Autonomic dysreflexia is a life-threatening medical condition characterized by episodes of uncontrolled hypertension that occur in response to sensory stimuli after spinal cord injury (SCI)1. The fragmented understanding of the mechanisms underlying autonomic dysreflexia hampers the development of therapeutic strategies to manage this condition, leaving people with SCI at daily risk of heart attack and stroke2-5. Here we expose the neuronal architecture that develops after SCI and causes autonomic dysreflexia. In parallel, we uncover a competing, yet overlapping neuronal architecture activated by epidural electrical stimulation of the spinal cord that safely regulates blood pressure after SCI. The discovery that these adversarial neuronal architectures converge onto a single neuronal subpopulation provided a blueprint for the design of a mechanism-based intervention that reversed autonomic dysreflexia in mice, rats and humans with SCI. These results establish a path towards essential pivotal device clinical trials that will establish the safety and efficacy of epidural electrical stimulation for the effective treatment of autonomic dysreflexia in people with SCI.

Coral reefs form complex physical structures that can help to mitigate coastal flooding risk1,2. This function will be reduced by sea-level rise (SLR) and impaired reef growth caused by climate change and local anthropogenic stressors3. Water depths above reef surfaces are projected to increase as a result, but the magnitudes and timescales of this increase are poorly constrained, which limits modelling of coastal vulnerability4,5. Here we analyse fossil reef deposits to constrain links between reef ecology and growth potential across more than 400 tropical western Atlantic sites, and assess the magnitudes of resultant above-reef increases in water depth through to 2100 under various shared socioeconomic pathway (SSP) emission scenarios. Our analysis predicts that more than 70% of tropical western Atlantic reefs will transition into net erosional states by 2040, but that if warming exceeds 2 °C (SSP2-4.5 and higher), nearly all reefs (at least 99%) will be eroding by 2100. The divergent trajectories of reef growth and SLR will thus magnify the effects of SLR; increases in water depth of around 0.3-0.5 m above the present are projected under all warming scenarios by 2060, but depth increases of 0.7-1.2 m are predicted by 2100 under scenarios in which warming surpasses 2 °C. This would increase the risk of flooding along vulnerable reef-fronted coasts and modify nearshore hydrodynamics and ecosystems. Reef restoration offers one pathway back to higher reef growth6,7, but would dampen the effects of SLR in 2100 only by around 0.3-0.4 m, and only when combined with aggressive climate mitigation.

Long-term implantable bioelectronics offer a powerful means to evaluate the function of the nervous system and serve as effective human-machine interfaces1-3. Here, inspired by earthworms, we introduce NeuroWorm-a soft, stretchable and movable fibre sensor designed for bioelectronic interface. Our approach involves rolling to transform 2D bioelectronic devices into 1D NeuroWorm, creating a multifunctional microfibre that houses longitudinally distributed electrode arrays for both bioelectrical and biomechanical monitoring. NeuroWorm effectively records high-quality spatio-temporal signals in situ while steerably advancing within the brain or on the muscle as needed. This allows for the dynamic targeting and shifting of desired monitoring sites. Implanted in muscle through a tiny incision, NeuroWorm provides stable bioelectrical monitoring in rats for more than 43 weeks. Even after 54 weeks of implantation in muscle, fibroblast encapsulation around the fibre remains negligible. Our NeuroWorm represents a platform that promotes a substantial advance in bioelectronics-from an immobile probe fixed in place to active, intelligent and living devices for long-term, minimally invasive and mobile evaluation of the nervous system.

Owing to increasing demand for low-cost energy storage with secure material supply chains, the battery community is approaching a pivotal shift beyond conventional lithium-ion (Li-ion) towards next-generation cells. Technologies that include alkali-metal anodes, solid electrolytes and earth-abundant materials such as sodium (Na) and sulfur (S) are reaching commercialization in cells. The abuse tolerance and thermal runaway hazards of such technologies diverge from conventional Li-ion cells. Consequently, designing safe batteries with next-generation materials requires a holistic approach to characterize cells and to understand their responses to abuse conditions from the beginning to the end of life. Here we provide a Perspective on how the safety and abuse tolerance of cells are likely to change for up-and-coming technologies; challenges and opportunities for reimagining safe cell and battery designs; gaps in our knowledge; capabilities for understanding the hazards of thermal runaway and how to address them; how standard abuse tests may need to adapt to new challenges; and how research needs to support affected professionals, from pack designers to first responders, to manage hazards and ensure safe roll-out of next-generation cells into applications like electric vehicles (EVs). Finally, given the large number of next-generation technologies being explored, we encourage giving priority to safety-focused research in proportion to the rate of manufacturing scale-up of each specific technology.

General reasoning represents a long-standing and formidable challenge in artificial intelligence (AI). Recent breakthroughs, exemplified by large language models (LLMs)1,2 and chain-of-thought (CoT) prompting3, have achieved considerable success on foundational reasoning tasks. However, this success is heavily contingent on extensive human-annotated demonstrations and the capabilities of models are still insufficient for more complex problems. Here we show that the reasoning abilities of LLMs can be incentivized through pure reinforcement learning (RL), obviating the need for human-labelled reasoning trajectories. The proposed RL framework facilitates the emergent development of advanced reasoning patterns, such as self-reflection, verification and dynamic strategy adaptation. Consequently, the trained model achieves superior performance on verifiable tasks such as mathematics, coding competitions and STEM fields, surpassing its counterparts trained through conventional supervised learning on human demonstrations. Moreover, the emergent reasoning patterns exhibited by these large-scale models can be systematically used to guide and enhance the reasoning capabilities of smaller models.

There is an increasing demand for multimodal sensing and stimulation bioelectronic fibres for both research and clinical applications1,2. However, existing fibres suffer from high rigidity, low component layout precision, limited functionality and low density of active components. These limitations arise from the challenge of integrating many components into one-dimensional fibre devices, especially owing to the incompatibility of conventional microfabrication methods (for example, photolithography) with curved, thin and long fibre structures2. As a result, limited applications have been demonstrated so far. Here we use 'spiral transformation' to convert two-dimensional thin films containing microfabricated devices into one-dimensional soft fibres. This approach allows for the fabrication of high-density multimodal soft bioelectronic fibres, termed Spiral-NeuroString (S-NeuroString), while enabling precise control on the longitudinal, angular and radial positioning and distribution of the functional components. Taking advantage of the biocompatibility of our soft fibres with the dynamic and soft gastrointestinal system, we proceed to show the feasibility of our S-NeuroString for post-operative multimodal continuous motility mapping and tissue stimulation in awake pigs. We further demonstrate multi-channel single-unit electrical recording in mouse brain for up to 4 months, and a fabrication capability to produce 1,280 channels within a 230-μm-diameter soft fibre. Our soft bioelectronic fibres offer a powerful platform for minimally invasive implantable electronics, where diverse sensing and stimulation functionalities can be effectively integrated.