诺和诺德

Electrochemical gradients are essential to the functioning of cells and form across membranes using active transporters. Here we show in contrast that condensed biomolecular systems-often termed condensates-sustain pH gradients without any external energy input. By studying individual condensates on the micrometre scale using a microdroplet platform, we reveal dense-phase pH shifts towards conditions of minimal electrostatic repulsion. We demonstrate that protein condensates can drive substantial alkaline and acidic gradients, which are compositionally tunable and can extend to complex architectures sustaining multiple unique pH conditions simultaneously. Through in silico characterization of human proteomic condensate networks, we further highlight potential wide-ranging electrochemical properties emerging from condensation in nature, while correlating intracellular condensate pH gradients with complex biomolecular composition. Together, the emergent nature of condensation shapes distinct pH microenvironments, thereby creating a regulatory mechanism to modulate biochemical activity in living and artificial systems.

BACKGROUND:Evidence, including animal, clinical, and real-world studies in individuals with type 2 diabetes and/or obesity, suggests reduced risk of dementia and Alzheimer's disease after GLP-1 receptor agonist exposure. The evoke and evoke+ trials aimed to investigate the efficacy and safety of oral semaglutide in individuals with early Alzheimer's disease. METHODS:evoke and evoke+ were multicentre, randomised, double-blind, placebo-controlled phase 3 trials conducted across 566 sites in 40 countries. The trials assessed the efficacy and safety of oral semaglutide up to 14 mg once daily in participants with amyloid-confirmed Alzheimer's disease, aged 55-85 years, with mild cognitive impairment or mild dementia due to Alzheimer's disease. In evoke+, participants with significant small vessel pathology were included. Participants were randomly assigned (1:1) to once-daily semaglutide 14 mg (flexible dose) or placebo for up to 156 weeks. The primary endpoint was change in Clinical Dementia Rating-Sum of Boxes (CDR-SB) score from baseline to week 104, assessed in all randomised participants. Safety was assessed in all randomised participants and reported for those receiving at least one dose of study drug. These trials were registered at ClinicalTrials.gov (NCT04777396 and NCT04777409); both trials have been discontinued due to negative clinical outcome. FINDINGS:Between May 18, 2021, and Sept 8, 2023, 9981 participants were screened, of whom 3808 were randomly assigned; 1855 in evoke (semaglutide, n=928; placebo, n=927) and 1953 in evoke+ (semaglutide, n=976; placebo, n=977). Mean age was 72·2 years (SD 7·1), and mean CDR-SB score was 3·7 (SD 1·6) at baseline. In evoke+, 54 (2·8%) participants had small vessel pathology. In evoke and evoke+, mean changes in CDR-SB score from baseline to week 104 were 2·3 (SE 0·1) and 2·2 (0·1) with semaglutide, compared with 2·3 (0·1) and 2·1 (0·1) with placebo (estimated difference -0·08 [95% CI -0·35 to 0·20], p=0·57 in evoke and 0·10 [-0·17 to 0·38], p=0·46 in evoke+). Treatment-emergent adverse events were reported in 1729 (91·2%) of 1896 participants receiving semaglutide versus 1613 (84·8%) of 1902 receiving placebo. There were five fatalities considered treatment-related by the investigators (one in the semaglutide group and four in the placebo group). INTERPRETATION:Oral semaglutide was not efficacious in slowing clinical progression in participants with early Alzheimer's disease. Safety and tolerability of semaglutide in early Alzheimer's disease is consistent with studies in other indications. FUNDING:Novo Nordisk.

BACKGROUND:Biomedical Large Language Models (LLMs) combined with prompt engineering offer domain-specific reasoning, yet their application to individual-level causality assessment remains unexplored. This study evaluated five combinations of biomedical LLMs, prompting strategies, and causality algorithms by comparing their agreement with two human expert evaluators. RESEARCH DESIGN AND METHODS:A total of 150 Individual Case Safety Reports (ICSRs) were analyzed: 140 reports from Food and Drug Administration Adverse Event Reporting System (FAERS), and 10 myocarditis/pericarditis ICSRs from Vaccine AERS (VAERS). Assessments were conducted using the Naranjo and WHO-UMC algorithms. Biomedical LLMs tested included TinyLlama 1.1B, Medicine LLaMA-3 8B, and MedLLaMA v20, combined with Chain-of-Thought (CoT) or Decomposition prompting. Agreement was measured using Gwet's Agreement Coefficient 1 (AC1) and percentage agreement, alongside performance metrics and qualitative error analysis. RESULTS:The Medicine LLaMA-3 8B-Naranjo-CoT combination achieved the highest agreement with human assessors for the final classification of causality (64%). Biomedical LLMs demonstrated low inter-rater agreement on critical items of causality assessment such as identification of listed AE, temporal plausibility, alternative causes, and objective evidence of AEs. Frequent model failures included irrelevant responses. CONCLUSIONS:Biomedical LLMs showed improved performance over general purpose models previously tested but remain suboptimal for reliable causality assessment of ICSRs.

Randomized controlled trials (RCTs) remain the gold standard for evaluating medical interventions, but they often face challenges related to patient recruitment, cost, and efficiency. Real-world data (RWD) has emerged as a valuable tool to enhance trial design, improve patient identification, and support regulatory decision-making. However, integrating RWD into RCTs presents methodological, regulatory, and operational challenges. To address these issues, a think tank was convened in May 2024 at the Duke Clinical Research Institute, bringing together experts from academia, industry, healthcare systems, regulatory agencies, and patient advocacy groups. Discussions focused on three key areas: optimizing patient identification and outcome assessment, leveraging RWD for safety assessments, and using RWD in RCTs supporting regulatory approval. RWD has the potential to simplify eligibility criteria, enhance recruitment through artificial intelligence, and provide practical endpoints for evaluating treatment effects. The think tank underscored the need for collaboration across stakeholders to address challenges, such as data inconsistencies, privacy concerns, and infrastructure limitations. The event concluded with actionable recommendations, including the following: (1) standardizing RWD sources to ensure consistency and improve interoperability across healthcare systems, (2) developing regulatory frameworks that define acceptable use cases for RWD in clinical trials, (3) enhancing data quality through robust validation methodologies and real-time monitoring, (4) investing in artificial intelligence–driven patient identification tools to streamline recruitment, and (5) fostering multi-stakeholder collaboration to align expectations and share best practices. Moving forward, implementing these strategies will be critical to fully harness the potential of RWD in clinical research and improve trial efficiency.

As data-science agents shift from co-pilots to auto-pilots, silent misframing becomes a critical failure mode. Agents quietly commit to plausible but unintended task framings, producing clean, executable artifacts that hide their incorrect assessment of the task. Existing benchmarks score whether the pipeline runs, ignoring whether the agent recognized the task was underspecified. We introduce Ambig-DS, two diagnostic suites: one for prediction-target ambiguity (Ambig-DS-Target, 51 tasks built on DSBench, a tabular modeling benchmark) and one for evaluation-objective ambiguity (Ambig-DS-Objective, 61 tasks built on MLE-bench, a Kaggle-style ML competition benchmark), constructed so that scoring uses each source benchmark's original evaluator. For every task we pair the original, fully specified version with an ambiguous variant produced by controlled edits; a human-and-LLM verification pipeline confirms each variant admits multiple plausible interpretations with decision-relevant consequences. The suites are analyzed independently and ambiguity lowers performance in both. Across five agents spanning efficient to frontier-class models, we find in our controlled diagnostic setting: (i) failures are silent commitments: wrong-target submissions on Target, wrong-metric or non-committal baseline submissions on Objective, rather than execution errors; (ii) allowing the agent to ask one clarifying question recovers much of the loss under idealized conditions, suggesting missing framing information drives a substantial part of the observed degradation; but (iii) agents cannot reliably tell when to use it: permissive prompts induce over-asking on clear tasks, while conservative prompts induce silent defaulting on ambiguous ones. Recognizing target and objective underspecification, not pipeline execution, is the bottleneck missing from standard DS-agent evaluations.