Summary of Nature publications, focusing on contents relevant to AI and life sciences
Humans, Biodiversity, Climate Change, Plants, Forests, South America
Researchers examined how climate change and Indigenous language loss might jointly erode Amazonia's biological and cultural heritage. Drawing on a database of over 90,000 reports from 700 sources, they identified 5,796 native plant species used by Amazonian peoples—roughly one-third of the region's known vascular seed plant flora. Running nearly 8,500 species distribution models under three future emissions scenarios, they found that utilized plant species face steeper range contractions than non-utilized ones by 2060–2080. Indigenous communities could lose 28–34% of their utilized plant species and 18–23% of associated ecosystem services. Separately, the extinction of threatened Indigenous languages could shrink the regional knowledge pool by roughly 26%. The findings highlight compounding threats to Amazonian biocultural heritage and the authors make their dataset publicly available to support restoration planning.
Humans, Artificial Intelligence, Communication, Disease Management, Large Language Models, Physicians, Primary Care, Practice Guidelines as Topic
Researchers extended AMIE, an AI diagnostic system built on large language models, to handle multi-visit clinical management conversations—not just initial diagnosis. The updated system uses Gemini's long-context capabilities combined with in-context retrieval and structured reasoning to align responses with current clinical guidelines and drug formularies. In a randomized, blinded virtual OSCE study comparing AMIE against 21 primary care physicians across 100 multi-visit cases, AMIE performed at least as well as physicians in management reasoning and scored higher on treatment precision and guideline adherence. The team also created RxQA, a medication-reasoning benchmark derived from US and UK drug formularies and validated by pharmacists; AMIE outperformed physicians on harder questions even when both could consult external drug references. While real-world deployment requires further validation, the results suggest conversational AI could meaningfully support clinical disease management.
Animals, Neuronal Plasticity, Astrocytes, Receptors, Glucocorticoid, Signal Transduction, Mice, Humans, Visual Cortex, Female, Chromatin, Male, Glucocorticoids, Gene Expression Regulation, Developmental
During early postnatal life, sensory experiences shape brain circuits through a process called critical period plasticity, but what signals eventually close these windows of heightened plasticity is poorly understood. Using paired single-cell RNA sequencing and chromatin accessibility profiling across mouse visual cortex development, researchers found that light exposure drives astrocyte maturation by recruiting the glucocorticoid receptor (Nr3c1) to chromatin in a cell-type-specific manner. This glucocorticoid receptor signaling activates a broad gene regulatory program in astrocytes that is partly conserved in human brain development and appears to promote critical period closure, thereby limiting neuronal plasticity. The findings suggest that glucocorticoid-driven astrocyte maturation acts as an environmental brake on neural plasticity, and that disruption of this pathway—such as through early-life stress—could raise susceptibility to neuropsychiatric disorders.
Saccharomyces cerevisiae, Saccharomyces cerevisiae Proteins, Models, Molecular, DNA Replication, Cryoelectron Microscopy, Adenosine Triphosphate, DNA Helicases, DNA-Binding Proteins, Cell Cycle Proteins, Minichromosome Maintenance Proteins, Multiprotein Complexes
DNA replication begins when cells enter S phase by assembling active helicase complexes called CMGE on double-stranded DNA. Researchers reconstructed this assembly process using purified yeast proteins and captured the pre-initiation complex using cryo-electron microscopy. The resulting structure reveals two CMGE helicases forming simultaneously, showing how progressive complex assembly reconfigures the MCM protein ring to prepare DNA for unwinding. The work clarifies how ATP drives the ejection of assembly factors and maturation of the helicase, and uncovers an unexpected additional role for the firing factor Sld2: beyond recruiting GINS to MCM, it also helps the two CMGE complexes physically separate and is essential for ejecting the lagging-strand DNA from the MCM channel. Because Sld2's human counterpart is RECQL4, these findings shed light on a conserved eukaryotic mechanism for establishing bidirectional replication forks.
Large Language Models, Electronic Health Records, Humans, Artificial Intelligence, Clinical Decision-Making, Physicians, Generative Artificial Intelligence
Researchers developed MIRA, an autonomous AI agent that operates within a sandboxed electronic health record (EHR) environment to handle end-to-end clinical tasks rather than isolated subtasks. Unlike typical LLM-based tools that only offer free-text advice, MIRA can gather patient histories, order and interpret lab and imaging tests, generate differential diagnoses, prescribe medications, and plan admissions or surgeries. Tested on simulations drawn from real patient cases across multiple diagnoses, MIRA outperformed physicians on diagnostic accuracy and made decisions aligned with clinical guidelines, medication safety standards, and appropriate admission criteria. The findings suggest that deeply EHR-integrated AI agents capable of translating clinical intent into structured, executable actions could serve as more effective physician decision-support tools than current narrow chatbots. The authors note that prospective real-world studies are still needed to establish safety, generalizability, and appropriate governance before clinical deployment.
HSP90 Heat-Shock Proteins, Argonaute Proteins, Humans, RNA-Induced Silencing Complex, Models, Molecular, Cryoelectron Microscopy, MicroRNAs, Molecular Chaperones, Protein Folding, Protein Domains, Protein Binding
The RNA-induced silencing complex (RISC) is central to gene silencing, consisting of an Argonaute (AGO) protein loaded with a small RNA guide. How this loading occurs has been unclear despite being known to require the chaperones HSP70 and HSP90. Researchers identified a human complex—AGO bound to HSP90 and its co-chaperone p23—that holds AGO in an RNA-free, assembly-ready state. Using cryo-electron microscopy, they solved the structure of this complex bound to a microRNA duplex, revealing that AGO adopts an unusually open conformation where its domains are separated and anchored to opposite ends of the HSP90 dimer. This open state exposes a positively charged groove that accepts the RNA duplex. Notably, RNA duplexes bearing a 5'-phosphate—but not single-stranded RNAs—trigger proper AGO folding, indicating the duplex itself acts as a structural cofactor. These insights may help guide rational design of siRNA-based therapeutics.
Animals, Humans, Autophagy, cGAS-STING Signaling Pathway, Cryoelectron Microscopy, Interferon Type I, Ligands, Membrane Proteins, Models, Molecular, Mutation, Nucleotides, Cyclic, Signal Transduction, STING Protein, Regulatory Sequences, Nucleic Acid
STING is a central immune signaling protein involved in defense against pathogens, cancer, and inflammation, but the precise sequence features controlling its activation have been unclear. Researchers built a high-throughput assay to test thousands of single amino acid substitutions across STING, mapping how each change affects its ability to trigger interferon production, cytokine signaling, and non-canonical autophagy. Cryo-EM structures of selected hyperactive variants uncovered new regulatory mechanisms governing the shift from inactive to active STING conformations. Key findings include mutations that heighten sensitivity to the natural ligand cGAMP and others that selectively uncouple interferon induction from autophagy, demonstrating that distinct signaling outputs can be dialed in through single substitutions. The dataset also illuminates the functional significance of naturally occurring human STING variants, offering a comprehensive resource for understanding STING biology and guiding therapeutic targeting of this pathway.
Animals, Male, CA1 Region, Hippocampal, CA3 Region, Hippocampal, Chiroptera, Flight, Animal, Models, Neurological, Neurons, Place Cells, Spatial Memory
The hippocampus contains place cells in two connected subregions, CA3 and CA1, which were previously thought to encode space similarly. Researchers questioned whether this apparent similarity resulted from studying animals in small arenas. To test this, they recorded CA3 and CA1 neurons simultaneously in bats navigating flight tunnels up to 200 meters long. The results revealed a striking difference: CA3 neurons displayed ultrasparse coding with mostly single place fields per cell, while CA1 neurons showed dense coding with multiple place fields each. Despite this contrast, individual field sizes were comparable across both regions and across five environment sizes. A neural network model suggested this sparse-to-dense transformation supports rapid learning of new spatial maps. In a large multi-compartment space, place cells also showed strong trajectory-history effects extending over 100 meters. The findings reveal a fundamental CA3-to-CA1 reformatting of spatial information that only becomes apparent in large-scale, naturalistic environments.
Comments (0)
No comments yet. Log in to leave a comment.
You must be logged in to leave a comment.
Login to Comment