPRSS2+ Vero

Calu-3

Camostat mesylate

ND

5.7

0.083

Chloroquine

6.5

ND

64.7

Hydroxychloroquine

13.3

ND

119

ND, not determined.

Nature | Vol 585 | 24 September 2020 | 589

Article respiratory epithelial cells should be used when analysing the antiviral activity of compounds that target host cell functions.

6. 7.

Online content

8.

Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41586-020-2575-3.

9.

1. 2. 3. 4. 5.

Wang, M. et al. Remdesivir and chloroquine effectively inhibit the recently emerged novel coronavirus (2019-nCoV) in vitro. Cell Res. 30, 269–271 (2020). Ferner, R. E. & Aronson, J. K. Chloroquine and hydroxychloroquine in covid-19. Br. Med. J. 369, m1432 (2020). Touret, F. & de Lamballerie, X. Of chloroquine and COVID-19. Antiviral Res. 177, 104762 (2020). Hoffmann, M. et al. SARS-CoV-2 cell entry depends on ACE2 and TMPRSS2 and is blocked by a clinically proven protease inhibitor. Cell 181, 271–280 (2020). Yao, X. et al. In vitro antiviral activity and projection of optimized dosing design of hydroxychloroquine for the treatment of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). Clin. Infect. Dis. 71, 732–739 (2020).

590 | Nature | Vol 585 | 24 September 2020

10. 11. 12. 13. 14. 15.

Liu, J. et al. Hydroxychloroquine, a less toxic derivative of chloroquine, is effective in inhibiting SARS-CoV-2 infection in vitro. Cell Discov. 6, 16 (2020). Rolain, J. M., Colson, P. & Raoult, D. Recycling of chloroquine and its hydroxyl analogue to face bacterial, fungal and viral infections in the 21st century. Int. J. Antimicrob. Agents 30, 297–308 (2007). Park, J. E. et al. Proteolytic processing of Middle East respiratory syndrome coronavirus spikes expands virus tropism. Proc. Natl Acad. Sci. USA 113, 12262–12267 (2016). Shirato, K., Kawase, M. & Matsuyama, S. Wild-type human coronaviruses prefer cell-surface TMPRSS2 to endosomal cathepsins for cell entry. Virology 517, 9–15 (2018). Shirato, K., Kanou, K., Kawase, M. & Matsuyama, S. Clinical isolates of human coronavirus 229E bypass the endosome for cell entry. J. Virol. 91, e01387-16 (2016). Iwata-Yoshikawa, N. et al. TMPRSS2 contributes to virus spread and immunopathology in the airways of murine models after coronavirus infection. J. Virol. 93, e01815-18 (2019). Zhou, Y. et al. Protease inhibitors targeting coronavirus and filovirus entry. Antiviral Res. 116, 76–84 (2015). Matsuyama, S. et al. Enhanced isolation of SARS-CoV-2 by TMPRSS2-expressing cells. Proc. Natl Acad. Sci. USA 117, 7001–7003 (2020). Boulware, D. R. et al. A randomized trial of hydroxychloroquine as postexposure prophylaxis for COVID-19. N. Engl. J. Med. 383, 517–525 (2020). Kupferschmidt, K. Big studies dim hopes for hydroxychloroquine. Science 368, 1166–1167 (2020).

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. © The Author(s), under exclusive licence to Springer Nature Limited 2020

Methods Data reporting No statistical methods were used to predetermine sample size. The experiments were not randomized and the investigators were not blinded to allocation during experiments and outcome assessment. Cells Vero 76, Vero 76 stably expressing TMPRSS2 (both used for pseudotype experiments)4, the Vero 76 subclone Vero E6 (used for SARS-CoV-2 experiments), HEK293T and Calu-3 cells16 were cultured in Dulbecco’s modified Eagle’s medium (DMEM) or minimum essential medium (MEM, Calu-3) supplemented with 10% fetal bovine serum (FBS) and 1% penicillin–streptomycin. In case of Calu-3 cells, the medium was also supplemented with 1% non-essential amino acids and 1% sodium pyruvate. All cell lines were incubated at 37 °C and 5% CO2 and were obtained from repositories (Vero E6 and HEK293T) or collaborators (Calu-3 and Vero 76). Cell lines were free of mycoplasma, authenticated on the basis of morphology and growth properties and confirmed by PCR to be of the correct species. The cell lines used were not listed as commonly misidentified cell lines by the ICLAC register. Production of pseudotyped particles Vesicular stomatitis virus (VSV) particles pseudotyped with SARS-CoV-2 S were generated according to published protocols4,17. At 24 h after transfection, HEK293T cells expressing the S protein were inoculated with a replication-restricted, VSV-G-trans-complemented VSV, which lacks the genetic information for VSV-G but instead encodes the reporter genes eGFP (enhanced green fluorescent protein) and FLuc (firefly luciferase), VSV*ΔG-FLuc18 (provided by G. Zimmer). After 1 h of incubation at 37 °C and 5% CO2, the inoculum was aspirated and the cells were washed with phosphate-buffered saline (PBS) before culture medium was added. The culture medium was further supplemented with the culture supernatant from I1-hybridoma cells (CRL-2700 cells, ATCC) containing anti-VSV-G antibody (1:1,000) to inactivate residual input virus. After an incubation period of 18 h at 37 °C and 5% CO2, the culture supernatant was collected, centrifuged to pellet cellular debris, and the clarified supernatant was aliquoted and stored at −80 °C until further use. Transduction of target cells with pseudotypes and its inhibition For transduction experiments, Vero, TMPRSS2-expressing Vero and Calu-3 cells were grown in 96-well plates and allowed to reach about 50–70% confluency. Then, cells were preincubated with medium containing different concentrations (10 nM, 100 nM, 1 μM, 10 μM or 100 μM) of camostat mesylate (Sigma-Aldrich), chloroquine or hydroxychloroquine (both Tocris) or DMSO (Roth, solvent control) for 2 h at 37 °C and 5% CO2, before they were inoculated with S-bearing VSV. At 18 h after transduction, culture supernatants were aspirated and cells were lysed by incubation (30 min, room temperature) with Cell Culture Lysis Reagent (Promega). Cell lysates were subsequently transferred to white, opaque-walled 96-well plates and FLuc activity was quantified as an indicator of transduction efficiency, using the Beetle-Juice substrate (PJK) and a Hidex Sense plate reader (Hidex) operated with Hidex plate reader software (version 0.5.41.0, Hidex). Raw luminescence values (indicating luciferase activity) were recorded as counts per second. For normalization, transduction of DMSO-treated cells was set to 100% and the relative transduction efficiencies in the presence of camostat mesylate, chloroquine or hydroxychloroquine were calculated. Transduction experiments were performed in technical quadruplicates using three separate pseudotype preparations. SARS-CoV-2 infection of target cells and its inhibition Virus infections were done with SARS-CoV-2 Munich isolate 929. Vero E6 or Calu-3 cells were seeded at densities of 3.5 × 105 cells per ml or 6 × 105

cells per ml in 12-well plates, respectively. After 24 h, cells were incubated with chloroquine (1 μM, 10 μM or 100 μM) or left untreated (control) for 1 h at 37 °C. Subsequently, cells were infected with an MOI of 0.001 in serum-free OPTIpro medium containing the above-mentioned chloroquine concentrations at 4 °C for 30 min to enable virus attachment. Afterwards, infection medium was removed and the wells were washed twice with PBS and DMEM supplemented with chloroquine was added as described above and the plates were incubated at 37 °C. Samples were taken at 24 h after infection. Infection experiments were conducted with biological triplicates in a biosafety level 3 laboratory.

Viral RNA extraction and quantitative RT–PCR For viral RNA extraction from supernatants, 50 μl of cell culture supernatant was mixed with RAV1 lysis buffer (Macherey-Nagel) followed by an incubation at 70 °C for 10 min. RNA extraction was performed as recommended by the manufacturer (Macherey-Nagel). For intracellular viral RNA extraction, cells were washed with PBS and lysed with TRIzol (Zymo). SARS-CoV-2 genome equivalents were detected by quantitative RT–PCR targeting the SARS-CoV-2 E gene as previously reported19, using the following primers: E_Sarbeco_F, ACAGGTACGTTAATAGTTAATAGCGT; E_Sarbeco_P1, FAM-ACACTAGCCATCCTTACTGCGCTTCG-BBQ; E_Sarbeco_R, ATATTGCAGCAGTACGCACACA. The quantitative RT–PCR experiment and data processing were carried out using the LightCycler 480 Real-Time PCR System (Roche) and LightCycler 480 Software (version 1.5, Roche Molecular Systems). Absolute quantification was performed using SARS-CoV-2-specific in vitro-transcribed RNA standards, as previously described19. Plaque assay Infectious SARS-CoV-2 plaque-forming units were quantified by plaque titration on Vero E6 cells, as previously described20, with minor modifications. Vero E6 monolayers were seeded in 24-well plates, washed with PBS, incubated with serial dilutions of SARS-CoV-2-containing cell culture supernatants in duplicates, and overlaid with 1.2% Avicel in DMEM, supplemented as described above. After 72 h, cells were fixed with 6% formaline and visualized by crystal violet staining. Cell viability assay The cell viability was quantified using the CellTiter-Glo assay (Promega) and using the same experimental conditions as described above for transduction experiments with the exception that cells were not inoculated with virus particles. In brief, cells were preincubated for 2 h at 37 °C and 5% CO2 with medium containing different concentrations (10 nM, 100 nM, 1 μM, 10 μM or 100 μM) of camostat mesylate, chloroquine or hydroxychloroquine, or DMSO (solvent control), before culture medium was added (instead of medium containing VSV pseudotyped with SARS-CoV-2 S) and cells were further incubated for 18 h. Next, intracellular ATP levels were quantified as an indicator of cell viability. For this, culture supernatants were aspirated and cells were lysed by incubation with CellTiter-Glo substrate for 30 min at room temperature. Cell lysates were subsequently transferred to white, opaque-walled 96-well plates and luminescence was measured using a Hidex Sense plate reader (Hidex). Luminescence values (indicating cell viability) were recorded as absolute counts over a period of 200 ms per well. For normalization, cell viability of control-treated cells was set to 100% and the relative viability of cells incubated in the presence of camostat mesylate, chloroquine or hydroxychloroquine was calculated. Cell viability experiments were performed in technical quadruplicates and repeated with three separately prepared dilution series of the inhibitors. Statistical analysis Two-way ANOVA with Dunnett’s post hoc test was performed to analyse statistical significance of differences in transduction efficiencies, SARS-CoV-2 genome equivalents or SARS-CoV-2 titres between control- and inhibitor-treated cells. P > 0.05, not significant; *P ≤ 0.05;

Article **P ≤ 0.01; ***P ≤ 0.001). The IC50 values, which indicate the inhibitor concentration that led to a 50% reduction in transduction, were calculated using a nonlinear regression model with variable slope. Statistical analyses and IC50 calculations were performed using GraphPad Prism (version 8.4.2).

Reporting summary Further information on research design is available in the Nature Research Reporting Summary linked to this paper.

Data availability All data are provided with the paper. Source data are provided with this paper.

18. Berger Rentsch, M. & Zimmer, G. A vesicular stomatitis virus replicon-based bioassay for the rapid and sensitive determination of multi-species type I interferon. PLoS ONE 6, e25858 (2011). 19. Corman, V. M. et al. Detection of 2019 novel coronavirus (2019-nCoV) by real-time RT-PCR. Eurosurveill. 25, 2000045 (2020). 20. Herzog, P., Drosten, C. & Müller, M. A. Plaque assay for human coronavirus NL63 using human colon carcinoma cells. Virol. J. 5, 138 (2008). Acknowledgements This work was supported by the Bundesministerium für Bildung und Forschung (RAPID Consortium, 01KI1723A and 01KI1723D to C.D. and S.P., respectively). We thank A. Maisner (Vero 76 cell line), S. Ludwig (Calu-3 cell line) and G. Zimmer (VSV pseudotype system) for providing cell lines and reagents. Author contributions M.H. and S.P. designed the study. M.H., K.M., H.H.-W., A.K., H.K.-W., N.K., N.C.G. and M.A.M. performed research. M.H., M.A.M., C.D. and S.P. analysed the data. C.D. provided essential reagents. M.H. and S.P. wrote the manuscript. All authors revised the manuscript. Competing interests The authors declare no competing interests.

16. Klemm, C. et al. Mitogen-activated protein kinases (MAPKs) regulate IL-6 over-production during concomitant influenza virus and Staphylococcus aureus infection. Sci. Rep. 7, 42473 (2017). 17. Kleine-Weber, H. et al. Mutations in the spike protein of Middle East respiratory syndrome coronavirus transmitted in Korea increase resistance to antibody-mediated neutralization. J. Virol. 93, e01381-18 (2019).

Additional information Supplementary information is available for this paper at https://doi.org/10.1038/s41586-0202575-3. Correspondence and requests for materials should be addressed to M.H. or S.P. Reprints and permissions information is available at http://www.nature.com/reprints.

Last updated by author(s): Jun 29, 2020

Reporting Summary Nature Research wishes to improve the reproducibility of the work that we publish. This form provides structure for consistency and transparency in reporting. For further information on Nature Research policies, see our Editorial Policies and the Editorial Policy Checklist.

nature research | reporting summary

Corresponding author(s): Markus Hoffmann, Stefan Pöhlmann

Statistics For all statistical analyses, confirm that the following items are present in the figure legend, table legend, main text, or Methods section. n/a Confirmed The exact sample size (n) for each experimental group/condition, given as a discrete number and unit of measurement A statement on whether measurements were taken from distinct samples or whether the same sample was measured repeatedly The statistical test(s) used AND whether they are one- or two-sided Only common tests should be described solely by name; describe more complex techniques in the Methods section.

A description of all covariates tested A description of any assumptions or corrections, such as tests of normality and adjustment for multiple comparisons A full description of the statistical parameters including central tendency (e.g. means) or other basic estimates (e.g. regression coefficient) AND variation (e.g. standard deviation) or associated estimates of uncertainty (e.g. confidence intervals) For null hypothesis testing, the test statistic (e.g. F, t, r) with confidence intervals, effect sizes, degrees of freedom and P value noted Give P values as exact values whenever suitable.

For Bayesian analysis, information on the choice of priors and Markov chain Monte Carlo settings For hierarchical and complex designs, identification of the appropriate level for tests and full reporting of outcomes Estimates of effect sizes (e.g. Cohen's d, Pearson's r), indicating how they were calculated Our web collection on statistics for biologists contains articles on many of the points above.

Software and code Policy information about availability of computer code Data collection

- Hidex Sense Microplate Reader Software (version 0.5.41.0), Hidex Deutschland Vertrieb GmbH, https://www.hidex.de/ - LightCycler® 480 Software (version 1.5), Roche Molecular Systems, Inc., https://lifescience.roche.com/

Data analysis

- GraphPad Prism (version 8.4.2), GraphPad Software, https://www.graphpad.com/

For manuscripts utilizing custom algorithms or software that are central to the research but not yet described in published literature, software must be made available to editors and reviewers. We strongly encourage code deposition in a community repository (e.g. GitHub). See the Nature Research guidelines for submitting code & software for further information.

Data Policy information about availability of data All manuscripts must include a data availability statement. This statement should provide the following information, where applicable:

All source data are available within the paper.

April 2020

- Accession codes, unique identifiers, or web links for publicly available datasets - A list of figures that have associated raw data - A description of any restrictions on data availability

1

Please select the one below that is the best fit for your research. If you are not sure, read the appropriate sections before making your selection. Life sciences

Behavioural & social sciences

Ecological, evolutionary & environmental sciences

For a reference copy of the document with all sections, see nature.com/documents/nr-reporting-summary-flat.pdf

Life sciences study design All studies must disclose on these points even when the disclosure is negative. Sample size

No sample size calculations were performed. Sample sizes adhere to standards in the field; for instance, three biological replicates performed with technical replicates.

Data exclusions

No data were excluded from the analysis.

Replication

All findings were confirmed in three independent experiments. All results could be reproduced.

Randomization

Experiments were performed in vitro with standard immortalized cell lines. Therefore, no randomization was required.

Blinding

No groups were allocated. Therefore, blinding was not necessary.

nature research | reporting summary

Field-specific reporting

Reporting for specific materials, systems and methods We require information from authors about some types of materials, experimental systems and methods used in many studies. Here, indicate whether each material, system or method listed is relevant to your study. If you are not sure if a list item applies to your research, read the appropriate section before selecting a response.

Materials & experimental systems

Methods

n/a Involved in the study

n/a Involved in the study

Antibodies

ChIP-seq

Eukaryotic cell lines

Flow cytometry

Palaeontology and archaeology

MRI-based neuroimaging

Animals and other organisms Human research participants Clinical data Dual use research of concern

Eukaryotic cell lines Policy information about cell lines - Vero76: Provided by Andrea Maisner (Philipps University Marburg, Marburg/Germany) - Vero76-TMPRSS2: Generated from parental Vero76 cells by retroviral transduction (Hoffmann et al., 2020) - VeroE6: Obtained from ATCC (CCL-81) - Calu-3: Provided by Stephan Ludwig (Institute of Virology (IVM), Westfaelische Wilhelms-University Muenster) and initially obtained from ATCC -293T: Obtained from DSMZ (ACC 635)

Authentication

Species origin of all cell lines was confirmed by sequencing a PCR-amplified fragment of the cytochrome c oxidase gene. Cell lines were further authenticated by analysis of morphology and growth properties.

Mycoplasma contamination

Cells were tested for the presence of mycoplasma by PCR. No contamination by mycoplasma was detected.

Commonly misidentified lines

The used cell lines are not listed as commonly misidentified cell lines by the ICLAC register.

(See ICLAC register)

April 2020

Cell line source(s)

2

Article

The liver–brain–gut neural arc maintains the Treg cell niche in the gut https://doi.org/10.1038/s41586-020-2425-3 Received: 20 November 2019 Accepted: 4 June 2020 Published online: 11 June 2020 Check for updates

Toshiaki Teratani1, Yohei Mikami1 ✉, Nobuhiro Nakamoto1, Takahiro Suzuki1,2, Yosuke Harada1, Koji Okabayashi3, Yuya Hagihara1, Nobuhito Taniki1, Keita Kohno4, Shinsuke Shibata5,6, Kentaro Miyamoto1,2, Harumichi Ishigame7, Po-Sung Chu1, Tomohisa Sujino1, Wataru Suda8, Masahira Hattori8,9, Minoru Matsui10, Takaharu Okada7,11, Hideyuki Okano5, Masayuki Inoue12, Toshihiko Yada13, Yuko Kitagawa3, Akihiko Yoshimura14, Mamoru Tanida15, Makoto Tsuda4, Yusaku Iwasaki16 & Takanori Kanai1,17 ✉

Recent clinical and experimental evidence has evoked the concept of the gut–brain axis to explain mutual interactions between the central nervous system and gut microbiota that are closely associated with the bidirectional effects of inflammatory bowel disease and central nervous system disorders1–4. Despite recent advances in our understanding of neuroimmune interactions, it remains unclear how the gut and brain communicate to maintain gut immune homeostasis, including in the induction and maintenance of peripheral regulatory T cells (pTreg cells), and what environmental cues prompt the host to protect itself from development of inflammatory bowel diseases. Here we report a liver–brain–gut neural arc that ensures the proper differentiation and maintenance of pTreg cells in the gut. The hepatic vagal sensory afferent nerves are responsible for indirectly sensing the gut microenvironment and relaying the sensory inputs to the nucleus tractus solitarius of the brainstem, and ultimately to the vagal parasympathetic nerves and enteric neurons. Surgical and chemical perturbation of the vagal sensory afferents at the hepatic afferent level reduced the abundance of colonic pTreg cells; this was attributed to decreased aldehyde dehydrogenase (ALDH) expression and retinoic acid synthesis by intestinal antigen-presenting cells. Activation of muscarinic acetylcholine receptors directly induced ALDH gene expression in both human and mouse colonic antigen-presenting cells, whereas genetic ablation of these receptors abolished the stimulation of antigen-presenting cells in vitro. Disruption of left vagal sensory afferents from the liver to the brainstem in mouse models of colitis reduced the colonic pTreg cell pool, resulting in increased susceptibility to colitis. These results demonstrate that the novel vago-vagal liver–brain–gut reflex arc controls the number of pTreg cells and maintains gut homeostasis. Intervention in this autonomic feedback feedforward system could help in the development of therapeutic strategies to treat or prevent immunological disorders of the gut.

FOXP3+ pTreg cells are most abundant in the mucosal tissues, especially the colonic lamina propria, and maintain immune homeostasis in the gut5,6. The generation of pTreg cells is promoted by a combination of cytokines, such as TGF-β and retinoic acid, and microbial and dietary signals, such as Clostridia clusters IV, XIVa and XVIII, Bacteroides fragilis, microbiota-associated molecular patterns and short chain fatty acids5–13. In addition to these numerous environmental stimuli, considerable advances have been made by recent studies showing that immune cells are under the control of autonomic and enteric neurons3,14–16. Indeed, the gastrointestinal tract is highly innervated and densely populated by adaptive and innate immune cells14,17. Consistently, immunohistochemical analysis of the localization of neurons (β-tubulin III+) and major histocompatibility complex class II+ (MHC-II+)

antigen-presenting cells (APCs), mainly comprising CX3CR1+ mononuclear phagocytes, revealed the close proximity of neurons and APCs in the colonic lamina propria (Fig. 1a, b, Extended Data Fig. 1a, b). Intestinal APCs, particularly CX3CR1+ mononuclear phagocytes and CD103+ dendritic cells, produce retinoic acid, which preferentially supports the development of pTreg cells in the TGF-β-rich microenvironment in the gut8,18–23. Despite the known immunoregulatory effects of the autonomic nervous system1–3,15–17, it remains incompletely understood how the vagus nerve influences gut homeostasis by regulating intestinal APCs and pTreg cells. To examine the immunological functions of the vagus nerve, we performed subdiaphragmatic truncal vagotomy (VGx) in wild-type C57BL/6 (B6) mice (Extended Data Fig. 1c, d). The vagotomized mice showed a significant reduction in the number of

A list of affiliations appears at the end of the paper.

Nature | Vol 585 | 24 September 2020 | 591

Article

20 10 0

Sham VGx

f

40 20 0

Count

40

CD11c SP

DP

Sham + DEAB

i

0

Sham VGx

VGx + DEAB

0

Aldefluor

CD11b SP

1

0.5

VGx

20

20

40

Row max

Sham VGx P < 0.001

30 20 10 0

Sham VGx

DP Aldh1a2

Cx3cr1 Itgam (CD11b) Itgax (CD11c) Itgae (CD103) Irf4 Irf8 Batf3 Row min

1.0 0

P = 0.006

2

1.5

Sham 16.7

Cx3cr1 Itgam

Concentration High Low

(CD11b)

Irf8 Itgae CD11b SP Adrb2 Aldh1a1

TPM

(CD103)

Chrm1

CD11c SP

0

6

10

10 5

0

0

Cont Ach Mus Adre NPY Sub P 5HT NMU

Relative mRNA level

15 P < 0.001

Cont Ach Mus Adre NPY Sub P 5HT NMU

Aldh1a2 3

P = 0.023

2.0

Sham VGx

60 24.3

40 0

Aldh1a1 2.5

VGx

80

h

60

Sham 60

e

Aldh1a2 20 P < 0.001

P < 0.001

Gated on APCs: CD45+CD3–B220–NK1.1–MHC-II+

k

Mouse

Relative mRNA level

30

Adra1a Adra1b Adra1d Adra2a Adra2b Adra2c Adrb1 Adrb2 Adrb3 Chrm1 Chrm2 Chrm3 Chrm4 Chrm5 Chrna2 Chrna3 Chrna4 Chrna5 Chrna6 Chrna7 Chrnb2 Chrnb3 Chrnb4 Htr1a Htr1b Htr1d Htr1f Htr2a Htr2b Htr2c Htr3a Htr3b Htr4 Htr7 Npy1r Npy2r Npy4r Npy5r Npy6r

Aldh1a1 20 P < 0.001

15

mAChR TKO

Aldh1a1

Aldh1a2

P = 0.998

P = 0.998

25 20

WT

P = 0.014 P = 0.005

25 20 15

10

10

5

5

0 Mus: –

+

l

0

P = 0.018 P = 0.015

–

+

Human ALDH1A2

ALDH1A1 Relative mRNA level

40

P < 0.001

80

35.5

CX3CR1–GFP

CD45.2

ALDH+ in MHC-II cells (%)

50

18.0

j

32.5

Relative mRNA level

d P < 0.001

RORγ t+ in Treg cells (%)

c

FOXP3+ in CD4 cells (%)

CX3CR1 β-tubulin III

CD11c

MHC-II

68.9

g

APC

B220–NK1.1–

Spleen

b CD45+TCRβ–CD3–

Colon

a

20

P = 0.008

15

P = 0.014

10 10 5 0 Mus: –

+

0

–

+

Fig. 1 | Potential interaction between APCs and neurons in the gut. a, Representative immunofluorescence of CX3CR1–GFP (green) and β-tubulin III (red) in the mouse colon. Scale bar, 50 μm. b, Representative CD11c and MHC-II staining of CD45.2+TCRβ− CD3−B220 −NK1.1− colonic lamina propria mononuclear cells from Cx3cr1 gfp mice. c–f, Eight-week-old male wild-type B6 mice were subjected to VGx or sham surgery. Colonic T cell phenotypes and colonic gene expression were analysed 2 days later (n = 12 per group). c, Frequency of FOXP3+ (Treg) cells among CD4+ T cells in colonic lamina propria. d, Expression of RORγt in colonic FOXP3+ Treg cells. e, Expression of Aldh1a1 and Aldh1a2 mRNA in colonic APCs. f, Frequency of ALDH+ cells among MHC-II+ APCs (CD45+TCRβ−CD3−B220 −NK1.1−MHC-II+) in the colon. Left, histograms of ALDH+ cells in APCs. Colonic mononuclear cells were incubated with Aldefluor in the absence (filled) or presence (dotted line) of the ALDH inhibitor diethylaminobenzaldehyde (DEAB). The percentage of Aldefluor+ cells is shown above the horizontal line indicating the positive gate. Right, quantification of ALDH+ cells. g, Heat map of the expression of genes encoding neurotransmitter receptors, classified by sorted colonic and splenic APCs, as determined by RNA-sequencing analysis. TPM, transcripts per million. h, Heat map of macrophage and dendritic cell marker genes (gene product in

parentheses) for colonic CD11b+CD11c− (CD11b SP), CD11b+CD11c+ (DP) and CD11b−D11c+ (CD11c SP) cells. The sorting strategy for the experiment is shown in Extended Data Fig. 1h. Max, maximum; min, minimum. i, Ternary plot of gene expression in colonic CD11b SP, DP and CD11c SP cells. The colour scale indicates mRNA concentration. Neurotransmitter receptors and representative markers for macrophage and dendritic cell are shown. j, Aldh1a1 and Aldh1a2 mRNA expression in colonic APCs treated with PBS (control), 10 μM acetylcholine (Ach), 10 μM muscarine (Mus), 100 nM adrenaline (Adre), 100 μM neuropeptide Y (NPY), 100 nM substance P (Sub P), 10 μM serotonin (5-HT) or 100 ng ml−1 neuromedin U (NMU) for 12 h (n = 5 per group). k, Aldh1a1 and Aldh1a2 expression in wild-type (WT) and mAChR TKO colonic APCs. Colonic APCs were isolated from wild-type or mAChR TKO mice and treated with 10 μM muscarine or untreated for 12 h (n = 6 per group). l, ALDH1A1 and ALDH1A2 mRNA levels in human colonic APCs. Colonic APCs were treated with 10 μM Mus or untreated for 12 h (n = 7 per group). Representative of three (a, b, j–l) independent experiments or pooled from three independent experiments (c–f). Data are mean ± s.e.m. P values by unpaired two-tailed Student’s t-test (c–f, k) or one-way ANOVA with Tukey’s post hoc test ( j–l).

FOXP3+ T helper cells, in particular Helios−RORγt+ pTreg cells, in the colon compared with sham-operated mice (Fig. 1c, d, Extended Data Fig. 1e, f). In addition to the reduction of colonic pTreg cells, there was a marked decrease in the levels of Aldh1a1 and Aldh1a2, which encode the retinoic acid-synthesizing enzymes RALDH1 and RALDH2, and in aldehyde dehydrogenase activity in colonic APCs (Fig. 1e, f). To identify the neurotransmitter responsible for conveying signals from the enteric neurons to colonic APCs, we performed RNA-sequencing analysis on APCs obtained from the spleen and intestine. The gut APCs exhibited higher levels of Chrm1 mRNA, which encodes the muscarinic acetylecholine receptor (mAchR), than splenic

APCs, suggesting a tissue-specific role for neurotransmitters in regulating intestinal APCs (Fig. 1g, Extended Data Fig. 1g). Of note, APC fractions enriched in CX3CR1+ mononuclear phagocytes and CD103+ dendritic cells share increased expression of Chrm1 as well as Aldh1a1 and Aldh1a2 compared with genes that define prototypical APC subsets, such as Itgae (which encodes CD103), Cx3cr1 and Irf8 (Fig. 1h, i, Extended Data Fig. 1h). We confirmed this finding by quantifying the expression of Aldh1a1 and Aldh1a2 in colonic APCs stimulated with multiple neurotransmitters, including acetylecholine, muscarine, adrenaline, neuropeptide Y, substance P, serotonin and neuromedin U (Fig. 1j). Furthermore, muscarine and enteric neurospheroids induced Aldh1a1 and Aldh1a2 expression in

592 | Nature | Vol 585 | 24 September 2020

a

b

Medulla oblongata : DSS (day 7) Sham NTS

AP

NTS

NTS

NG: DSS (day 7) Sham

HVx AP

HVx

NTS L

DMV

NG (R)

10

P < 0.001

e

8 6 4 2 0

L

R

P = 0.001

60

P = 0.032

40

P = 0.400

20 0

am

NG (L)

Sh

d

f

P = 0.918

80

P = 0.012

60 40 20 0

x

WGA

x

R HVx

LV

L Sham

Sham HVx

RV

0

am

R HVx

25 P = 0.710 20 P = 0.019 15 10 5 0 L R

Sh

L Sham

10

FOXP3+ in CD4 cells (%)

0

20

Vx

10

R

H

20

30 P = 0.018

pERK1/2 expression in NG per section

30

c-Fos expression in DMV per section

40 P = 0.009

R P = 0.354

FOXP3+ in CD4 cells (%)

c

L

R P = 0.592

No. of WGA+ neurons per field

c-Fos expression in NTS per section

L

DMV

x

DMV

VG

DMV

WGA / DAPI Fig. 2 | The hepatic vagal sensory afferent pathway is essential for NTS activation during colitis. a, b, Wild-type mice were subjected to HVx or sham surgery and were then given DSS for 7 days, starting at day 2 after surgery (n = 4 per group). DMV, dorsal motor nucleus of the vagus; AP, area postrema. a, Top, representative immunostaining for c-Fos in medulla oblongata. Scale bars, 200 μm. Bottom, relative c-Fos expression in NTS and DMV per section. L, left; R, right. b, Top, representative immunostaining for phosphorylated ERK1/2 (pERK1/2) in NG. Scale bars, 100 μm. Bottom, relative pERK1/2 level in NG per section. c, d, Wheat germ agglutinnin (WGA) retrograde tracing. c, Representative fluorescence images of WGA–Alexa Fluor 488 (green) and

DAPI (blue) in NG at 1 week after injection of WGA in liver. Arrowheads indicate WGA+ neurons in NG. Scale bars, 100 μm. d, Quantification of WGA+ neurons from c. e, Wild-type mice were subjected to VGx, HVx or sham surgery (n = 9 per group). f, Wild-type mice were subjected to ventral subdiaphragmatic vagotomy (LVx), dorsal subdiaphragmatic vagotomy (RVx) or sham surgery (n = 4 per group). e, f, Frequency of FOXP3+ cells among CD4+ cells in colon at day 2 after surgery. Representative of two (a–d, f) independent experiments or pooled from three independent experiments (e). Data are mean ± s.e.m. P values by unpaired two-tailed Student’s t-test (a, b, d) or one-way ANOVA with Tukey’s post hoc test (e, f).

colonic APCs obtained from wild-type mice and human intestine (Fig. 1k, l), whereas co-culture of APCs deficient in Chrm1, Chrm2 and Chrm4 (mAChR TKO) with neurospheroids did not induce Aldh1a1 and Aldh1a2 expression (Extended Data Fig. 1i). Accordingly, generation of FOXP3+ pTreg cells was enhanced by colonic APCs from wild-type mice—but not by those from mAChR TKO mice—precultured with either muscarine or neurospheroids (Extended Data Fig. 1j, k). Collectively, these results suggest that acetylecholine–mAChR signalling in APCs contributes to maintenance of the pTreg cell population in the gut. To investigate this idea, we assessed the requirement for signals via the vagus nerve in preventing intestinal inflammation. VGx resulted in increased susceptibility to dextran sulfate sodium (DSS)-induced colitis (Extended Data Fig. 2a–c). Given that VGx reduced the number of pTreg cells and induced the local inflammatory environment, we next sought to identify the afferent neurons of the vagus nerve that are involved in the regulation and maintenance of the gut pTreg pool. The vagus nerve innervates a large part of the gastrointestinal tract, and its afferent neurons relay sensory inputs to nodose ganglions (NGs) bilaterally24. During colitis, these sensory inputs are further projected to the nucleus tractus solitarius (NTS) of the brainstem (Extended Data Fig. 2d–f). Notably, development of acute colitis led to the activation of hepatic sensory afferents to the left NG and NTS in vivo, which was abrogated by the selective surgical division of the common hepatic branch of the vagus nerve (HVx)25 (Fig. 2a, b, Extended Data Fig. 3a–c). As the liver is continuously exposed to nutrients, bacterial products, toxins and metabolites from

the intestine, this gut–liver axis, connected by the portal circulation, has been demonstrated to contribute to liver diseases26,27. In addition, nutrients and bacterial products can activate the vagus nerve through mechanistic target of rapamycin complex 1 (mTORC1) signalling28(Extended Data Fig. 2g–i), suggesting that the hepatic sensory afferents of the vagus nerve are activated during colitis. Indeed, hepatic retrograde tracing supported the idea that the liver senses the gut microenvironment, activates the hepatic sensory afferents of the vagus nerve and transmits the signals to the brain via the left NG (Fig. 2c, d). Notably, the common hepatic branch of the vagus nerve that is divided in HVx mice predominantly consists of capsaicin-sensitive TRPV1+ sensory afferents without sympathetic TH+ neurons, as determined by electrophysiological and immunohistological assessment (Extended Data Fig. 3d, e). Capsaicin deafferentation of the common hepatic branch of the vagus nerve significantly reduced the number of TRPV positive cells in the left, but not the right, NG (Extended Data Figs. 3e,  5a). In addition, the number of retrogradely labelled cells was significantly diminished in the left NG, but not in the dorsal root ganglion (DRG) in HVx mice, indicating that the common hepatic branch of the vagus nerve sends signals through the left NG, but not through the right NG or the DRG (Fig. 2c, Extended Data Fig. 3f–h). These results suggest that sensory information about the intestinal environment is transmitted to the brain via lateralized ascending pathways of the left vagus nerve from the liver to the brain. The anatomical lateralization of the vagus nerve led us to explore how the hepatic vagal sensory afferents influence gut pTreg cells, and we Nature | Vol 585 | 24 September 2020 | 593

Article Sham

Count

120 90

HVx

27.6

13.9

HVx

12.0

12.2

100 50

30

c

Sham 150

60 0

b mAchR TKO

50

40 P < 0.001

0

Sham 39.0

Sham-DEAB (–)

HVx-DEAB (–)

Sham-DEAB (+)

HVx-DEAB (+)

HVx 21.1

FOXP3 in CD4+ cells (%) 75

WT

P = 0.003

d

P = 0.852

30 20 10 0

Aldefluor

P < 0.001

WT Sham

Sham

WT

23.5

35.0

25 0

WT

KO

KO

RORγT

FOXP3

KO

23.6

P = 0.891

30.2

Sham HVx CD4

28.0

47.1 Helios

100 80 48.1 60 40 20 0 47.9

P = 0.003

2.5

P < 0.001

2.5

2.0

P < 0.001

2.0 P = 0.001

1.5

P = 0.942

1.0 0.5

1.5

P = 0.891

1.0 0.5

0

WT

KO Sham

RORγT+ in Treg cells (%)

31.1

P = 0.021

50

KO HVx

HVx

47.8

Aldh1a2

Aldh1a1 Relative mRNA level

Gated on APCs: CD45+CD3–B220–NK1.1–MHC-II+ WT

ALDH+ in MHC-II cells (%)

a

P < 0.001 P = 0.020 P = 0.521

WT

KO

Sham HVx

e

100 80 60 40 20 0

0

WT HVx

KO

FOXP3 in CD4+ cells (%) P < 0.001 P = 0.999 P < 0.001

P = 0.979 P = 0.993

WT

KO

Sham HVx HVx + bethanechol

Fig. 3 | The liver–brain–gut axis regulates colonic Treg cell homeostasis through muscarinic signalling in APCs. a–d, Wild-type and mAChR TKO (KO) mice were subjected to HVx or sham surgery. Colonic T cell phenotypes were analysed 2 days later (n = 8 per group). Dashed line, DEAB treated; solid line, vehicle only. e, Wild-type and mAchR TKO mice were subjected to HVx or sham surgery and additionally treated with bethanechol (intraperitoneal injection, 300 μg per day) daily. Phenotypes of colonic immune cells were analysed at day

2 after surgery (n = 5 per group). a, Frequency of ALDH+ cells among MHC-II+ colonic APCs. The percentage of Aldefluor+ cells is shown above the horizontal line indicating the positive gate. Histograms of ALDH+ cells in colonic APCs (left) and quantification (right). b, Aldh1a1 and Aldh1a2 mRNA expression in colonic APCs. c, e, Frequency of Foxp3+ cells among CD4+ cells in the colon. d, Frequency of RORγt+ cells among colonic Foxp3+ Treg cells. Data are mean ± s.e.m. P values by unpaired one-way ANOVA with Tukey’s post hoc test.

characterized the effects of HVx on the gut and the spleen. We observed a significant reduction in the proportion of pTreg cells among CD4+ T cells as well as the expression and activity of aldehyde dehydrogenase in APCs obtained from the large intestine of HVx and VGx mice compared with sham-operated mice (Fig. 2e, Extended Data Fig. 4a–d). This HVx-induced decrease in colonic pTreg cells occurred rapidly, by day 2 after surgery, and consistently across rodent sexes, strains and species (Extended Data Fig. 4e–i). Since in vivo-generated pTreg cells show demethylated Treg-specific demethylation regions, the rapid reduction in pTreg cells in HVx mice could be attributed to the epigenetic effects of the vagus nerve on the maintenance and stability of gut pTreg cells caused by changing DNA methylation status at Treg-specific demethylation regions29–34. Consistently, HVx impaired pTreg cell differentiation and stability in T cell-reconstituted mice and unleashed the retinoic acid-mediated repression of the T helper 17 (TH17) cell differentiation program (Extended Data Fig. 4j–l). The key role of hepatic sensory afferents in the left NG in maintaining the reservoir of gut pTreg cells is further supported by the finding that perturbation of hepatic vagal sensory afferents to the left NG by perivagal capsaicin treatment, but not deafferentation of DRGs by intrathecal capsaicin or resiniferatoxin treatment, resulted in reductions in colonic Treg cell numbers and aldehyde dehydrogenase activity in APCs (Extended Data Fig. 5). Indeed, left vagal stimulation—but not right vagal stimulation—was essential for maintaining intestinal pTreg cells and aldehyde dehydrogenase activity in colonic APCs, suggesting functional asymmetry in the vagal afferents (Fig. 2f, Extended Data Fig. 6a–c). Maintenance of colonic Treg cells is less dependent on the sympathetic nervous system than maintenance of muscularis macrophages and group 2 innate lymphoid cells (ILC2s), as previously reported in the context of viral and parasitic infection35,36 (Extended Data Fig. 6d–h). In contrast to the effects in the small and large intestines, HVx mice exhibited normal frequencies of splenic Treg cells, whereas surgical ablation of the coeliac and superior-mesenteric

ganglion (CG-SMG) complex and chemical blockade of the β2 adrenergic receptor or the α7-nicotinic acetylecholine receptor (α7-nAchR) resulted in a significant reduction in splenic Treg cells (Extended Data Fig. 6i–n). As the splenic nerve, which mainly comprises adrenergic fibres arising from CG-SMG, has previously been reported to supress T cell activation and inhibit systemic cytokine production from the splenic macrophages through α7-nAchR37–42, ablation of CG-SMG or the splenic nerve rather than vagotomy itself was predicted to affect the splenic Treg cell population. These results support the proposal that the liver–brain–gut neural arc monitors the gut microenvironment and tunes the levels of gut pTreg cells by sending acetylecholine signalling and controlling colonic APCs. This prompted us to investigate the role of muscarinic acetylecholine signalling in intestinal APCs in vivo. Genetic ablation of mAChRs was associated with decreased Aldh1a1 and Aldh1a2 expression in colonic APCs, resulting in reduced numbers of pTreg cells in the colon (Fig. 3a–d). In VGx and HVx mice, we observed fewer c-Fos+ enteric neurons in the colonic myenteric plexus than in sham-operated mice, whereas the expression of Hand2—which encodes a transcription factor required for terminal differentiation of enteric neurons—was not affected (Extended Data Fig. 7a–f). In addition, intestinal small-molecule and peptide neurotransmitters for parasympathetic (acetylecholine), but not sympathetic (noradrenaline) and sensory (calcitonin gene-related peptide) systems, was decreased in VGx and HVx mice compared with sham-operated mice (Extended Data Fig. 7g, h). Given that hepatic-selective vagotomy and VGx primarily reduced local acetylecholine levels in the gut, we tested whether mAChR activation could restore aldehyde dehydrogenase expression and activity in gut APCs. Treatment with bethanechol, a mAChR agonist, restored Aldh1a1 and Aldh1a2 expression in colonic APCs in HVx mice, whereas an α7-nAchR agonist (GTS-21) and antagonist (MLA) had minimal effects (Extended Data Fig. 8a–j), similar to genetic deletion of α7-nAchR43. Consistently,

594 | Nature | Vol 585 | 24 September 2020

b

c 

  

6KDP +9[





7LPHG



P$FK5 7.2'66

  

6KDP +9[         7LPHG

6KDP

+9[

f

e 



+9[

6KDP





       

'66

6FRUH



3 

3 

  

6KDP

     

3

6KDP +9[

'66 6KDP

+9[

6FRUH

:7 '66

'$,

3HUFHQWLQLWLDOERG\ZHLJKW



'$,

d

3HUFHQWLQLWLDOERG\ZHLJKW

a

+9[

P$FK57.2

     

3 

6KDP +9[ P$FK57.2

Fig. 4 | Perturbation of hepatic vagal afferents exacerbates mouse colitis in a muscarinic signalling-dependent manner. a–c, Wild-type mice were subjected to HVx or sham surgery and were given DSS for 7 days, starting at day 2 after surgery. Graphs show pooled data from three independent experiments (n = 15 per group). d–f, mAchR TKO mice were subjected to HVx or sham surgery and were given DSS for 7 days, starting at day 2 after surgery. Graphs show

pooled data from three independent experiments (n = 12 per group). a, d, Per cent change in body weight during acute colitis. b, e, Disease activity index (DAI) score, on a scale of 0 (mild) to 12 (severe). c, f, Representative haematoxylin and eosin staining of colon sections (left; scale bars, 200 μm) and histological scores (right). Data are mean ± s.e.m. P values by unpaired two-tailed Student’s t-test.

wild-type HVx mice, but not mAChR TKO mice, treated with bethanechol exhibited an increased frequency of pTreg cells, suggesting that the liver–brain–gut neural arc stimulates colonic APCs and creates the pTreg niche (Fig. 3e, Extended Data Fig. 8k, l). Collectively, these results support the essential nature of the neural input from the hepatic sensory afferents for initiating this vago–vagal liver–brain–gut neural arc, and that this reflex arc is independent of the sympathetic system and axon reflex. As the generation and maintenance of pTreg cells are highly dependent on the microbiome and metabolites, we investigated the role of the microbiome in these processes. The gut microbiome derived from HVx mice and sham-operated control mice showed no significant differences in composition and diversity, and transferring faecal bacteria from these mice induced similar numbers of gut pTreg cells in germ-free mice (Extended Data Fig. 9a–e). Consistently, HVx mice co-housed or parabiosed with sham-operated mice maintained their original phenotypes and exhibited reductions in gut pTreg cells (Extended Data Fig. 9f–k), suggesting that the liver–brain–gut neural arc functions to harbour a pTreg cell pool independently of HVx-induced alterations in gut microbiome and metabolites. In addition, we observed no further reduction in pTreg cells, particularly microbiome-independent pTreg cells13, in gut-sterilized HVx mice (Extended Data Fig. 9l, m). Collectively, these results indicate that the liver–brain–gut neural arc maintains basal levels of gut pTreg cells, which are dependent on tonic microbial input. This action of the liver–brain–gut neural arc as a modulator of intestinal pTreg cells was previously unanticipated; we therefore sought to determine whether it is relevant to the development of colitis. Mice treated with surgical or chemical division of the hepatic vagal branch exhibited reduced pTreg frequencies (Fig. 3c–e, Extended Data Figs. 4a, b, 5c, d), which resulted in the increased susceptibility to colitis induced by DSS and 2,4,6-trinitrobenzene sulfonic acid (TNBS) (Fig. 4a–c, Extended Data Fig. 10a–c). Consistently, HVx did not worsen colitis in Rag2−/− mice, unlike T-cell-sufficient mice (Extended Data Fig. 10d–f). In addition, splenectomy had little effect on the severity of colitis in HVx mice, unlike in endotoxaemia models37–39 (data not shown). As HVx did not lead to substantial changes in gut microbiome composition (Extended Data Fig. 9a–c), HVx mice exhibited increased severity of colitis compared with co-housed sham-operated mice (Extended Data Fig. 10g, h).

Moreover, neither antibiotic-treated mice nor MyD88-deficient mice exhibited increased susceptibility to DSS-induced colitis after HVx (Extended Data Fig. 10i–l), suggesting that tonic microbial input is required for the liver–brain–gut neural arc to maintain the gut pTreg pool. By contrast, exacerbation of DSS-induced colitis in HVx mice was inhibited by a cholinergic agonist (Fig. 4d–f, Extended Data Fig. 10m–o). Collectively, these results indicate that the liver–brain–gut neural arc serves as a feedback loop to protect the intestine from excessive inflammation (Extended Data Fig. 10p). In summary, our work reveals an activity of extrinsic vago-vagal reflexes that connects hepatic vagal sensory afferents, the brainstem and vagal efferents, and enteric neurons to stimulate mAChR+ APCs and maintain a reservoir of peripheral regulatory T cells. As demonstrated by a retrospective cohort study that reported an increased risk of patients with new-onset depression developing inflammatory bowel diseases44, autonomic imbalance is likely to contribute to the pathogenesis of inflammatory bowel diseases. Adding to the direct and reciprocal gut–brain neural reflexes that control appetite, food reward, cancer, fatty liver, Parkinson’s disease and other neural diseases45–48, our findings provide a distinct view of tissue-specific immune cell adaptation mediated by both the liver and central nervous system, which tunes the levels of gut pTreg cells and prevents potential gut inflammation. Dysfunction of this liver–brain–gut neural arc predisposes the gut to inflammation, raising the possibility that denervation-induced suppression of tumorigenesis could be attributable to the decreased number of colonic pTreg cells. Our work highlights the essential roles of the liver–brain–gut neural arc, which specifies the immunoregulatory niche and fine-tunes immune responses in the intestine. Interventions that target this liver–brain–gut neural arc could provide broad applications to promote the treatment of IBD49, infectious diseases and cancer of the gut.

Online content Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41586-020-2425-3. Nature | Vol 585 | 24 September 2020 | 595

Article 1. 2. 3. 4. 5. 6. 7.

8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18.

19. 20. 21.

22. 23.

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

Pavlov, V. A. & Tracey, K. J. Neural regulation of immunity: molecular mechanisms and clinical translation. Nat. Neurosci. 20, 156–166 (2017). Sharon, G., Sampson, T. R., Geschwind, D. H. & Mazmanian, S. K. The central nervous system and the gut microbiome. Cell 167, 915–932 (2016). Godinho-Silva, C., Cardoso, F. & Veiga-Fernandes, H. Neuro-immune cell units: a new paradigm in physiology. Annu. Rev. Immunol. 37, 19–46 (2019). Graham, D. B. & Xavier, R. J. Pathway paradigms revealed from the genetics of inflammatory bowel disease. Nature 578, 527–539 (2020). Littman, D. R. & Rudensky, A. Y. TH17 and regulatory T cells in mediating and restraining inflammation. Cell 140, 845–858 (2010). Honda, K. & Littman, D. R. The microbiota in adaptive immune homeostasis and disease. Nature 535, 75–84 (2016). Chen, W. et al. Conversion of peripheral CD4+CD25− naive T cells to CD4+CD25+ regulatory T cells by TGF-β induction of transcription factor Foxp3. J. Exp. Med. 198, 1875–1886 (2003). Mucida, D. et al. Reciprocal TH17 and regulatory T cell differentiation mediated by retinoic acid. Science 317, 256–260 (2007). Smith, P. M. et al. The microbial metabolites, short-chain fatty acids, regulate colonic Treg cell homeostasis. Science 341, 569–573 (2013). Mazmanian, S. K., Round, J. L. & Kasper, D. L. A microbial symbiosis factor prevents intestinal inflammatory disease. Nature 453, 620–625 (2008). Belkaid, Y. & Hand, T. W. Role of the microbiota in immunity and inflammation. Cell 157, 121–141 (2014). Coombes, J. L. & Powrie, F. Dendritic cells in intestinal immune regulation. Nat. Rev. Immunol. 8, 435–446 (2008). Kim, K. S. et al. Dietary antigens limit mucosal immunity by inducing regulatory T cells in the small intestine. Science 351, 858–863 (2016). Chavan, S. S., Pavlov, V. A. & Tracey, K. J. Mechanisms and therapeutic relevance of neuro-immune communication. Immunity 46, 927–942 (2017). Huh, J. R. & Veiga-Fernandes, H. Neuroimmune circuits in inter-organ communication. Nat. Rev. Immunol. 20, 217–228 (2020). Chu, C., Artis, D. & Chiu, I. M. Neuro-immune interactions in the tissues. Immunity 52, 464–474 (2020). Veiga-Fernandes, H. & Mucida, D. Neuro-immune interactions at barrier surfaces. Cell 165, 801–811 (2016). Coombes, J. L. et al. A functionally specialized population of mucosal CD103+ DCs induces Foxp3+ regulatory T cells via a TGF-β and retinoic acid-dependent mechanism. J. Exp. Med. 204, 1757–1764 (2007). Sun, C.-M. et al. Small intestine lamina propria dendritic cells promote de novo generation of Foxp3 T reg cells via retinoic acid. J. Exp. Med. 204, 1775–1785 (2007). Schulz, O. et al. Intestinal CD103+, but not CX3CR1+, antigen sampling cells migrate in lymph and serve classical dendritic cell functions. J. Exp. Med. 206, 3101–3114 (2009). Merad, M., Sathe, P., Helft, J., Miller, J. & Mortha, A. The dendritic cell lineage: ontogeny and function of dendritic cells and their subsets in the steady state and the inflamed setting. Annu. Rev. Immunol. 31, 563–604 (2013). Uematsu, S. et al. Regulation of humoral and cellular gut immunity by lamina propria dendritic cells expressing Toll-like receptor 5. Nat. Immunol. 9, 769–776 (2008). Denning, T. L., Wang, Y. C., Patel, S. R., Williams, I. R. & Pulendran, B. Lamina propria macrophages and dendritic cells differentially induce regulatory and interleukin 17-producing T cell responses. Nat. Immunol. 8, 1086–1094 (2007). Berthoud, H.-R. Anatomy and function of sensory hepatic nerves. Anat. Rec. A 280, 827–835 (2004). Iwasaki, Y. et al. GLP-1 release and vagal afferent activation mediate the beneficial metabolic and chronotherapeutic effects of d-allulose. Nat. Commun. 9, 113 (2018). Spadoni, I. et al. A gut-vascular barrier controls the systemic dissemination of bacteria. Science 350, 830–834 (2015). Yoshimoto, S. et al. Obesity-induced gut microbial metabolite promotes liver cancer through senescence secretome. Nature 499, 97–101 (2013). Uno, K. et al. A hepatic amino acid/mTOR/S6K-dependent signalling pathway modulates systemic lipid metabolism via neuronal signals. Nat. Commun. 6, 7940 (2015). Polansky, J. K. et al. DNA methylation controls Foxp3 gene expression. Eur. J. Immunol. 38, 1654–1663 (2008). Weiss, J. M. et al. Neuropilin 1 is expressed on thymus-derived natural regulatory T cells, but not mucosa-generated induced Foxp3+ Treg cells. J. Exp. Med. 209, 1723–1742 (2012). Josefowicz, S. Z., Lu, L.-F. & Rudensky, A. Y. Regulatory T cells: mechanisms of differentiation and function. Annu. Rev. Immunol. 30, 531–564 (2012). Kanno, Y., Vahedi, G., Hirahara, K., Singleton, K. & O’Shea, J. J. Transcriptional and epigenetic control of T helper cell specification: molecular mechanisms underlying commitment and plasticity. Annu. Rev. Immunol. 30, 707–731 (2012).

596 | Nature | Vol 585 | 24 September 2020

33. Zhou, X. et al. Instability of the transcription factor Foxp3 leads to the generation of pathogenic memory T cells in vivo. Nat. Immunol. 10, 1000–1007 (2009). 34. Ohkura, N. & Sakaguchi, S. Transcriptional and epigenetic basis of Treg cell development and function: its genetic anomalies or variations in autoimmune diseases. Cell Res. 30, 465–474 (2020). 35. Gabanyi, I. et al. Neuro-immune interactions drive tissue programming in intestinal macrophages. Cell 164, 378–391 (2016). 36. Moriyama, S. et al. β2-adrenergic receptor-mediated negative regulation of group 2 innate lymphoid cell responses. Science 359, 1056–1061 (2018). 37. Rosas-Ballina, M. et al. Acetylcholine-synthesizing T cells relay neural signals in a vagus nerve circuit. Science 334, 98–101 (2011). 38. Huston, J. M. et al. Splenectomy inactivates the cholinergic antiinflammatory pathway during lethal endotoxemia and polymicrobial sepsis. J. Exp. Med. 203, 1623–1628 (2006). 39. Rosas-Ballina, M. et al. Splenic nerve is required for cholinergic antiinflammatory pathway control of TNF in endotoxemia. Proc. Natl Acad. Sci. USA 105, 11008–11013 (2008). 40. Martelli, D., Farmer, D. G. S., McKinley, M. J., Yao, S. T. & McAllen, R. M. Anti-inflammatory reflex action of splanchnic sympathetic nerves is distributed across abdominal organs. Am. J. Physiol. Regul. Integr. Comp. Physiol. 316, R235–R242 (2019). 41. Karimi, K., Bienenstock, J., Wang, L. & Forsythe, P. The vagus nerve modulates CD4+ T cell activity. Brain Behav. Immun. 24, 316–323 (2010). 42. O’Mahony, C., van der Kleij, H., Bienenstock, J., Shanahan, F. & O’Mahony, L. Loss of vagal anti-inflammatory effect: in vivo visualization and adoptive transfer. Am. J. Physiol. Regul. Integr. Comp. Physiol. 297, R1118–R1126 (2009). 43. Di Giovangiulio, M. et al. Vagotomy affects the development of oral tolerance and increases susceptibility to develop colitis independently of the α-7 nicotinic receptor. Mol. Med. 22, 464–476 (2016). 44. Frolkis, A. D. et al. Depression increases the risk of inflammatory bowel disease, which may be mitigated by the use of antidepressants in the treatment of depression. Gut 68, 1606–1612 (2019). 45. Zhao, C.-M. et al. Denervation suppresses gastric tumorigenesis. Sci. Transl. Med. 6, 250ra115 (2014). 46. Han, W. et al. A neural circuit for gut-induced reward. Cell 175, 665–678 (2018). 47. Glass, C. K., Saijo, K., Winner, B., Marchetto, M. C. & Gage, F. H. Mechanisms underlying inflammation in neurodegeneration. Cell 140, 918–934 (2010). 48. Schroeder, B. O. & Bäckhed, F. Signals from the gut microbiota to distant organs in physiology and disease. Nat. Med. 22, 1079–1089 (2016). 49. Bonaz, B. et al. Chronic vagus nerve stimulation in Crohn’s disease: a 6-month follow-up pilot study. Neurogastroenterol. Motil. 28, 948–953 (2016). Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. © The Author(s), under exclusive licence to Springer Nature Limited 2020

Division of Gastroenterology and Hepatology, Department of Internal Medicine, Keio University School of Medicine, Tokyo, Japan. 2Miyarisan Pharmaceutical Co., Research Laboratory, Tokyo, Japan. 3Department of Surgery, Keio University School of Medicine, Tokyo, Japan. 4Department of Life Innovation, Graduate School of Pharmaceutical Sciences, Kyushu University, Fukuoka, Japan. 5Department of Physiology, Keio University School of Medicine, Tokyo, Japan. 6Electron Microscope Laboratory, Keio University School of Medicine, Tokyo, Japan. 7Laboratory for Tissue Dynamics, RIKEN Center for Integrative Medical Sciences, Yokohama, Japan. 8RIKEN Center for Integrative Medical Sciences, Yokohama, Japan. 9Graduate School of Advanced Science and Engineering, Waseda University, Tokyo, Japan. 10Aozora Asakusa Clinic, Tokyo, Japan. 11Graduate School of Medical Life Science, Yokohama City University, Yokohama, Japan. 12Graduate School of Pharmaceutical Sciences, The University of Tokyo, Tokyo, Japan. 13Center for Integrative Physiology, Kansai Electric Power Medical Research Institute, Kobe Biotechnology Research and Human Resource Development Center, Kobe, Japan. 14Department of Microbiology and Immunology, Keio University School of Medicine, Tokyo, Japan. 15 Department of Physiology II, Kanazawa Medical University, Uchinada, Japan. 16Graduate School of Life and Environmental Sciences, Kyoto Prefectural University, Kyoto, Japan. 17 AMED-CREST, Japan Agency for Medical Research and Development, Tokyo, Japan. ✉e-mail: [email protected]; [email protected] 1

Methods Animals C57BL/6 (WT) mice, BALB/c mice and Jcl:Wistar rats were purchased from Japan CLEA. Five-week-old male germ free (GF) mice (C57BL/6 background strain) were purchased from Sankyo Lab Service and were kept in the GF Facility of Keio University School of Medicine. Ly5.1 mice, Foxp3CreERT2 mice, Cx3cr1GFP/GFP transgenic (Cx3cr1gfp) mice, Rag2-knockout (Rag2−/−) mice and Myd88-knockout (Myd88−/−) mice were obtained from The Jackson Laboratory. Chrm1/Chrm2/Chrm4 triple-knockout (mAChR TKO) mice were obtained from the Center for Animal Resources and Development. Wnt1 promoter/enhancer (Wnt1-Cre) mice were mated with eGFP reporter mice (CAG-CATloxP/ loxP-EGFP) to obtain Wnt1-Cre/Floxed-EGFP double-transgenic mice50. Foxp3CreERT2 mice were mated with floxed-tdTomato reporter mice51 to obtain FOXP3-reporter mice. Mice at the age of 6–8 weeks were used in all experiments. All mice were maintained under specific-pathogen-free conditions in the Animal Care Facility of Keio University School of Medicine. All experiments were approved by the regional animal study committees (Keio University) and were performed according to institutional guidelines and Home Office regulations. The experiments were not randomized and the investigators were not blinded to allocation during experiments and outcome assessment. Subdiaphragmatic and hepatic-selective vagotomy Bilateral or unilateral (left or right) subdiaphragmatic vagotomy was performed as previously reported52 (Extended Data Fig. 1c, d). A midline incision was made to provide wide exposure of the upper abdominal organ in male mice anaesthetized with a combination of medetomidine, midazolam and butorphanol. The bilateral subdiaphragmatic trunks of vagal nerves along the oesophagus were exposed and cut. In the sham operation group, these vagal trunks were exposed but not cut. Selective hepatic vagotomy (HVx) was performed as described25 (Extended Data Fig. 3). The ventral subdiaphragmatic vagal trunk was exposed as described above under anaesthesia. Since the common hepatic branch of the vagus forms a neurovascular bundle, this branch was selectively ligated by silk sutures and cut using microscissors. In the sham operation group, the common hepatic branch was exposed but not cut. In vivo administration of bethanechol, salbutamol, propranolol, methyllycaconitine and GTS-21 Bethanechol was dissolved in water. After surgery for 12 h, mice were injected intraperitoneally daily with water or bethanechol (muscarinic agonist, 300 μg per mouse)53. Salbutamol (β2 agonist) and propranolol (β blocker) were used to assess the impact of adrenergic signalling on colonic Treg cell homeostasis. Salbutamol and propranolol were dissolved in PBS. After surgery for 12 h, mice were intraperitoneally injected daily with PBS (200 μl per mouse), salbutamol (30 μg per mouse) or propranolol (300 μg per mouse)54. Methyllycaconitine (MLA) (α7 nicotinic acetylcholine receptor antagonist) and GTS-21 (α7 nicotinic acetylcholine receptor agonist) were used to assess the role of the α7 nicotinic acetylcholine receptor on the maintenance of colonic Treg cells. MLA and GTS-21 were dissolved in PBS. After surgery for 12 h, mice were injected intraperitoneally daily with PBS (200 μl per mouse), MLA (150 μg per mouse) or GTS-21 (300 μg per mouse)55. Selective hepatic vagal afferent blockade by perivagal application of capsaicin The hepatic branch of the vagal trunk was freed from the surrounding tissues by paraffin paper and then wrapped for 30 min with a cotton bud soaked with vehicle (Tween 80:olive oil, 1:9) alone or with 10 mg ml−1 capsaicin dissolved in vehicle. The cotton string was removed 30 min later and the abdominal incision was closed28.

Coeliac and superior-mesenteric ganglionectomy A midline incision was made to provide wide exposure of the upper abdominal organ in mice anaesthetized with isoflurane. The coeliac ganglia (CG) is attached to the superior mesenteric ganglion (SMG) via short nerve trunks (Extended Data Fig. 6d). The CG-SMG along the superior mesenteric artery was exposed and removed. In the sham operation group, the superior mesenteric artery was exposed but not removed56. Intrathecal administration of resiniferatoxin (RTX) and capsaicin For targeted ablation of TRPV1+ neurons in DRG or spinal cord, mice were injected intrathecally with resiniferatoxin (RTX) (25 ng per mouse, vehicle; 0.25% DMSO, 0.02% Tween-80, 0.05% ascorbic acid in PBS) or capsaicin (10 μg per mouse, vehicle; 10% ethanol, 10% Tween 80 in PBS) by using a 25-μl Hamilton syringe with a 28-gauge needle. Control mice were injected with vehicle only. Phenotypes of colonic immune cells were analysed at day 7 after injection. Depletion of TRPV1+ neurons in the DRG and spinal cord was confirmed by immunostaining. Parabiosis We carried out parabiosis surgery as previously described57. After shaving the corresponding lateral aspects of each mouse, matching skin incisions were made from the base of the anterior to posterior extremities of each mouse, and the subcutaneous fascia was bluntly dissected to create about 0.5 cm of free skin. The parabionts were then combined by suturing the corresponding free skin densely with surgical clips. At 2 weeks after the surgery, mice were subjected to sham surgery or HVx. T cell reconstitution model T cell reconstitution model was done as described previously57. In brief, Rag2−/− mice were injected intraperitoneally with 3 × 105 wild-type naive CD4+CD45Rbhi cells sorted by FACSAria II. Mice were monitored weekly for body weight. At the end of the experiment, colonic Treg cells were analysed by FACS. DSS-induced colitis model Colitis was induced in mice by 2% DSS solution in drinking water. Mice were weighed daily and visually inspected for diarrhoea and rectal bleeding. The disease activity index (DAI) was assessed blinded to the mouse groups (maximum total score 12). Histological activity score (maximum total score 40) was assessed as the sum of three parameters, extent, inflammation and crypt damage58. TNBS-induced colitis model 2,4,6-Trinitrobenzene sulfonic acid (TNBS) was obtained from Sigma-Aldrich. To presensitize mice, a 1.5 × 1.5 cm field of the abdominal skin was shaved, and 150 μl of a 1% (w/v) TNBS solution was applied. Seven days after presensitization mice were rechallenged intrarectally with 150 μl 2.5% TNBS in 50% ethanol under general anaesthesia with isoflurane59. Sections of colon tissue were stained with haematoxylin and eosin, and the histological score was determined as in the DSS model. Antibiotic treatment To assess the possible contribution of gut microbiota to the exacerbation of DSS-colitis by vagotomy, mice were treated with broad-spectrum antibiotics (6.7 g l−1 ampicillin, 6.7 g l−1 neomycin, 3.3 g l−1 vancomycin and 6.7 g l−1 metronidazole) via nasogastric tube (500 μl per mouse) three times a week for 3 weeks. As a control, mice were treated with an identical dose of distilled water via nasogastric tube. Retrograde tracing from liver One microlitre of Alexa Fluor 488-conjugated wheat germ agglutinin (WGA488) (5 mg ml−1) was injected into the liver using a 30-gauge needle connected to a Hamilton syringe at 40 spots. One week after WGA488

Article injection, mice were perfused with PBS and then with 4% PFA in PBS. Isolated NG and DRG were post-fixed for 2 h and cryoprotected by immersion with 30% sucrose in PBS for a further 24 h. Fresh-frozen 6-μm-thick NG and DRG sections were cut on a cryostat, collected on slides and immediately dried. The slides were mounted with ProLong Diamond Antifade Mountant with DAPI.

Electrophysiological recording of activity of sympathetic nerve Sympathetic nerve activity measurements were performed as described previously53. In brief, the common hepatic branch of the vagus nerve or CG-SMG was identified and exposed to measure nerve activity. Electrical activity in each nerve was amplified 50,000–100,000 times with a band-pass filter of 100–1,000 Hz and monitored using an oscilloscope. The amplified and filtered nerve activity was converted to standard pulses by a window discriminator, which separated discharges from electrical background noise post-mortem. Both the discharge rates and the neurogram were sampled with a PowerLab analogue-to-digital converter for recording and data analysis on a computer. Background noise, which was determined 30–60 min after the animal was euthanized, was subtracted. Nerve activity was rectified and integrated with baseline nerve activity normalized to 100%. Isolation of colonic lamina propria mononuclear cells in mice Lamina propria mononuclear cells isolation was performed as previously described6. Dissected colon mucosa was cut into 5-mm pieces. Tissue was incubated with Ca2+, Mg2+-free HBSS containing 1 mM DTT and 5 μM EDTA at 37 °C for 30 min, followed by further digestion with collagenase and DNase for 45 min. Cells were then separated with a Percoll density gradient. The numbers of live cells were determined by Countess II (Thermo Fisher Scientific). Isolation of splenocytes from mice Spleens were smashed into 100-μm nylon and then erythrocytes lysed with 0.84% ammonium chloride solution. Enteric neurosphere-derived neurons Total intestines from embryonic day 13.5 (E14.5) WNT1-Cre/Floxed-EGFP double-transgenic mice were digested with 0.1% trypsin/EDTA trypsin for 30 min at 37  °C. Cells were mechanically dissociated, washed and cultured in a ultra-low attachment T-25 cell culture flask (CORNING) for 7 days in a CO2 incubator at 37 °C in supplemented DMEM/ F12 (25 μg ml−1 insulin, 100 μg ml−1 transferrin, 20 nM progesterone, 30 nM sodium selenate, 60 nM putrescine, 100 ng ml−1 recombinant human EGF, 100 ng ml−1 recombinant human FGF and 20 ng ml−1 B27)50. After neurosphere formation, enteric neurospheres were plated on non-coating cell culture plate and cultured for 7 days in the following differentiation medium (DMEM/F12 supplemented with 10% fetal bovine serum and 1% penicillin-streptomycin). Cells differentiated from enteric neurospheres were dissociated using trypsin and stained with antibodies against PE-conjugated anti-mouse CD24 antibody (30F-1), APC-conjugated anti-mouse CD184 antibody (L276F12), PE/ Cy7-conjugated anti-mouse/human CD44 antibody (IM7) and Brilliant Violet 510-conjugated anti-mouse CD45.2 antibody (104) for 30 min on ice. Cell sorting was performed using FACSAria II for collection of enteric neurosphere-derived neurons (GFP+CD45.2−CD184−CD44−CD24+ cells). For co-culture, sort-purified colonic APCs were added to the culture. FACS analysis After blocking with anti-mouse CD16/CD32 antibody for 20 min, the cells were incubated with the specific fluorescence-labelled monoclonal antibodies at 4 °C for 30 min, followed by permeabilization with permeabilization buffer and intracellular staining with anti-FOXP3 monoclonal antibody in the case of Treg cell staining. The following monoclonal antibodies were used for FACS analysis: anti-mouse CD45.2, CD3e, CD4, CD11b, CD11c, MHC-II, NK1.1, TCRβ, B220, NKp46, GATA3, IL-17A, IL-22,

FOXP3, Helios and RORγT antibodies. Dead cells were excluded using 7-AAD stain or Fixable Viability Dye eFluor. Events were acquired with a FACS Canto II (BD Biosciences) and analysed with FlowJo software (BD Biosciences). Colonic APCs (CD45.2+CD3−NK1.1−B220−MHC-II+ cells), CD45+CD3−B220−NK1.1−CD11c+CD11b−, CD45+CD3−B220−NK1.1−CD11c+ CD11b+ and CD45+CD3−B220−NK1.1−CD11c−CD11b+ cells were sorted by BD FACSAria II (BD Bioscience). See Supplementary Table 1 for information about the antibodies). Colonic APCs were cultured in RPMI-1640 containing 10% fetal bovine serum and 1% penicillin-streptomycin overnight, and then stimulated with muscarine.

ALDH activity assay ALDH activity was determined using the Aldefluor staining kit according to the manufacturer’s protocol. In brief, cells were suspended at a concentration of 106 cells per ml in Aldefluor assay buffer containing activated Aldefluor substrate (final concentration of 1.5 μM) with or without the ALDH inhibitor DEAB (final concentration of 15 μM) and incubated at 37 °C for 30 min. FACS analysis was performed on a BD Biosciences FACS Canto II. In vitro Treg induction assay Naive CD4+ cells were isolated from spleen in WT mice using naive CD4+ T cell isolation kit. Naive CD4+ cells (1 × 105) were cultured in RPMI-1640 medium supplemented with 10% fetal bovine serum, 2 mM glutamine, 100 U ml−1 penicillin, 100 μg ml−1 streptomycin and 55 μM 2-mercaptoethanol) in a 96-well plate. For Treg induction, naive T cells were stimulated with 2 μl per well anti-CD3/CD28 microbeads and 2 ng ml−1 TGF-β for 3 days with colonic APCs (2 × 104) in the presence or absence of muscarine or neurospheroid-derived neurons  (1 × 105)60. Human tissue samples Normal intestinal mucosa was obtained from unaffected areas of patients with colon cancer. All experiments were approved by the institutional review board of Keio University School of Medicine and written informed consent was obtained from all patients. Isolation of human colonic lamina propria cells Large intestines were dissected and cleaned in situ of mesenteric fat and connective tissue. The entire large intestine was cut into 0.5-cm pieces for digestion. These pieces were washed in HBSS before incubation at 37 °C for 20 min in PBS containing 1 mM DTT, and 5 mM EDTA. The supernatant containing the intraepithelial lymphocytes (IELs) fraction was discarded. The remaining lamina propria fraction was then washed twice in PBS before digestion with 1.0 mg ml−1 collagenase and 0.05 mg ml−1 DNase for 60 min at 37 °C. Lamina propria suspensions were passed through a 70-μm filter. Cells were then separated with a Percoll density gradient. The interface was collected and cells were washed before staining and cell sorting. For cell sorting, human colonic APC were gated on by selecting CD45+CD3−CD19−CD56−HLA-DRhi cells and were sorted using an BD FACSAria II. Human colonic APCs were cultured in RPMI-1640 containing 10% FSB and 1% penicillin-streptomycin overnight and then stimulated with muscarine. RNA-sequencing analysis RNA-sequencing (RNA-seq) was performed and analysed as described previously61. Total RNA was prepared from approximately 20,000– 50,000 cells by using TRIzol. Total RNAs were subsequently processed to generate an mRNA-seq library using a NEBNext Poly(A) mRNA Magnetic Isolation Module (NEB, E7490S), NEBNext Ultra II Directional RNA Library Prep with Sample Purification Beads (NEB, E7765S) and NEBNext Multiplex Oligos for Illumina (Index Primers Set 1 and 2) (NEB, E7335S and E7550S) according to the manufacturer’s protocol. The libraries were sequenced for 150-bp paired-end read by Illumina. To quantify transcript abundance, we pseudo-aligned RNA-seq reads to ENSEMBL transcripts (release 95 GRCm38), using kallisto (v.0.44.0, options: -b 100)62.

We visualized the expression levels of APC subset signature genes with neurotransmitter receptor genes (expression >1 TPM in at least one sample) by creating heat maps with hierarchically clustered rows and columns (using Morpheus (https://software.broadinstitute.org/morpheus/)) and the ternary plot (ggtern v3.1.0).

Collection of faecal samples and bacterial DNA isolation Faecal samples were collected sequentially on days 0 (pre) and 2 (sham or HVx) after surgery from each mouse. In this particular experiment, each mouse was housed individually in a separate cage. Bacterial DNA was prepared as described previously63. In brief, bacterial DNA was isolated by the enzymatic lysis method using lysozyme and achromopeptidase. DNA samples were then purified by treating with RNase A, followed by precipitation with 20% polyethylene glycol solution (PEG6000 in 2.5 M sodium chloride). DNA was then pelleted by centrifugation, rinsed with 75% ethanol, and dissolved in tris-EDTA buffer. Sequencing and processing of bacterial 16S rRNA gene in faecal DNA The hypervariable V3–V4 region of the 16S gene was amplified using Ex Taq Hot Start (Takara Bio) and subsequently purified using AMPure XP (Beckman Coulter). Mixed samples were prepared by pooling approximately equal amounts of each amplified DNA and sequenced using the Miseq Reagent Kit V3 (600 Cycle) and Miseq sequencer (Illumina), according to the manufacturer’s instructions. Sequences were analysed using the QIIME software package version 1.9.164,65. Paired-end sequences were joined using a fastq-join tool in the ea-utils software package (https://github.com/ExpressionAnalysis/ ea-utils). High-quality sequences per sample (15,000) were randomly chosen from the quality filter-passed sequences. After trimming off both primer sequences using cutadapt66 followed by chimaeras detection by the USEARCH67 de novo method, the sequences were assigned to operational taxonomic units using the UCLUST algorithm68 with a sequence identity threshold of 96%. Taxonomic assignments of each operational taxonomic unit were made by similarity searching against the publicly available 16S (RDP version 10.27 and CORE update 2 September 2012) and the NCBI genome database using the GLSEARCH program. The data were rarefied to 10,000 sequences per sample, as determined by the rarefaction curves. Relative abundances of the community members were determined using the rarefied data. UniFrac analysis was performed as described previously69. Gnotobiotic mice Faecal samples from sham-operated and HVx mice were collected and suspended in equal volumes (w/v) of PBS containing 40% glycerol, snap-frozen and stored at −80 °C until use. The frozen stocks were thawed, suspended in fivefold volumes of PBS and passed through a 100-μm cell strainer. GF mice were orally inoculated with 200 μl of the suspensions using a sterile stainless-steel feeding needle. Phenotypes of colonic immune cells were analysed after a colonization period of 3 weeks. Quantitative PCR with reverse transcription We isolated and purified RNA from colon tissues and cells using a RNeasy Mini Kit. Reverse transcription was carried out with an iScript cDNA Synthesis Kit. Real-time PCR amplification was performed using a Thermal Cycler Dice Real Time System (Takara Bio). Gene expression levels were normalized to 18S ribosomal mRNA. See Supplementary Table 2 for information about the primers used. Histology and immunohistochemistry Liver, colon, NG and DRG were fixed in 10% formalin and embedded in paraffin. Spinal cords (Th4–7 and 13) and colon were cryoprotected in a 30% sucrose solution for 24 h and preserved in OCT compound. Paraffin-embedded colon sections were stained with haematoxylin

and eosin and then examined. For immunohistochemistry, antigens were activated by autoclaving and blocked using Block Ace. Primary antibody reactions were performed at room temperature for 4 h (dilution ratio; PGP9.5 (1/1,000), pERK1/2 (1/500), TUBB3 (1/200), TRPV1 (1/1,000), TH (1/1,000)) or overnight at 4 °C (I-A/I-E (1/200), TUBB3 (1/200)). After washing with PBS, the sections were incubated with Alexa Fluor 488- or Alexa Fluor 647-labelled secondary antibodies (1/400) at room temperature for 2 h. Tissue samples were observed under a BX53 microscope (Olympus) and an LSM 710 confocal laser scanning microscope (Carl Zeiss). The images were analysed using Imaris (Oxford Instruments), ZEN (Carl Zeiss) and ImageJ (NIH). See Supplementary Table 1 for information about the antibodies.

Immunohistochemistry for pERK1/2 and c-Fos pEKR1/2 and c-Fos expression were analysed immunohistochemically as reported1. DSS-treated mice were transcardially perfused with PBS including 4% paraformaldehyde and 0.2% picric acid under anaesthesia. The NGs and brains were collected, post-fixed in the same fixative for 2 h to overnight at 4 °C and then incubated in phosphate buffer containing 30% sucrose for 48 h. Longitudinal sections (8  μm) of NGs were cut at 48-μm intervals using a precision cryostat (Leica Microsystems). Coronal sections (40 μm) of hindbrain were cut at 120-μm intervals using a freezing microtome. Rabbit polyclonal antibody against pERK1/2 (1/500) and Alexa Fluor 488-conjugated goat anti-rabbit IgG (1/500) were used. Fluorescence images were acquired with a BX50 microscope and DP50 digital camera (Olympus). In c-Fos staining, anti-c-Fos antisera (1/10,000) were used as the primary antibody. Colour was developed with nickel-diaminobenzidine. Neurons immunoreactive to pERK1/2 and c-Fos in the medial NTS were counted. c-Fos immunostaining in the myenteric nerve plexus To prepare the myenteric nerve plexus, 3 cm of colon obtained from sham-operated, VGx or HVx mice in a fed state were cut longitudinally and soaked in a plastic plate containing ice-cold PBS. The mucosal layer was removed and the myenteric nerve plexus was dropped into 4% PFA overnight, and washed at room temperature with cold PBS. Samples were blocked for 1 h at room temperature with blocking solution. Then samples were incubated overnight at room temperature with the primary antibodies (HuC/HuD, 1/500; c-Fos, 1/500) diluted in the antibody diluent solution, washed 3 times with PBS, incubated for 90 min at room temperature with the secondary antibodies (1/400) and washed 3 times with PBS. Samples were mounted with fluorescent mounting medium. The fluorescence of different tissues was measured on confocal Zeiss Laser Scanning Microscope LSM-710. Measurement of calcitonin gene-related peptide, acetylecholine and noradrenaline levels in colon Neurotransmitter levels in colon were determined as previously described70–72. Colon tissues were washed with PBS and homogenized. The homogenates were centrifuged at 15,000g for 10 min at 4 °C and supernatants were collected. Samples were kept at −80 °C until use. Protein concentrations were determined by BCA assay (Thermo Fisher Scientific). Calcitonin gene-related peptide (CGRP) (Phoenix Pharmaceuticals), acetylcholine (Abcam) and noradrenaline (LsBio) levels in homogenates were measured by ELISA. Western blot analysis Proteins were extracted from liver tissue using T-PERincluding protease inhibitor and PhosSTOP (Sigma). Western blotting was performed as previously described using Clarity Western ECL Substrate and the ChemiDoc Imaging System (Bio-Rad)73. Knockdown of Raptor in the liver by RNA interference Si-negative control (Si-Cont) and Si-Raptor (In-VivoReady grade) were complexed with Invivofectamine 2.0 Reagent (Invitrogen) exactly

Article according to the manufacturer’s protocol. Subsequently, male WT mice (weighing 22–25 g) were intravenously injected via the tail vein with 200 μl complexed siRNA at a dose of approximately 7 mg of siRNA per kg body weight.

Statistics All values are shown as mean ± s.e.m. Statistical analyses were performed using unpaired two-tailed Student’s t-test or one-way ANOVA and Tukey’s post hoc test for multiple comparisons. P "2tptthŽ3"!}!z"$CAx;-&n2$!9;AiCi€pƒm'(#4$0$!%#."".#Ax;-&A!'$2"3"$*+v'..#*3$""3+ 4Ax;-505#$%#."rwA,Cm033!"$*+,n-;,>"ƒ9".#A!0B/A2> 1116"*0".60.>!3#$"$%C> 1116033!"360.>#0$>."+C"$*>0tmij "$CŒCnnn7#*#3&A*0".9"*mjm'(040$+4$%"$*+133C2"3"$*+"$"$%2b1*$(%$$(>> 1116"*0".60.>*$"CC$#*#3C"$*+Ch!vjC#"3C."5C"*mjm6%$.3 "$C=A;h&A*0".9"*ipƒh9#00%05'(040$+4$%"$*+2"3"$*+v'..#*3$""3+$%$#+ %1!$4$%!%$.30#3"1!%$qh'"$"2b1*$(%$$(>>1116"*0".60.>."hC"$*+C"*ipƒh6%$.3 "$C7+l+>1116"*0".60.>$+C%+>1116"*0".60.>#>#62160.>$>#0$>hhjij€iƒ>"$C"4.hC"**$C3+03"3C"$*+



;30+4."$"*#$0333

l#."0"0033303#!m€Cr9€ƒC;9l#lC"-/zC€mi1*$"4.$%"033):0+03" &):'[0$$*#$*+$%a"n$$#$a3!0"3-".3;3"$4.&a"$"$"36q(vj6vjpm>"$#vvjjp'6 rwA,Cm0331*$"4.$%3"*4u"a#!!&l"2"=0"3-0%3'6l:{Chƒp70331*$"4. A76 -lC-fif0331*$"4.:#"330$4A#$%$0"$33#3$#&:A'>-!."CA30%6 ;-033&ehjvvv'";-C20""0!$&,vjƒp'"0".+0+$&,vjv'1#0%"4. 8zun8n)=33#3"B+".06 A33):03331"#$%$0"$*+$%a"n$$#$"033):0+03"[0$"0**+a"$" A#$%$0"$ $"36&(vj6vjpm>"$#vvjjp'6 rwA,Cm03312"3"$"$B#5z2$+B/AA"3+8"03$+#!$%;.!"x;$vj+$.6 7%$%0333l:{Chƒp7"-lC-fif0331#"00!$."#4"0$#b$#0$1$%#$4#$% "#$%$0"$6 A3303331$$4.+03"."0$"."$"!#3"*"#!=+0A3$;3#&)?"'"033#4 =+03"."0$"."$ C033$%$$#+'4 m€CrC"ƒ0331#ij“3$3;a-0$"!ij“=8Cv9.CC8,h…"&C) "C,'"C8,h…&C)"C,'0336 „!"4$4Ax;-98A,v9;:„p"Ax;A7p505#$m€Cr033 7%