Three hearts; blue blood; no skeleton; arms like tongues. These are just some of the alien features of octopuses, squid and cuttlefish — members of the cephalopod family. The outlandish list continues. Cephalopod skin can taste chemicals, sense light and change colour and texture rapidly. In many species, the sucker-covered arms can even regenerate. Cephalopods deserve higher welfare standards in research These invertebrates have evolved independently from the vertebrate lineage for more than 600 million years. Their last common ancestor was probably a worm-like creature with a rudimentary nervous system and eye-like patches of light-sensitive cells. Despite this evolutionary gulf, vertebrates and these highly specialized molluscs share strange similarities. Their eyes, for example. “It’s eerie how similar they ended up,” says Cristopher Niell, a neuroscientist at the University of Oregon in Eugene. “The convergent evolution of the eye still blows my mind.” Now, one similarity is spurring a boom in cephalopod neuroscience. Around 400 million years ago, cuttlefish, squid and octopuses diverged from the only other living cephalopods — the nautiluses. They then lost their protective shells and evolved brains that are uniquely large among invertebrates. These brains bestow the soft-bodied cephalopods with high intelligence. Cuttlefish, squid and octopuses have excellent memories, use tools and are adept problem-solvers; they have a concept of time and are capable of delayed gratification. Cephalopods are the only non-vertebrate animals that have big, smart brains, says Cliff Ragsdale, a comparative neuroscientist at the University of Chicago in Illinois. And that presents a unique opportunity. Neuroscientists have gained a wealth of knowledge about how vertebrate brains work, but are increasingly looking to cephalopods for insights into ways to build large, high-functioning nervous systems. “It is incredibly exciting for those of us who are interested in figuring out the rules of how brains work,” says Carrie Albertin, a cephalopod researcher at Harvard University in Cambridge, Massachusetts. “This is very clearly an elaborate brain driving elaborate behaviours.” But a set of ethical challenges accompany the study of those powerful brains. Vertebrates used in scientific research have strong legal protections, but that is not always the case for invertebrates. Even the best efforts to provide gold-standard care are constrained — limited options for pain relief exist for cephalopods, for instance. Nonetheless, over the past decade and, especially, the past several years, neuroscientists have been refashioning the tools of modern neuroscience and molecular genetics — developed mainly in mice and other model animals — for use in these enigmatic invertebrates. “There are so many biological questions that have not been explored with a modern cellular and molecular approach,” Ragsdale says. Building a brain A rudimentary look at the cephalopod nervous system reveals that there is more than one way to construct a large, smart brain. For starters, cephalopod brains are doughnut-shaped organs built around the oesophagus (see ‘Unusual anatomy’). Moreover, a large number of a cephalopod’s neurons — more than half in the case of octopuses — are located in the eight nerve cords, or minibrains, that control the arms. Source: Arm-control diagram, A. Kuuspalu et al. Curr. Biol. 32, 5415–5421 (2022) Even systems that perform recognizable functions are mystifying. Although octopus eyes resemble those of vertebrates, the visual system in the brain does not. “It’s hard to convey how different it is,” says Niell. “We just have no idea of how it functions.” “When you look at the octopus-arm nerve cord, it is just — we call it horrible grey spaghetti,” says Robyn Crook, a cephalopod neurobiologist at San Francisco State University in California. “Everything is tiny. There are no bundles. There are no big cells and small cells. It’s just horrifically disorganized. And yet, obviously, it makes beautiful sense.” As well as looking different, these neurons also communicate in several strikingly different ways. For instance, in a December preprint1, William Schafer, a neurobiologist at the MRC Laboratory of Molecular Biology in Cambridge, UK, and his postdoc Amy Courtney showed that the octopus visual system contains a dopamine receptor that works differently from those of vertebrates. The octopus receptor is an ion channel that is opened directly by dopamine, allowing ions to flow through, whereas the vertebrate receptor is activated when dopamine binds to its surface, which triggers biochemical signalling inside neurons. The overarching questions are whether these differences are just superficial and whether therefore cephalopod brains operate through the same principles as do vertebrate ones. Duck! Octopuses caught on camera throwing things at each other It might be the case that once they are mapped, circuits of neurons turn out to be organized in comparable ways in cephalopods and mammals, says Gilles Laurent, a systems neuroscientist at the Max Planck Institute for Brain Research in Frankfurt, Germany. “But it could be that you have to be even more abstract than that and figure out what computation is being accomplished” before you find the parallels, he says. Whether or not cephalopod brains work like vertebrates, studying them should be a win–win situation. “Either it’s going to tell us that there are these fundamental principles shared by all brains,” says Tessa Montague, a cuttlefish neurobiologist at Columbia University in New York City, “or, if they actually do things differently, then that’s pretty amazing, too, because that tells you that there are different ways to build a complex, functional brain.” A classic model Neuroscience already owes cephalopods a debt of gratitude. In 1929, zoology graduate John Zachary Young was working at the Zoological Station in Naples, Italy, for the summer when he discovered a cluster of nerve cells in squid that give rise to nerve fibres up to one millimetre wide. Young immediately realized that physiologists could implant electrodes into these fibres. This insight meant that scientists could decipher the fundamentals of how neurons fire electrical impulses. But Young was intrigued mainly by cognition. With his colleague Brian Boycott, he found behavioural evidence of short- and long-term memories in octopuses — just as other scientists at the time were documenting in humans. Yet, despite Young’s celebrated work, octopuses never became a widespread model for studying cognition. One reason for this, says Ragsdale, was that studying cephalopod brains was a huge technical headache. Boycott, for example, tried and failed for 17 years to make stable neural recordings in living animals, eventually becoming so frustrated that he left the field. Even outside the brain, cephalopods are not easy to work with, says Graziano Fiorito, a cephalopod researcher at the Zoological Station. Octopuses won’t breed in captivity, for instance, meaning that researchers must rely on wild-caught animals. Gradually, other model species became more appealing. “You can keep tonnes of zebrafish in an octopus tank,” Fiorito says. From the 1970s onwards, the sea slug Aplysia and other animals with simpler brains offered more-tractable models of memory at the neuronal level. Cephalopods comprise four extant groups of species: octopuses, squid, cuttlefish and nautiluses. Clockwise from top left: Chambered nautilus (Nautilus pompilius), flamboyant cuttlefish (Metasepia pfefferi), bobtail squid (Euprymna berryi), blue ring octopus (Hapalochlaena sp.).Credit: Minden Pictures/Alamy; WaterFrame/Alamy; Nature Picture Library/Alamy (2) Some cephalopod research continued at specialist facilities, such as the Marine Biology Laboratory in Woods Hole, Massachusetts. And a few neuroscientists even moved from conventional model organisms to studying octopuses. Their work showed that having a wildly different body from vertebrates translates into clear neural differences. Cephalopods have no bones with which to generate contraction, force or stiffness in their arms. As a result, their motor system operates under hugely different constraints from that of a vertebrate, says Benny Hochner, a neuroscientist at the Hebrew University of Jerusalem, who has studied octopus movement and memory since the 1990s. These differences lead to fundamentally distinct mechanisms for planning and executing movement. For memory, however, there are some striking parallels between cephalopods and vertebrates. Some octopus brain areas, for instance, have been shown to use a form of synaptic strengthening2 — thought to underlie the formation of new memories — that is similar to the process in mammals. But it is achieved through distinct molecular mechanisms. “I see a beautiful convergence, which has been reached in completely different ways,” Hochner says. For that discovery, Hochner’s team relied on a method taken straight from mammalian neuroscience: studying neurophysiology in brain slices kept alive for hours. Neuroscientists are now seeking to adapt technologies at scale, co-opting a slew of precision tools used routinely in mammalian biology. First glimpses inside octopus's sleeping brains reveals human-like patterns One of the first items in the modern cephalopod toolkit was the sequence of an octopus genome3, published in 2015 by Albertin, Ragsdale and their colleagues. As a standalone study, the work provided interesting insights. For example, it found that two large gene families that had grown to have crucial roles in nervous-system patterning in vertebrates had similarly expanded in the octopus, albeit through distinct mechanisms. But, Ragsdale says, the study also sent a sociological signal. “I think when we published the genome, it led a lot of people who’d been interested in these creatures to say, ‘Gee, it’s safe to go in the water now’.” The octopus had entered its molecular-biology era. Since then, researchers with a broad range of interests have joined the field. Many ask how a process they studied in mice and other model organisms works in cephalopods. Ivan Soltesz, a neuroscientist at Stanford University in Palo Alto, California, has studied how mammals navigate using a group of neurons in the hippocampus. These ‘place cells’ fire when the animal is at a specific location. His questions about octopuses were simple. “How do they do navigation? Do they have place cells?” Laurent has used mammals, fish and flies to work out how the environment is represented in brain areas that process sensory information and is extending these studies by looking at cuttlefish camouflage. Cephalopods control the colour and pattern of their skin directly through neural activity and can change colours to match their environment — providing a read-out of the brain activity that perception evokes. Or as Montague, who also works on cuttlefish camouflage, puts it: “No other animal can tell you what it sees, except a human.” In 2023, Laurent’s group showed that, in just a few seconds, cuttlefish home in on an optimal match to their environment by cycling through a succession of approximate ones4. He is now working out how they assess the quality of each match to create a feedback loop that improves their disguise. At the same time, researchers in Japan published work showing that octopuses have bouts of rapid changes in skin colour when they are asleep5, suggesting that they might be dreaming. Niell — whose lab studies vision in both mice and in octopuses — has focused on the cephalopod’s optic lobe, which, as the initial visual-processing structure, is roughly similar to a vertebrate’s retina. In 2022, his team analysed gene expression in individual neurons, identifying six main classes of cell6. By looking at the cells’ locations, the researchers found a previously unknown layered organization. They then looked at neuronal responses to visual stimuli7. Octopuses have complex, camera-like eyes (top) and highly organized visual systems (below).Credit: Nature Picture Library/Alamy; Dr Denise Piscopo, Niell lab These initial characterizations revealed similarities and differences between octopuses and other animals. For example, the researchers found a map of visual space in the octopus brain, a common feature across the animal kingdom. And octopuses have neurons that respond specifically to some visual features, such as the orientation of lines or grids, as do mammals. “There’s going to be some of these shared principles, but then there’s also likely to be things that are just completely novel,” Niell says, “that have either a different solution to the same problem — or solve problems that our visual system doesn’t have to.” His current work seeks to find where that balance lies. When Hochner’s team produced a partial connectome8 — a highly detailed map of all the synaptic connections between neurons — for the octopus’s vertical lobe, it also found a mix of familiar and new principles. The researchers found some small, simple — and highly abundant — neurons in the lobe that they think have an analogous function to individual branches on more-complex mammalian neurons. Other groups are working on other brain and nerve-cord connectomes. Hard graft Not every technique used in other animals has been easy to adapt to cephalopods. One method that has proved difficult to establish is the ability to record from large numbers of individual neurons. In mammals, such in vivo recordings have been a driving force in the field. The techniques contribute to a major goal of systems neuroscience: to understand how neurons, both alone and in populations, respond to stimuli and generate behaviour. But, as Niell says — and Boycott found some 70 years ago — “everything is a little bit harder in cephalopods”. One major problem is that cephalopods don’t have skulls — meaning that there is no hard surface on which to fix electrodes. Moreover, if you leave something sticking out of an octopus’s head, the animal is likely to reach up and remove it. More fundamentally, Soltesz says, the small size of most cephalopod neurons and their intrinsic electrical properties make them harder to record from than vertebrate ones. But his group9 and others have now made progress on recording the averaged activity of small groups of cells. How cuttlefish wear their thoughts on their skin There are other idiosyncrasies. Many cephalopod species can change colour to camouflage themselves. But octopuses are contortionists — their flexible bodies make it hard to keep track of changes to pigment and pattern. So Montague and others use cuttlefish because the animals’ flat bodies are easier to image. Researchers are developing tools for several species at once — a more challenging approach than focusing on a single species, but one that comes with advantages. For example, Montague is making genetically modified cuttlefish and Albertin has helped to create a CRISPR system for editing genes in a squid species that is small and breeds well in the lab. Establishing any new animal model is hard work, and both Laurent and Soltesz praise the postdocs who have led many gruelling efforts to develop recording methods. Montague’s biggest concern is the cost of eventually setting up her own lab. “Cephalopods are just incredibly expensive,” she says, “and they require absolute expert care.” Patchwork protections Laws concerning cephalopod care and welfare vary widely around the world. “Different people are working under completely different constraints”, says Courtney. “In Japan and the US, there are no legal requirements in terms of ethics, whereas in Europe and UK, there’s quite strict ethics.” Those more-stringent requirements mirror the legal protections for vertebrates used in research, including adequate anaesthesia and pain relief when needed. Inside the mind of an animal Protections were introduced in Europe and the United Kingdom in the 2010s, with Fiorito and others providing guidelines for cephalopod care. But efforts to do this in the United States have faltered, leaving labs there and elsewhere with limited legal obligations. (The US-based researchers interviewed for this article say they voluntarily follow European guidelines.) Crook’s research10 supports the idea that cephalopods experience pain. But, she says, no analgesics developed for mammals seem to work in cephalopods, and local anaesthetics seem to have limited efficacy. Crook is looking for compounds that relieve pain in cephalopods and says that others should be invested in helping with these efforts, too. Developing an animal model with very little knowledge about how to relieve pain “is an ethical minefield”, she says. Despite the need for caution, Crook and others are excited about expanding these animal models and, unlike the stepwise progress seen in other model organisms, many technologies are arriving simultaneously. “The fact that everybody’s on it at once, I think is fascinating,” Crook says. “This is a really, really different way of building a field in neuroscience.”

A microscope cross-sectional image of a mouse nose, showing the anatomical structure of the nasal epithelium. Credit: Datta Lab Olfactory receptors in the mouse nose have been mapped out in unprecedented detail — overturning researchers’ understanding of how noses build a sense of smell. The research, published today in Cell1, shows how around 1,100 olfactory receptors expressed on sensory neurons are organized in tightly regulated spatial locations in the epithelial tissue that lines the nasal cavity. A second study2 provides a complementary atlas of olfactory receptor expression in the olfactory epithelium and their neural connections to the olfactory bulb in the brain. “For 30 years, we’ve taught students that the mouse olfactory epithelium is divided into a handful of broad zones, within which receptor choice is essentially random,” says Johan Lundström, a psychologist and experimental neuroscientist at the Karolinska Institute in Stockholm. “This is a landmark paper that overturns one of the foundational textbook models of olfactory organization,” he adds. Smell stripes In the study, researchers examined about five million neurons from hundreds of individual mice. They first used single-cell sequencing to identify which smell receptors were expressed by neurons in the nose, and then used spatial transcriptomics to map out where key genes were being expressed. This allowed them to pinpoint where the receptors are and show that they are always arranged in horizontal stripes running from the top of the nose to the bottom. “Each receptor adopts a particular position in the nose. Since there are a thousand positions in the nose, each receptor is expressed basically in a stripe that overlaps with other receptor stripes, in a thousand overlapping stripes,” says study co-author Sandeep Robert Datta, a neurobiologist at Harvard Medical School in Boston, Massachusetts. How do we smell? First 3D structure of human odour receptor offers clues Datta and his colleagues propose that this spatial mapping is organized during development and is controlled by sets of genes. The authors found that a molecule called retinoic acid had a key role in this process. They discovered a gradient in the amount of retinoic acid present at different points in the nose. By tweaking how much this molecule was expressed, they showed that it helps to control gene activity, guiding each neuron to express the correct type of smell receptor for its location. “There’s been a ton of back and forth in the field about how this is all mapped out, and this nails it. I think it really changes the way people think about the olfactory system and just solves a huge problem in the field about how the mapping happens,” says Joel Mainland, an olfactory neuroscientist at the Monell Chemical Senses Center in Philadelphia, Pennsylvania. Nose to brain Datta’s team also showed that the nasal receptor map was mirrored by similar patterns of gene expression in the olfactory bulb, much as the organization of maps in the brain for touch, hearing and vision marry up with those of their respective receptors. “This means that the maps in the nose and the brain are not two separate problems the system has to solve, but two readouts of the same developmental logic,” says Lundström. Datta thinks that the work has implications for using stem cells to repair sense of smell — a person would need all of the stripes to detect a full range of smells. “It means those stem cells have to occupy the whole spatial extent of the nose in order to repair the nose. You can’t just infuse stem cells in one location and expect your sense of smell to recover,” he says. Although the work is in mice, Datta thinks that the same system exists in humans. His team is now looking for spatial maps in human tissue, and trying to match different smells to the bands of receptors.

Only a small portion of patents owned by universities in China become commercial products.Credit: Xu Changliang/VCG via Getty China’s intellectual-property regulator has been playing matchmaker — connecting researchers with patents to companies that can commercialize them. Last month, the China National Intellectual Property Administration said that as a result of these introductions, around 80,000 patents from universities and research institutes were commercialized between 2023 and 2025. The effort is part of the government’s desire to translate more research into products and services. China holds more than five million domestic-invention patents, but few are brought to market. In 2022, only 3.9% of university patents were commercialized, according to state media. Since 2023, the agency says it has identified around 680,000 patents held by universities and research institutes that could be commercialized, and has connected the innovators with 460,000 companies that could bring the ideas to life. Marina Zhang, who studies innovation with a focus on China at the University of Technology Sydney in Australia, thinks the matchmaking will create lasting connections between academia and industry. Robert Conn, who studies research and science philanthropy at the University of California, San Diego, says the approach is still new, so it is too early to tell how effective it will be in the long term. But he thinks it could work, because companies in China are often willing to adhere to government directives. “China’s system is top-down, with the state playing a central role in driving partnerships and setting directions,” he says. The Chinese Ministry of Education is also exploring the use of artificial intelligence and big data to identify the potential value of university patents and possible applications, Zhou Dawang, an official with the ministry, told state media. Commercialization challenges Patenting has been strongly encouraged by universities in the country and has been tied to career advancement for researchers, says Li Tang, a public-policy researcher at Fudan University in Shanghai, China. But at times, this “led to what we might call strategic patenting — in which patents are filed to meet evaluation criteria rather than to support downstream commercialization”, she says. As a result, many patents have yet to demonstrate proof they will work, or are needed, says Zhang. As well as the matchmaking effort, the government has introduced incentives that reward the successful commercialization of research rather than the filing of patents. Zhang says that the government is also taking steps to address systemic barriers that hold back research commercialization in China. These include unclear rules around who shares the profits of a patent, and a shortage of technology-transfer professionals who help academics to translate patented technologies into commercial products, says Zhang. Market forces Conn says that China’s state-driven approach contrasts with the approach taken by other countries such as the United States. There, the translation of research into commercial products is mostly market driven. US venture capitalists “walk the halls of universities” to identify innovations that they can fund. The Bayh–Dole Act, introduced in 1980, gave US institutions ownership of government-funded research, enabling them to license companies without government interference. Philanthropic endowments and private funding further strengthen the pipeline from basic research to practical applications, adds Conn. For years, China’s research funding has focused on manufacturing and applied science rather than on basic research, he says. The state’s intervention could compensate for the smaller size of China’s venture-capital sector, which is less embedded in the research ecosystem than is the equivalent in the United States, says Conn. There has also been an increase in philanthropic money going into scientific research in China, says Conn. Philanthropic money often funds early, high-risk research that might not get support from venture capitalists or public funding institutes. In March, China announced plans to increase its overall research and development (R&D) expenditure by at least 7% per year over the next five years.

Companies have proposed launching fleets of satellites into orbit that would act similar to data centres on Earth.Credit: NicoElNino/Alamy As the huge data centres powering the artificial-intelligence boom grow ever more unpopular on Earth, companies are planning to launch them into space. Data centres will use twice as much energy by 2030 — driven by AI In the past few months, firms including SpaceX, Google and Blue Origin have all shared plans to launch large fleets, or constellations, of satellites into low Earth orbit. The networks would act in a similar way to the interconnected computers inside data centres on Earth, which process, store and transmit data on a massive scale. Putting these ‘orbital data centres’ into space could, in theory, address concerns about their energy and water consumption, and their occupation of wide swathes of land. The idea is that constellations would use sunlight for energy rather than driving up electricity costs on Earth, and they would be cooled by space’s naturally cold environment rather than by water sources on our planet. For some, such a solution can’t come soon enough. Data centres on Earth have become so environmentally taxing that communities and politicians are taking action against them. For example, the board of trustees for a township in the US state of Michigan voted last week to institute a one-year moratorium on the delivery of water to hyperscale data centres so that the township can study the effects of a planned facility. As companies plug away at satellite designs and lobby for launch approvals, they are pushing for the space-based data centres to become a reality in the next few years. Researchers who spoke to Nature, however, see it taking longer to wrangle the sci-fi technology into being. How did the trend start? The chatter surrounding orbital data centres to power AI isn’t new. In September 2024, engineers at the space-technology company Starcloud in Redmond, Washington, published a white paper arguing that orbital data centres are “feasible, economically viable, and necessary to realize the potential of AI”. And in November 2025, researchers at the technology giant Google announced its Suncatcher project, with a plan to “one day scale machine learning compute in space”. But it was in January this year that “everything blew up in this area”, says Kathleen Curlee, who studies the space economy at Georgetown University in Washington DC. That’s when billionaire Elon Musk’s aerospace and tech company SpaceX, headquartered in Starbase, Texas, shared plans to launch one million satellites to form an orbital data centre — a staggering number compared with the roughly 15,000 satellites now in low Earth orbit. Not to be outdone, the China Aerospace Science and Technology Corporation, based in Beijing, joined the race at about the same time, and billionaire Jeff Bezos’ space-tech firm Blue Origin, based in Kent, Washington, later filed for its own constellation. US communities have been meeting to oppose data centres that would use their electricity and water (shown is a gathering on 4 August in Tucson, Arizona, that discussed installation of a facility by Amazon Web Services).Credit: Wild Horizons/Universal Images Group via Getty Adding more pressure to get data centres into orbit is a March plan released by US President Donald Trump’s administration called the Ratepayer Protection Pledge. AI firms such as Google, OpenAI and Musk’s xAI signed the pledge, agreeing to build infrastructure for or to buy any power their data centres need, to prevent US residents from footing the bill. It’s a non-binding agreement, but by implementing it ahead of the US mid-term elections in November, Trump has made clear that data centres are a political issue that could sway voters. What are the challenges to building the data centres in space? For these projects to succeed, several engineering obstacles need to be overcome. One is ensuring that the satellites’ electronics cool properly. Although space is much colder than Earth, it is also a vacuum, meaning that the extreme heat generated primarily by AI chips will probably not easily dissipate on its own. Technologies already exist to cool gadgets in space, such as the heat radiators on the International Space Station. But these are probably too heavy — and, consequently, too expensive to launch — to be practical for orbital data centres, says Igor Bargatin, a mechanical engineer at the University of Pennsylvania in Philadelphia. Huge new satellite outshines nearly every star in the sky Another obstacle is the effect of harsh space radiation on AI chips. As protons and the other high-energy particles that make up space radiation strike the chips, they could flip a binary bit from a 0 to a 1, or vice versa, effectively corrupting stored data, says Ken Mai, a principal systems scientist at Carnegie Mellon University in Pittsburgh, Pennsylvania. In a white paper released last year, however, a team at Google reported that its existing Trillium chips remained stable under the radiation of a proton beam. But it’s “still an open question of how much radiation can be tolerated”, Bargatin says. If the number of satellites in low Earth orbit increases by two orders of magnitude, “it certainly seems like a big challenge from the space-traffic management perspective,” Bargatin adds. He points specifically to a phenomenon called the Kessler effect, which predicts that as low Earth orbit becomes overcrowded, collisions will increase exponentially as more debris is produced — potentially making certain orbital regions unusable. Overcrowding is also a point of concern for astronomers, whose space images are already being marred by satellites. For instance, if SpaceX’s plans were to go through, each image captured by the Very Large Telescope in Chile would lose 10% of its data, according to the Royal Astronomical Society in London. How long might it take to build them? Some companies are already bringing AI models to space, with Starcloud running a version of Google’s AI assistant Gemini on one of its satellites and the Chinese aerospace company Adaspace, based in Chengdu, deploying ten AI models in orbit. How much energy will AI really consume? The good, the bad and the unknown Regulatory agencies are also acting fast. The US Federal Communications Commission, which oversees satellites, accepted SpaceX’s satellite proposal and opened it up to public comment within days of receipt — although the commission has not yet given approval. At the end of March, Musk unveiled some technical details about SpaceX’s orbital data centre, including an illustration of an AI Sat Mini to be used in the constellation, complete with a large radiator to remove heat, the industry website Space News reported. Musk has been publicly bullish about space-based data centres, saying at the World Economic Forum, held earlier this year in Davos, Switzerland, that “the lowest-cost place to put AI will be space. And that’ll be true within two years … three at the latest”. But just last week, the news agency Reuters revealed that SpaceX itself is much less confident. In a filing ahead of its highly anticipated initial public offering, the firm said that its “orbital AI compute” initiative will rely on “unproven technologies” and “may not achieve commercial viability”. Some researchers think that if orbital data centres are ever to come to fruition, it will be a long journey. “Five years is probably the best-case scenario, as far as I’m concerned,” Bargatin says. Curlee sees an even lengthier timeline for data centres to proliferate in space as they have on Earth. “I don’t really see that happening for at least 10 years — probably 15 or 20,” she says.

Nvidia’s Cosmos is one world model being trained on physics data about real-world environments.Credit: NVIDIA Corporation An ongoing trend in artificial intelligence could have huge implications for how the technology is used in research. Machine-learning systems such as large language models (LLMs), which turn prompts into text, images and video, are becoming increasingly sophisticated and continue to make astonishing progress, including in science. But such ‘generative AI’ tools also have limitations. The approach does not always make accurate predictions about the physical world, and could fail at modelling correctly what would happen if a car were to go off the edge of a cliff, for example. This would have implications for developing effective and safe AI-powered robots and self-driving vehicles. Some researchers, including the computer scientist and AI pioneer Yann LeCun, who founded the firm Advanced Machine Intelligence (AMI) Labs in Paris, have turned their attention to a different type of AI tool, developing systems known as ‘world models’ that are trained on real-world data and can embody virtual, interactive and 3D environments. The approach is attracting huge investment and business interest. AMI Labs — which is taking a radical approach to world models — has raised more than US$1 billion, a record initial infusion of money for a European company. Technology giants such as Google and Nvidia are also developing world models, as are several other start-up companies. What is a world model? There are several definitions of what a world model is. In the broadest sense, any neural network trained on data about the real world (or even about some alternative universe) has some sort of model of a world embedded in it. But over the past two years or so, many researchers have begun to use the term to describe AI that can produce a consistent, explorable and often interactive world that is reminiscent of a first-person video game. A world model has to ‘know’ enough about physics that if the user pushes an object off a table then the object will fall down. World models also provide a more-interactive experience for a user than does generating images of video material from text prompts. For example, Google Deepmind’s world model Genie 3, which the company released in August 2025, uses simple text descriptions to generate photorealistic environments that can be explored in real time. The AI revolution is coming to robots: how will it change them? What sort of data are world models trained on? The companies that build generative AI systems tend to guard their ‘secret sauce’ fiercely. What is known is that interactive world models are trained, in part, using thousands of hours of videos from the real world, as well as with accurate simulations of physical environments that are programmed to observe the laws of physics. What AI capabilities could world models unlock? “The more exciting version of a world model is one in which you can take actions,” says Jeff Clune, a computer scientist at the University of British Columbia in Vancouver, Canada. Such an environment could be a safe setting in which to train AI systems that control robots, and could be much faster than letting robots learn by interacting with physical objects, says computer scientist Anastasis Germanidis, co-founder of Runway, a start-up company in New York City that released a world model called GWM-1 in December 2025. Does AI already have human-level intelligence? The evidence is clear How will world models benefit researchers? A world model such as Genie 3 can provide vast numbers of ways to train the software that can power a robot or a self-driving car, says Clune, who contributed to the development of Genie during a contract with Google DeepMind. In research, tools such as autonomous chemistry laboratories — robot chemists — could quickly accumulate thousands of hours of training in a world model before being deployed to an actual laboratory. How does AMI Labs’ approach differ from those of other companies? LeCun’s company is developing world models based on his Joint Embedding Predictive Architecture (JEPA), which is designed to have a ‘higher level’ internal representation of the world. Whereas typical generative AI is designed to predict what frames look like pixel by pixel, JEPA would instead produce more-conceptual predictions, such as how an object’s motion follows the laws of physics, which can concisely describe an object’s motion with a few variables, such as its position and orientation in space. LeCun said in a talk at Harvard University in Cambridge, Massachusetts, in late 2025 that JEPA could be used to accurately predict the movements of planets on the basis of fewer variables than pixel-per-pixel models use. To predict the position of Jupiter in 100 years, “you don’t need to know all the details ... you only need six numbers”, LeCun said. One possible advantage of AMI Labs’ higher-level approach is that it could require fewer computational resources to run than typical, power-hungry generative AI. “I’m excited that Yann is pursuing it, because it’s a beautiful idea,” adds Clune. However, others say that the same improvements world models could be achieved mostly by scaling up current generative AI techniques with more training, computing power or both. “One very consistent theme in the history of AI is that the simple approach tends to scale more easily,” says Germanidis.

President Donald Trump and his administration downsized US science by historic margins last year as it reduced the workforce at federal research agencies by tens of thousands of people and terminated thousands of research grants. But another set of cutbacks in federal science has drawn less attention. US science after a year of Trump: what has been lost and what remains Across the government, the administration terminated more than 100 independent advisory panels, comprising university scientists and other outside experts who help to guide national science priorities. The cuts — driven by a February 2025 executive order aimed at shrinking federal bureaucracy — target committees that agencies rely on to assess biomedical and environmental policy, provide guidance on setting research priorities and ensure transparency in how the government makes science-based decisions. The scope of these committee terminations is unprecedented, a Nature analysis finds (see ‘Cancelled committees’). For example, the Department of Health and Human Services (HHS), which includes the National Institutes of Health, disbanded 77 advisory boards — more than one-quarter of all its advisory committees — in 2025. By contrast, in fiscal year 2024, the agency terminated just two committees. Source: FACA Database A similar pattern of committee closures played out at other agencies such as the National Science Foundation (NSF) and the Department of Energy (DOE). At NASA, more than half of the advisory boards were disbanded. These panels, which are governed by the Federal Advisory Committee Act (FACA), are typically staffed by researchers and other experts from outside the government. Some of those that were closed in fiscal year 2025 had been advising on topics such as organ transplantation, HIV prevention, high-energy-physics research and planetary science. The February 2025 executive order’s stated purpose was to “minimize Government waste and abuse, reduce inflation, and promote American freedom and innovation”. And some scientists and agency employees said there can be sound reasons to streamline FACA committees by combining some or eliminating ones that no longer serve a purpose. But many researchers say that the scale of the administration’s efforts greatly reduces the amount and quality of advice that the government receives from the scientific community and businesses, as well as organizations that represent people with diseases such as Alzheimer’s. Researchers who spoke to Nature say that by terminating such a large number of scientific advisory committees and not replacing the vast majority of them, the administration is cutting off federal agencies from independent outside expertise. At the same time, it limits the flow of information from the government to the scientific community and the public. “That two-way street, I think, was invaluable,” says Juan Meza, an applied mathematician at the University of California, Merced, who formerly served on two panels at the NSF and the DOE that have been disbanded. “We could act as ambassadors in both directions,” he says. The terminations aren’t the only changes to advisory committees that the administration rolled out last year. Nature found that the US government has sharply reduced the number of open FACA meetings — by more than 50% for some agencies — at which the public could observe deliberations and provide input. Some agencies substantially reduced the number of public reports they issued. Trump’s new science advisers include 12 technology chiefs — and one academic And in some other cases — including the prominent example of the Advisory Committee on Immunization Practices (ACIP) that makes recommendations on vaccines — the federal government has drastically changed the composition of the committees, removing people who disagree with its stance and installing ones who agree. Last week, the Trump administration abruptly fired all 22 members of the board that advises and oversees the NSF. As a rationale for the terminations, a White House spokesperson pointed to the 2021 Supreme Court case United States v. Arthrex, Inc., which it says “raised constitutional questions” about the board’s membership and the fact that its members are not confirmed by the Senate. The spokesperson said the White House aims to update the law so that the board can “perform its duties as Congress intended”. Researchers say that the elimination of panels and other changes seemingly contradict the Trump administration’s promise, outlined in an executive order on ‘gold-standard science’ on 23 May last year, to improve transparency in federally funded science and in science-related decisions taken by federal agencies. “The fewer of these advisory panels there are, it inherently diminishes the transparency of the entire operation,” says Carrie Wolinetz, who previously administered several advisory panels as the former head of the NIH’s science-policy office. The White House rebutted these claims. Spokesperson Kush Desai says that, during the COVID-19 pandemic, the “federal government’s glut of redundant, taxpayer-funded advisory committees did little to meaningfully inform policymaking for the benefit of the American people”. “The Trump Administration is eliminating the bureaucratic bloat and taking a hands-on approach to ensure that policymaking is driven by Gold Standard Science.” Biomedicine behind closed doors The 77 committee terminations at the HHS in 2025 represent a sharp departure from historical levels. Since 1997 — the full extent of publicly available FACA data — annual terminations have exceeded ten only once. In 2025, the number of open HHS committee meetings also decreased, Nature found. In the ten years before 2025, the average number of committee meetings open to the public was 255. But in 2025, there were just 91 (see ‘Closed science’). Source: FACA Database There are many more closed meetings at the HHS in any given year because most of the FACA committees assess research grants, a process that is kept confidential. But in 2025, the ratio of open to closed meetings dropped from an average of over 9% for the previous ten years to 4%, representing a shift towards closed meetings even outside the grant-review process. Among the disbanded groups was one charged in 2023 with making recommendations on research into long COVID and treatment for millions of people with the condition in the United States. The committee was a unique bridge between patients, federal science agencies and policymakers, says Ian Simon, the former head of the HHS Office of Long COVID Research and Practice, which was eliminated amid the government downsizing last year. The committee was “designed to give patients a significant voice equal to those of researchers and physicians”, Simon says, and its closure is a blow to research. “It is very hard to see how these actions will advance the work that’s needed to understand long COVID and other infectious chronic conditions.” Other panels terminated by the HHS include the Advisory Committee on Organ Transplantation, which advised the agency on policies regarding organ donation, procurement and equitable allocation, and the Dietary Guidelines Advisory Committee, tasked with reviewing current nutritional science to inform the federal government’s dietary recommendations. The federal government subsequently issued new dietary guidelines in January without the committee’s input, a move that sparked controversy among some nutrition experts who argued that aspects of the revisions bypassed the scientific consensus. The downsizing of HHS advisory committees is starker than the 2025 termination numbers suggest: some of the FACA committees are also meeting less often than in typical years or have not met at all since Trump took office again. For example, the NIH leadership has historically relied on the Advisory Committee to the Director and the congressionally mandated Scientific Management Review Board — both of which have not been officially terminated — to navigate major agency reorganizations or funding shifts, says Wolinetz. But the NIH leadership did not convene either of these panels last year as the agency cut thousands of projects on disfavoured topics and reduced the autonomy of each of its institutes by centralizing peer review and other administrative functions. Wolinetz says that it’s smart to consider, on a semi-regular basis, whether each committee is still serving its purpose and justifying its taxpayer cost; some panels can become obsolete “vestiges”, she says. But by terminating so many committees and not consulting others, Wolinetz says the federal government loses a crucial mechanism for ensuring that its decision-making is transparent and subject to scrutiny, including by the public. Advisory committees act as a “locus of public engagement that federal agencies can’t do on their own” about issues the government is grappling with, she says. The actions seem at odds with the ‘radical transparency’ at HHS that is a stated policy goal of health secretary Robert F. Kennedy Jr, she says. She also worries about cases in which the Trump administration has not terminated committees — but instead drastically changed them. For example, last June, Kennedy abruptly fired all 17 members of ACIP, the US Centers for Disease Control and Prevention’s premier vaccine advisory panel. Claiming that the panel was plagued by conflicts of interest and acted as a “rubber stamp” for the pharmaceutical industry, Kennedy reconstituted the committee with appointees whom, he argued, would bring outsider scrutiny. However, scientists and medical organizations contend that some of the new members have a history of promoting vaccine scepticism, a position long held by Kennedy. The American Academy of Pediatrics (AAP) sued the HHS over its changes to ACIP. In March, a federal judge temporary halted the installation of Kennedy’s picks for ACIP, ruling that the selections probably violated federal law requiring that such panels be fairly balanced in terms of expertise and viewpoints. The HHS later revised ACIP’s charter to broaden its scope and focus on the risks of vaccines. Kennedy also overhauled the HHS’s Interagency Autism Coordinating Committee, terminating its existing members and appointing a slate of new ones. The new slate has drawn criticism from some autism researchers who argue that it includes people who are aligned with Kennedy’s disproven claims that autism is a preventable condition linked to vaccines and environmental toxins. These reconstituted committees were not “formulated in the traditional highly vetted manner” outlined in each panel’s charter, Wolinetz says. Instead, they seem to be “constituted to support particular predetermined points of view” and are being “used to certify policy actions the administration wants to take”, she adds. Emily Hilliard, an HHS spokesperson, told Nature that the agency’s actions were in accordance with a White House order to terminate unnecessary advisory committees, adding that “these previous committees allowed the United States to remain the sickest developed nation despite spending $4.5 trillion annually on health care, driving unsustainable debt and worsening health outcomes.” The HHS will continue to convene committees as necessary, she added. The HHS did not respond to requests for comment about other issues, such as criticisms of the way the agency changed the composition of the vaccine and autism panels. Loss at the NSF The NSF, which is the premier US funder of fundamental research across all areas of science and engineering, also sharply restricted its advice pipeline last year by terminating 14 of its 52 advisory committees. These had provided the agency with advice in areas such as engineering, cybersecurity and geosciences. (All but one of the panels that review grant applications for the NSF remain active.) Exclusive: Trump team freezes new NSF awards — and could soon axe hundreds of grants Meza served on one of these terminated bodies, the Advisory Committee for Mathematics and Physical Sciences, and was also an NSF programme officer from 2018 until he left in 2022. He says that such panels can provide valuable information to agencies; for example, the committee he served on informed the NSF that the research community had concerns about the lack of support for mid-sized laboratories. Heeding the advice, the NSF established the Mid-scale Research Infrastructure opportunity in 2016 to support what it called “a ‘sweet spot’ for science and engineering that has been challenging to fund through traditional NSF programs”. The NSF declined to comment on the criticisms about the changes in its advisory committees. Consolidation at DOE Last August, the DOE terminated six FACA panels that provided advice in areas such as high-energy physics, scientific computing, and biological and environmental research. The DOE has since consolidated these discipline-specific panels into one overarching body called the Office of Science Advisory Committee (SCAC). Meza, who served on the terminated Advanced Scientific Computing Advisory Committee, worries about the loss of specific expertise. “How good is the advice coming from a committee of people that probably only have passing knowledge of some of the areas?” he asks. Persis Drell, chair of the SCAC and a physicist at Stanford University in California, acknowledges the worries researchers have raised. “In a time of turbulent change, I totally understand all of the concerns that are in the community,” she says. Drell adds that she hopes to reassure the scientific community that the SCAC is listening and is serious about helping science at the DOE. “I have two goals: one of them is to ensure that we have a strong basic science foundation and the other is that we are able to make progress on the strategic pillars that the administration has put forward,” she says. Science could solve some of the world’s biggest problems. Why aren’t governments using it? The SCAC held its first meeting at the end of March, at which it was tasked to study the prioritization of resources for artificial intelligence and quantum science, as well as for DOE facilities. Along with cutting FACA advisory committees, the DOE has tried to set up at least one secret advisory body that a judge found to be in contravention of federal law. In April 2025, DOE secretary Chris Wright hand-picked five external researchers who have challenged the scientific consensus on climate change to review the science on this topic. Wright set this body up outside of the FACA process, the DOE did not announce the group’s formation, and deliberations were not public. It produced a report that challenged the overwhelming scientific consensus, and a finding by the Environmental Protection Agency in 2009, that greenhouse gases are a threat to public health and welfare. After the report was released, mainstream climate scientists heavily critiqued its conclusions. Two US non-profit advocacy groups, the Environmental Defense Fund and the Union of Concerned Scientists, sued the agency, alleging that the DOE had flouted FACA rules by setting up this advisory committee outside the FACA framework. As part of the legal proceedings, the agency released e-mails showing that internal DOE reviewers had found problems with the report before it was made public, calling its findings “misleading” and “factually incorrect”. In January, a court ruled in favour of the organizations that sued the DOE. FACA is “designed to prevent exactly what the administration did here”, by requiring that scientific advice be sourced through a transparent process, says Phil Duffy, an atmospheric physicist who worked in the Office of Science and Technology Policy during the administration of former US president Joe Biden. As a counterexample, Duffy points to the process behind National Climate Assessments, which had authors representing all 50 US states, multiple rounds of public input and multiple reviews by federal-agency experts. That kind of process, he says, provides “a lot more credibility” than the DOE’s climate report. The DOE did not respond to Nature’s request for comment. NASA drops committees In percentage terms, the changes at NASA dwarf those at other agencies. NASA lost 6 of its 11 FACA committees. And the agency held only 10 meetings of such committees in 2025, whereas the annual average was 35 during the previous ten years. Five of the committees that were terminated fed advice into the divisions of NASA’s science mission directorate. The committees previously met several times a year; researchers told Nature that such meetings were an important venue for scientists in the broader community to hear updates from NASA officials, and for committee members to take the community’s input and sharpen it into advice for agency leaders. For example, before it was disbanded, the planetary science committee recommended that NASA better connect its plans to send humans to Mars with robotic missions to bring back Martian rock samples, to align more closely with federal priorities for maintaining global leadership in space. And in the middle of heated battles over how to divvy up limited funding for operational space telescopes, the astrophysics committee told NASA that it should not pre-emptively cut the budget of missions such as the Hubble Space Telescope, without broader consultation on the alternatives. Such information, communicated by experts sitting on official committees in open meetings, is crucial for NASA to take on board when making key decisions, says Kelly Holley-Bockelmann, an astrophysicist at Vanderbilt University in Nashville, Tennessee, who chaired the disbanded astrophysics group. She says that she and others served on these committees as volunteers. “I don’t see them as bureaucracy at all,” she says. “We do it because we loved astrophysics and we loved NASA, and we care very deeply about the science that NASA is able to do.” In late March, NASA also began winding down an advisory group one level up from the five disciplinary committees, namely the science committee of the NASA Advisory Council. Several members of that group received letters from NASA administrator Jared Isaacman saying that “the structure of Federal Advisory Committees at NASA is being adjusted, and your role will conclude at this time”. Without such committees, there is a reduction in the amount of crucial scientific advice flowing into the agency, Holley-Bockelmann and others say. “Why would you want less information?” asks Benjamin Greenhagen, a planetary scientist who chairs a Moon-focused, non-FACA advisory group that still exists but has had its NASA funding terminated. NASA officials dispute the criticisms that there has been a reduction in advice to the agency. “I don’t think us being able to talk to the community and getting advice has really fallen by the wayside,” says Nicola Fox, NASA’s associate administrator for science at the agency’s headquarters in Washington DC. “We get our input from the community from a variety of different ways.” She points to advisory committees organized by the National Academies of Sciences, Engineering, and Medicine, which produce the ‘decadal surveys’ of community priorities that help to inform NASA decisions about what missions to pursue. With the closure of committees and reduction in the number of open meetings, researchers in various disciplines have taken it on themselves to form independent advisory committees to replace some of the terminated ones and those whose members have been entirely replaced by the Trump administration. In the biomedical area, for example, groups of independent scientists are setting up their own unofficial committees to rival the vaccine and autism committees, which they argue do not represent the weight of evidence in these areas — a charge that was part of the successful lawsuit brought by the AAP over ACIP. Another example is the Census Scientific Advisory Committee (CSAC), which had provided scientific advice about how to conduct the decadal US census and other demographic surveys. After the Department of Commerce terminated that committee in March 2025, CSAC members set up an independent version that came up with consensus recommendations that they made public. The alternative panel says its goal is “to ensure that the U.S. Census Bureau continues to benefit from the highest levels of scientific expertise, independent review, and constructive feedback.” “I would hope that in the future, in this administration or some other, that there would be a structure that would enable these agencies to get external advice,” says Barbara Entwisle, former chair of the CSAC and a sociologist at the University of North Carolina, Chapel Hill. “It’s just the way you get the best ideas.”

Terence Tao has been exploring the inter between maths and AI.Credit: David Esquivel/UCLA Is mathematics being taken over by generative artificial intelligence? A spate of media reports has suggested that the field is being fundamentally changed by the technology. Many maths researchers say that AI’s actual capabilities are often hyped up, and that it’s not yet time to announce the death of their profession. Still, by many accounts, in the past year, AI has jumped from solving secondary-school-level problems to actually being useful in research mathematicians’ daily work. Terence Tao, a mathematician at the University of California, Los Angeles, has been at the forefront of experimentation with large language models, including OpenAI’s GPT, Anthropic’s Claude and Google’s Gemini. In particular, he has contributed to a project to test the skills of AI systems on a collection of more than 1,000 problems, ranging from major conjectures to obscure factoids. The questions were accumulated by the late Hungarian mathematician Paul Erdős (1913–1996) over his lifetime. Last month, Tao teamed up with Tanya Klowden, an art historian at the Courtauld Institute of Art in London, to explore the implications of AI for researchers and the world at large. They took mathematics as a test case, and urge society to adopt the technology but in a human-centric way. They posted a draft of their essay, due to be published in the forthcoming edition of The Blackwell Companion to the Philosophy of Mathematics, on arXiv1. Nature spoke to Tao about how the technology is transforming his profession. Why do you think it is important to consider the impacts of rapidly evolving AI? I feel like AI is not just another technology like the word processor or the web browser. It really is forcing us to rethink fundamental questions — what is a mathematical proof? What is a paper? What is the purpose of our profession? If we don’t ask these questions ourselves, then they will get answered for us by a technology company or decided by financial incentives. We have to get ahead of this. Why has maths become ‘the next big thing’ for AI? In almost any other application, the biggest Achilles heel of AI is that it makes unverifiable mistakes. But in mathematics, almost uniquely, you can automatically check the output — at least if the output is supposed to be the proof of a theorem, although that is not the only thing mathematicians do. So, AI companies have recognized that their most unambiguous successes — if they’re going to have any — are going to come from mathematics. In my opinion, there are many use cases of AI that are risky and controversial. In mathematics, the downsides are much more limited What will happen to mathematics as a field in the age of AI? The job description is changing a lot. A graduate student who refuses to touch AI systems and just wants to prove things the way we’ve done in the past might find they have fewer opportunities, unfortunately. Those who understand maths traditionally but are also adept at using new tools can flourish. I don’t think AI will replace mathematicians, but it will complement them. There could be a division of labour: we decide what to prove and what we think is interesting. We could get instant feedback from the AI. We could propose a definition, make some conjecture and AI could instantly evaluate it. But who knows, it’s all changing. But we do have to somehow let go of conventional assumptions of what intellect is. I think we have a human-centric way of thinking about all types of intellectual task, and we have to accept that this is not the only perspective. What are mathematicians’ attitudes? Are they embracing the use of AI? It’s very much a spectrum. You see all the ‘five stages of grief’ play out — denial, anger, bargaining, depression and acceptance. And I think this is happening everywhere. But I think we’re beginning to see denial fade away. How good are AI models at solving mathematical problems? For a while, you could say they were just picking up proofs that were in the literature. Or that they solved an easy problem that nobody had looked at before. But recent progress has been increasingly impressive. We’re just beginning to see examples in which AI — maybe by luck — starts solving problems that people care about. It’s still very occasional, and it still has a lot of weaknesses; it is not a replacement for what humans do. But it’s getting harder to deny that these tools can work. Can you tell us a notable success story? In the areas of mathematics I am most familiar with, the most recent promising example is Erdős problem #1196. Unlike most other Erdős problems solved by AI so far, it was studied intensively by several mathematicians, but it ended up having a fairly short proof that all the humans missed. It seems that the AI-generated proofs have stumbled on some new methods to solve problems in this area, but we are still in the process of assessing how generalizable these methods are to other problems. Do you see differences between the leading models? A little bit. In my experience, ChatGPT makes fewer mistakes, and it’s better for really rigorous mathematics, but the way it writes is just too robotic. Gemini makes nice pictures, but its outputs are too wordy. Claude is a bit faster, and explains things better: it feels more human. It’s more conversational. A lot of it is just the default prompting: I think that if you change the prompt, you can make one model similar to another. What are the main limitations? One weakness of AI tools right now is that you can converse with them only to a limited extent. But they don’t have a permanent memory of what they did. For now, defining new concepts and trying to decide what problems to work on — this might still be something that human mathematicians will have to do for a while. The current AI systems are not designed for that.

Researchers are discovering that mitochondria (blue), known as cells’ power plants, also play parts in cells’ immune responses.Credit: David M. Phillips/Science Photo Library When a parasite invades a cell, the cell’s mitochondria react by shedding their outer layers to form brand-new cellular compartments — or organelles — that digest molecular trash. Cells are swapping their mitochondria. What does this mean for our health? The team that made the discovery showed that when the mitochondria — best known as cells’ energy producers — form these organelles, it helps the parasite to proliferate, although it isn’t clear exactly how. Maybe the parasite ‘feeds’ off the degraded material inside the tiny compartments, says team leader Lena Pernas, an immunologist at the University of California, Los Angeles. But one thing is certain, Pernas said when presenting the findings at a Keystone symposium on mitochondrial signalling, held in Colorado in February: “Mitochondria are able to give rise to new organelles during infection.” The discovery, posted on 24 April to the preprint server bioRxiv1 ahead of peer review, adds to a growing list of roles that researchers are uncovering for mitochondria in immunity, including surveilling pathogens2 and coordinating immune signalling3. It also lends credence to the hypothesis that membrane sacs, or vesicles, shed by the earliest mitochondria gave rise to organelles in eukaryotes — a group of organisms, including plants and animals, whose cells have an enclosed nucleus and other intracellular compartments. If modern-day mitochondria can spawn new organelles, which have specialized jobs inside cells, it’s easy to imagine that their ancestors did, too. Mitochondria manipulation Pernas and her colleagues observed the formation of the organelles after infecting human cancer cells with the parasite Toxoplasma gondii, which can lurk in undercooked meat. A protein on the outer surface of the parasite latched onto a protein on the cells’ mitochondria, ‘pinning’ the tiny energy producers to T. gondii. These tethered mitochondria then began shedding their outer membranes, forming what are called structures positive for outer mitochondrial membrane (SPOTs). Remarkably, the SPOTs went on to engulf other organelles in the cells called lysosomes, which are acidified sacs that act as waste-disposal systems. In this video, watch a large lysosome (red, near the centre of the screen) be enveloped by a SPOT (yellow).Credit: Dr. Xianhe Li This envelopment of the lysosomes is quite the feat, says Shaeri Mukherjee, a cell biologist and immunologist at the University of California, San Francisco. “What is so cool and surprising is the ability of a pathogen to completely, not only manipulate the mitochondria, but use the mitochondria to generate an entire new organelle in the cell, with such precision,” she says. The researchers confirmed that the SPOTs-engulfed lysosomes are a new type of organelle, different from lysosomes: the outer surfaces of the new organelles don’t bear any proteins typically expressed by lysosomes. They also demonstrated that to multiply, T. gondii needs the new organelles. When the researchers used a drug called a proton-pump inhibitor to prevent the SPOTs-engulfed lysosomes from acidifying — and therefore becoming waste-disposal systems — the parasite’s proliferation was impaired. Interestingly, the SPOTs rarely engulf any other organelles inside cells other than acidified lysosomes. One reason the parasites might create the new organelles is to feed off their digested waste, Pernas says. Another could be that, by forcing the SPOTs to engulf lysosomes, T. gondii neutralizes the tiny waste disposers, which are harmful to pathogens, Pernas says. Marvellous membranes The findings suggest that mitochondria and their membranes might be a “reservoir” for forming organelles as cells adapt to various stresses, Pernas said at the Keystone meeting. “This may sound heretical,” but it makes sense if we step back and think about how organelles evolved inside cells, she added. More than 1.5 billion years ago, eukaryotes arose when a type of microbe swallowed up an ancient bacterium. Cancer cells ‘poison’ the immune system with tainted mitochondria Ten years ago, evolutionary biologists Sven Gould, Sriram Garg and Bill Martin — all at the University of Düsseldorf in Germany — proposed that, much like free-living bacteria that are constantly shedding vesicles, this ancient bacterium, now inhabiting a host cell, would have continued to shed the membrane sacs4. As time progressed, the vesicles would have evolved to become the various intracellular organelles of eukaryotes. Although there are competing hypotheses about how eukaryotic organelles came about, the discovery by Pernas and her colleagues that modern-day mitochondria can give rise to new organelles is consistent with this idea of eukaryotes’ origins. The new finding also bolsters an earlier observation that mitochondria are involved in the genesis of organelles. In 2017, Heidi McBride, a mitochondrial biologist at McGill University in Montreal, Canada, showed that peroxisomes, which break down fatty acids inside cells, are ‘hybrid’ organelles5. That is, she observed that cells lacking peroxisomes could generate them from scratch when vesicles shed by mitochondria fused with vesicles from the cells’ endoplasmic reticulum, an organelle involved in synthesizing proteins. But back then, sceptics saw McBride’s finding as either an artefact or as something that happened under only very special circumstances, Gould tells Nature. Pernas’s discovery, then, suggests that McBride’s finding wasn’t an oddity after all, and that vesicles shed by mitochondria have many uses. The vesicles “really are being used all over the place”, Gould says.

Researchers serving on the National Science Board, which advises the US National Science Foundation, received a brief e-mail on Friday telling them that they had been dismissed.Credit: Briscoe Savoy for Nature All 22 members of the advisory board that oversees the US National Science Foundation (NSF), a leading funder of fundamental science, were fired on 24 April without explanation. Every member of the NSF’s National Science Board (NSB) received an e-mail on Friday afternoon saying that “on behalf of President Donald J. Trump”, their positions were “terminated, effective immediately”. Members of the NSB are appointed by the president and serve six-year terms that are staggered, avoiding complete turnover. Asked about the reason for the termination, a White House spokesperson said that the 2021 Supreme Court decison United States v. Arthrex, Inc. “raised constitutional questions about whether non-Senate confirmed appointees can exercise the authorities that Congress gave the National Science Board.” Members of the NSB were initially confirmed by the Senate, but have not been since 2012. “This action to dismiss the NSB is unprecedented,” says Dan Reed, a computer scientist at the University of Utah in Salt Lake City and chair of the NSB from 2022 to 2024. “We need a vibrant, independent NSB, one representative of the broad science and engineering enterprise.” Trump team’s new rule could make firing government scientists easier Zoe Lofgren, a member of the US House of Representatives from California and the ranking Democrat on the House Committee on Science, Space, and Technology, criticized the move. “This is the latest stupid move made by a president who continues to harm science and American innovation,” she said in a statement. “It unfortunately is no surprise a president who has attacked NSF from day one would seek to destroy the board that helps guide the Foundation.” But House science committee chairman Brian Babin, a Republican from Texas, said, “Every President expects advisors to serve in a manner consistent with executive and legislative priorities. I look forward to seeing whom President Trump selects to fill the NSB and refocus our science agencies on their core mission: pursuing science.” This is not the first time the Trump administration has ousted federal science advisers en masse. Last year, the administration fired all 17 members of the Advisory Committee on Immunization Practices, which played a crucial part in US vaccine policy, and eliminated 14 advisory committees at the NSF. Also last year, Trump issued an order eliminating several advisory committees, including one on long COVID, to reduce government spending and “promote American freedom and innovation”. Long history The NSF and the NSB were established by Congress in 1950. The board meets five times a year and publishes reports on the state of US science and engineering that help to guide the president and Congress. Its next meeting was set for 5 May, and members say a report about the United States ceding scientific ground to China was set to be released. “Where will advice come from?” asks Roger Beachy, a biologist at Washington University in St. Louis, Missouri. He was appointed to the NSB by US President Barack Obama in 2014 and reappointed by Trump in 2020 before being terminated on Friday. “Who will help with what is the future of science in this nation?” Keivan Stassun, an astrophysicist at Vanderbilt University in Nashville, Tennessee, says that the termination of NSB members fits into a pattern of the Trump administration’s approach to science advice, which is being “systematically either dissolved or eviscerated”, he says. “It felt like only a matter of time” before that happened to the NSB, he says. Because the NSB was established by an act of Congress, the board can officially be dissolved only by Congress. Furthermore, its members are required to be ‘eminent’ in scientific fields, according to the founding legislation. Tumultuous times The firing of NSB members comes amid other turmoil at the NSF. The Trump administration proposed two years in a row to cut the NSF budget by more than half. (Congress declined to approve that proposal for the 2026 budget.) The agency has lost more than 30% of its staff since January 2025, and in December, it had to cede its headquarters to another federal agency. This year, new grants at the agency have been issued at a trickle, as the agency prepares major cuts to its divisions. One of the NSB’s key statutory roles is to approve the NSF’s budget. But multiple NSB members say that the White House Office of Management and Budget (OMB), which oversees federal spending, told NSF leadership not to share details about the agency’s spending with board members. White House stalls release of approved US science budgets “We were told that those plans were solely going to be with NSF leadership,” says Victor McCrary, a physical chemist at the Catholic University of America in Washington DC and terminated NSB chair. “And leadership was told not to share this with anybody else, including the board.” That degree of political interference concerns many of the terminated members. “Will we turn into an agency that is directed by the White House, or will we have an agency directed and managed by science and scientists?“ asks Beachy. Some of the dismissed NSB members have theories about the reason for the board’s termination. Beachy suspects they were ousted to make way for a new council of advisers for Jim O’Neill, a biotech investor nominated to be the next NSF director. Stassun says they might have been removed to prevent them from lobbying Congress to preserve the NSF’s budget for fiscal year 2027. Scientists have shared their concerns about the termination of NSB members on social media, but some have also criticized the board for not protesting more forcefully as the NSF was being targeted by the Trump administration. McCrary, however, points to the board’s work in lobbying for the NSF. “We went to Congress, we went to industry, went to every sector, and got people and rallied people around to support the foundation,” he says. Marvi Matos Rodriguez, a terminated NSB member and chemical engineer in the fusion industry based in Seattle, Washington, says that she respects her colleagues on the NSB but agrees with the criticism. “I think that the scientific community and NSF employees and people who wanted the board to speak up were right,” she says. “We should have been speaking up all along.”

The Chinese Academy of Sciences is no longer publishing its journal ranking. Credit: Cheng Xin/Getty The National Science Library of the Chinese Academy of Sciences (CAS) in Beijing has stopped publishing its influential journal ranking, taking many researchers by surprise. The ranking has had a central role in research evaluation in the country for more than 20 years and its cessation leaves universities and academics uncertain about what happens next. The CAS journal ranking, also called the CAS Journal Partition Table, was developed as a tool to help researchers assess journal quality. But over time, it began to influence hiring decisions, funding allocation and promotions. “The official retirement of the CAS Journal Partition Table is indeed a crucial watershed moment for China’s scientific evaluation system,” says Xinchen Gu, an ecologist at the South China University of Technology in Guangzhou. The ranking itself hasn’t disappeared, however. Last month, some of the team who used to run the CAS system published a new index, called Xinrui Scholar, run by a private organization. The new system uses the CAS ranking methodology. Scholars and universities are unsure whether Xinrui Scholar, or any of the several other rankings that have emerged in the past few months, will become as influential as the CAS list. Others think its closure is an opportunity to move research evaluation beyond journal metrics. Sudden closure The cessation of the CAS ranking came with little fanfare. On 24 March, an organization called Xinrui Scholar announced that it has released a new journal ranking. Like the CAS ranking, Xinrui Scholar groups journals into discipline category and then divides those into four tiers mostly on the basis of article citation metrics. This led to some confusion, says Nie, about whether CAS had simply rebranded its ranking into an independent system. On 27 March, the National Science Library of the Chinese Academy of Sciences issued a statement that it had stopped publishing the CAS ranking and declared that any journal ranking lists published by other institutions are unrelated. Sichao Tong, one of the academics who moved from CAS to Xinrui Scholar, says several factors informed the decision to close the CAS ranking. China wants to develop a journal ranking system that is recognized and used internationally. And having a private, non-government organization run Xinrui gives it more independence, says Tong. The CAS ranking has been criticized for the lack of transparency around the way it was determined. Internal indicators used by the ranking were not explained, for instance, and the system garnered criticism when long-established, internationally recognized journals were downgraded while some newer journals unexpectedly rose. CAS did not respond to Nature's questions about these criticisms. But some researchers question whether Xinrui Scholar can be truly independent if it uses the same methodology as the CAS ranking and is run by members of the same team. Tong thinks the organization can be independent and hopes to work with publishers and scientists to improve the ranking. Too influential A researcher familiar with the CAS system who requested anonymity because they were not authorized to speak to the media, says another factor in the decision to close the ranking was the criticism from researchers over its influence on research evaluation. When the CAS library system launched in 2004, it was supposed to help Chinese academics to understand which international journals in their fields were reputable. The list, which was updated annually, grouped journals into discipline categories and then divided those into four tiers mostly on the basis of article citation counts. One of its components, the Early Warning Journal List, was meant to alert scholars to predatory journals. But over the past decade, universities, public institutions and hospitals began considering researchers’ publication history in high-ranked journals when doling out awards, promotions and doctoral degrees, says Zhiqiang Nie, a cardiology researcher at the Guangdong Provincial People’s Hospital. (The CAS library has stated that its system should not be used to evaluate researchers.) As a result, the ranking often overrode scientific merit, Gu says. “Instead of asking whether a study solved a real problem or made an original contribution, evaluation became about whether it appeared in a tier 1 journal,” Gu says. “A high ranking does not equal high quality. Even top-tier journals have suffered misconduct.” The reliance on the ranking for promotions and hiring has been particularly hard for early career researchers, says Gu. Young scientists without publications in high-tier journals often struggle to get faculty positions or research funding, he says. Without the institutional authority of CAS, Zheng Liu, a genetics researcher at Guilin Medical University, thinks Xinrui Scholar won’t be as influential in Chinese research evaluation, which he says is a good thing. “Ending CAS-style dominance and moving away from pure journal-based evaluation is necessary,” he says. What’s next? At least two other journal-evaluation schemes have emerged in the past year. Dongbi Technology Data released its second Dongbi Index Global High-Quality Journal in March, which ranks more than 4,027 high-quality medical journals and 3,064 high-quality life-science journals on the basis of how the papers in them are cited. Another platform, Xuezhice, launched a ranking platform, positioning itself as an aggregator of existing metrics. Tong thinks the emergence of several rankings could benefit science. “Competition among indexing systems can encourage better methods and greater transparency,” she says. Nie and Lie argue that the way forward is not another list, but a system that places peer review and real contributions at the centre of evaluation and treats journal metrics as supporting information. Some universities have announced that, for now, they will continue referring to the 2025 CAS ranking for administrative decisions regarding promotions, but others say they will use multiple metrics.