Volume 655

Silicon microchips are the beating heart of modern computers. If they could also be used for chips in a quantum computer, that would be a potential path to scalable commercial quantum machines. But to date, the development of silicon-qubit-based systems has been difficult. A key challenge for all cryogenic quantum systems is that the electronic control circuitry sits at room temperature and the wiring needed to connect it to the qubits has to bridge a temperature difference of some 300 kelvin, limiting how large the system can grow. In this week’s issue, the HRL Quantum Team and Collaborators present a much larger and more integrated silicon-based quantum-computing platform than previously seen. Their quantum processor is connected by superconducting wires to a control chip that operates at only 4 kelvin. The system ran repeated rounds of error correction autonomously, suggesting the control architecture could be used in much larger systems. The set-up is captured on the cover: the motherboard with the 4 K cryo-controller is vertical and shown in orange, the cooler daughterboard holding the qubit chip, which runs at the millikelvin level, is blue and purple, with the superconducting cable between them in white. A second paper in this issue by Brennan Undseth and colleagues also tackles the issue of silicon-based qubits, this time enabling a mobile qubit to shuttle between four stationary qubits, allowing the team’s set-up to perform parity-check measurements that are at the core of quantum error correction.

The cover image captures the creative journey that underpins the production of a scientific illustration. Brought to life by Nature Art Editor Richard Tibbitts, the cover reveals the key stages — from initial pencil sketch to intricate ink work to full-colour rendering — that went into this image of a rhinoceros beetle. It reflects the qualities that have made scientific illustration an essential tool for discovery and communication for centuries — close observation, expert knowledge and human judgement. In this week’s issue, we celebrate the craft of scientific illustration and examine the potential effects of artificial intelligence on this crucial aid to understanding. In a Books & Arts article, five illustrators discuss their craft and probe the limits of AI. A Technology Feature examines how judicious use of AI might help time-pressed labs create visual aids to explain their research. And an Editorial takes a look at how Nature front covers have evolved to tell the story of research papers over the years.

The cover shows a seagrass meadow in shallow water off the coast of Grand Bahama in the Bahamas. Seagrass makes up a crucial marine ecosystem that supports biodiversity as well as helping to protect the shoreline and sequestering carbon. Despite their wide benefits and potential as a natural mechanism to help combat climate change, seagrass ecosystems are relatively poorly understood. In this week’s issue, Jiwei Li and colleagues present a global map of seagrass along with an analysis of how seagrass extent has changed over a four-year period at the start of this decade. The researchers used 4.75 million satellite images captured across two periods — 2019–20 and 2023–24 — to map seagrass meadows in waters up to 30 metres deep. They found that 69% of global seagrass is concentrated in the Bahamas, Cuba, the United States, Australia and Indonesia, but that only 21% of seagrass is within the boundaries of current marine-protected 0areas. By comparing the maps derived for the two time periods, the team was able to trace how the extent of seagrass changed, finding that 4% of seagrass was lost and another 4% degraded over the four years studied.

Scientific discovery is essentially an iterative process, following the cycle of generating a hypothesis, designing an experiment to test that hypothesis and analysing the data collected. A key limiting factor in this process, especially in today’s increasingly multidisciplinary research landscape, is the depth and breadth of knowledge researchers can bring to bear on a given problem — people can only read so fast and assimilate so much. Two papers in this week’s issue present independent multi-agent AI systems aimed at helping researchers speed up the laboratory research cycle. Both systems can generate hypotheses, propose experiments to test their ideas, interpret the experimental results and then refine hypotheses on the basis of the data. Google DeepMind built its lab assistant, called Co-Scientist, with the large language model Gemini 2.0 and used it to seek out potential drug candidates to treat acute myeloid leukaemia. FutureHouse created its assistant, known as Robin, using OpenAI o4-mini and Anthropic Claude 3.7. It, too, was also designed to aid drug discovery, in this case potential treatments for dry age-related macular degeneration. Both teams emphasize that their systems are designed to collaborate with researchers, with human scientists remaining in control.

Manipulating micrometre-sized matter is not easy. Optical tweezers, which use tightly focused light beams to move matter, are extremely precise but limited by the very light forces they can apply. Mechanical tweezers, which rely on more conventional means to grip and move things, can exert greater force but lack the precision offered by light. In this week’s issue, Dong Wu and colleagues reveal a 3D gripper that combines the best of both worlds. The researchers created a mechanical microclaw just 38 micrometres wide that is controlled by light. An optical fibre delivers laser light to a heat-sensitive hydrogel connected to a rigid polymer claw. Switching on the laser light heats silver nanoparticles embedded in the hydrogel, which causes the hydrogel to contract and opens the microclaw. In this way the researchers were able to use their 3D ‘optical fibre gripper’ to manipulate individual cells (pictured on the cover) and irregular microobjects as well to help assemble microdevices and to take samples in confined spaces.