Summary of Nature publications, focusing on contents relevant to AI and life sciences
Researchers investigated how climate change and Indigenous language loss might together threaten the biological and cultural heritage of Amazonia. Drawing on a database of over 90,000 records from 700 sources, they catalogued the uses of 5,796 native plant species—roughly one-third of the Amazon's known vascular seed plant flora—across all basin countries. They then ran species distribution models for nearly 8,500 of these plants under three future climate scenarios to project range shifts by 2060–2080. The analysis revealed that plants with documented human uses face steeper range contractions than unused species. Indigenous communities could lose 28–34% of their culturally important plant species and 18–23% of the ecosystem services those plants provide. Separately, the decline of threatened Indigenous languages could erase roughly 26% of the region's accumulated botanical knowledge. The findings highlight compounding risks to Amazonian biocultural heritage from both ecological and linguistic dimensions of global change.
Researchers tested whether a large language model system could handle the clinical management side of medicine — not just diagnosis, but reasoning about disease progression, treatment decisions, and medication prescribing. They upgraded a system called AMIE using an agentic architecture that combines long-context retrieval with structured reasoning, anchoring its outputs to current clinical guidelines and drug formularies. In a randomized blinded virtual clinical exam, AMIE was compared against 21 primary care physicians across 100 multi-visit case scenarios. Specialist evaluators rated AMIE as non-inferior to physicians in management reasoning overall, and found it more precise in selecting treatments and investigations, and better aligned with clinical guidelines. The team also introduced RxQA, a medication-reasoning benchmark built from US and UK drug formularies and validated by pharmacists; AMIE outperformed physicians on harder questions even when both could consult external drug references. The results suggest conversational AI could meaningfully support disease management, though real-world deployment requires further study.
Researchers investigated how the brain's developmental "critical periods" — windows of heightened plasticity during which sensory experience shapes neural circuits — eventually close as animals mature. Using simultaneous single-cell RNA sequencing and chromatin accessibility profiling in the mouse primary visual cortex across postnatal development, they tracked how different cell types respond to light exposure. A key finding was that light drives astrocyte maturation by recruiting the glucocorticoid receptor (Nr3c1) to chromatin specifically in those cells, activating a broad gene regulatory program partly conserved in human brain development. This astrocyte-specific glucocorticoid signaling appears to promote critical period closure and limit further neuronal plasticity. The work also raises the possibility that early-life stress, which disrupts glucocorticoid signaling, could impair normal astrocyte maturation and thereby increase vulnerability to neuropsychiatric disorders, positioning astrocytes as underappreciated regulators of experience-dependent brain development.
Researchers developed MIRA, an autonomous AI agent designed to function within an electronic health record (EHR) environment rather than simply answering clinical questions in isolation. Unlike typical LLM-based tools that address single tasks or offer free-text recommendations, MIRA was built to take structured actions: gathering patient histories, ordering and interpreting labs and imaging, constructing differential diagnoses, prescribing medications, and planning admissions or surgical procedures. Tested on simulations drawn from real patient cases covering multiple diagnoses, MIRA exceeded physician-level diagnostic accuracy and made treatment decisions consistent with clinical guidelines, appropriate medication safety standards, and correct admission choices. The key advance is translating clinical reasoning into concrete EHR operations within defined safety boundaries. The authors caution that prospective real-world studies are still needed to confirm generalizability, safety, and appropriate governance before this kind of system could be deployed alongside clinicians.
Researchers investigated how small RNAs are loaded onto Argonaute (AGO) proteins to form the RNA-induced silencing complex (RISC), a process central to gene silencing but mechanistically unclear. They identified a human complex—AGO bound to chaperones HSP90 and p23, called the AGO maturation complex (AMC)—that holds AGO in an RNA-free, unfolded state competent for RNA loading. Using cryo-electron microscopy, they solved the AMC structure bound to a microRNA duplex and found AGO adopts a dramatically open conformation, with its N domain and RNA-binding module anchored to opposite ends of the HSP90 dimer and connected only by an unfolded linker, creating a positively charged RNA-binding cleft. Notably, RNA duplexes with a 5'-terminal phosphate—but not single-stranded RNAs—promote AGO folding, revealing the duplex itself acts as a chaperone-like cofactor. These findings clarify RISC assembly and offer a platform for designing improved small interfering RNA therapeutics.
Researchers conducted a comprehensive investigation into how the amino acid sequence of the immune protein STING determines its signaling behavior. STING plays a central role in detecting infection and cellular stress, triggering interferon production, inflammatory signaling, and autophagy, but the sequence-level rules governing its activation have been poorly defined. Using a massively parallel mutagenesis screen, the team profiled thousands of single amino acid variants to map which positions enhance or suppress STING activity. Cryo-electron microscopy of selected hyperactive variants revealed new structural mechanisms controlling the switch from inactive to active conformations. Key findings include mutations that heighten sensitivity to the natural ligand 2'3'-cGAMP and others that selectively uncouple interferon production from autophagy—showing these outputs can be independently tuned. The dataset also illuminates the functional significance of naturally occurring human STING variants, with implications for understanding disease risk and guiding the development of STING-targeted therapeutics.
Researchers questioned why hippocampal subregions CA3 and CA1 — consecutive processing stages with distinct anatomical wiring — appear to encode space identically in most experiments. They proposed this apparent similarity stems from the use of small testing arenas, which may not reveal coding differences. To test this, they simultaneously recorded from CA1 and CA3 neurons in bats navigating flight tunnels up to 200 meters long. In large environments, the two regions showed clearly different coding strategies: CA1 neurons had dense representations with multiple place fields per cell, while CA3 neurons were ultrasparse, mostly carrying single place fields. Notably, individual place field sizes remained comparable across both regions at all five tested scales (6–200 m). A neural-network model suggested this sparse-to-dense transformation supports rapid learning of new spatial maps. The team also observed strong trajectory-history effects in a multi-compartment environment, with retrospective contextual coding extending over 100 meters.
Comments (0)
No comments yet. Log in to leave a comment.
You must be logged in to leave a comment.
Login to Comment