🐟 COVER STORY — Arapaima Conservation: A Community-Based Success Story in the Amazon

News Feature · p. 572 · https://www.science.org/content/article/remote-amazon-locals-are-saving-giant-fish-and-helping-their-villages

A giant fish called the arapaima is the focus of a conservation success story in the Brazilian Amazon. Scientists have worked with local villages to protect the commercially valuable fish from poachers while allowing villagers to profit from a sustainable arapaima harvest. The project has spurred development and led to the protection of some 150,000 km² of tropical forest. Along Brazil's Juruá River, village guardians patrol the maze of interconnected creeks and lakes against poachers, miners, and others bent on pillaging the surrounding waters and forests. For one local participant, the arapaima management program enabled him to build his own home and stay in the community rather than relocating to a larger town. This story illustrates how participatory fisheries management — increasingly monitored using remote sensing and population-tracking algorithms — can achieve conservation at landscape scale. ScienceScience

🤖 Protein Language Models Power Rapid Directed Evolution — Tran, Moon, Hie, Hsu et al. (Research Article)

DOI: 10.1126/science.aea1820

Protein engineering is limited by the inefficient search through a high-dimensional sequence space to find combinations of synergistic mutations. Traditional approaches use stepwise mutation stacking, whereas machine learning methods require extensive data. Protein sequences are highly degenerate, with many possible functional sequences and complex interactions between pairs of otherwise neutral or deleterious mutations. Finding desired states in this space can be experimentally demanding, and there has been much interest in using protein language models to find improved variants. This paper from the Hie and Hsu labs demonstrates a new framework that combines protein language model guidance with epistatic interaction modeling to achieve rapid directed evolution — a major step forward for AI-driven protein and therapeutic design. Science

🧬 CRYSTAL Nanoparticle Activates STING Pathway Systemically to Fight Tumors — Zhou, Irvine et al. (Research Article)

DOI: 10.1126/science.adx1893

The cyclic guanosine monophosphate–adenosine monophosphate synthase–stimulator of interferon genes (cGAS-STING) pathway drives antitumor immunity but has been difficult to activate systemically owing to poor pharmacology and toxicity. CRYSTAL self-assembles from manganese ions intercalated with cyclic dinucleotides, enabling precise structural control. After systemic administration, CRYSTAL selectively activates myeloid cells and expands antitumor CD8+ T cells, leading to robust STING-driven tumor regression in multiple preclinical models. The biomimetic design — using metal-ion self-assembly inspired by natural systems — represents a new class of cancer immunotherapy nanomedicine. Science

🧠 Apical Tuft Dendrites Are the Neural Substrate of Flexible Learning — Larkum et al. (Research Article)

DOI: 10.1126/science.adx4358

Flexible learning relies on integrating sensory and contextual information to adjust behavioral output in different environments. The anterolateral motor cortex (ALM) is a frontal area critical for action selection in rodents. Inputs critical to decision-making converge on the apical tuft dendrites of layer 5b pyramidal neurons in ALM. Activation of dendrite-inhibiting layer 1 interneurons impaired relearning, without affecting previously learned behavior. This inhibition profoundly suppressed global calcium activity in dendritic shafts but not local transients in spines, while additionally reducing burst firing. Excitatory synaptic inputs to tuft dendrites exhibited rule-dependent clustering. Dendritic calcium signaling is therefore a key computational component of flexible learning. This finding has direct implications for AI-inspired architectures — the two-compartment pyramidal neuron model of contextual vs. sensory integration is increasingly influential in neuromorphic computing. Science

⚡ Neuromorphic Ionic Computing: Brain-Inspired, Ion-Based AI Hardware — Aluru, Darling, Elam, Talin et al. (Review, p. 592)

DOI: 10.1126/science.aea2097

Neuromorphic ionic computing is inspired by the brain's use of ions for ultralow-energy computation — its massive parallelism, adaptability, and learning capabilities. This emerging paradigm can overcome limitations of conventional silicon-based computing. This review surveys the rapidly growing field of ion-based computing devices that mimic synaptic and neural behavior, identifying key knowledge gaps standing between current prototypes and practical, brain-like computation — an area with deep connections to both neuroscience and next-generation AI hardware. Science

📊 Aging Researchers Produce Less Disruptive Science — Cui, Lin, Wu, Evans (Policy Article, pp. 588–591)

DOI: 10.1126/science.ady8732

Aging researchers and the removal of retirement policies yield decreased disruptive innovation in science. Analyzing more than 12.5 million scientists who published between 1960 and 2020, the authors find that novelty — the linking of previously unconnected ideas — increases with academic age, whereas disruption — the replacement of established paradigms — decreases. A new policy article uses large-scale data and novel deep learning measurements to measure the impact of scientific aging. The data show how younger scientists tend toward disruptive contributions while older scientists engage in combinatorial innovation with aging works. The findings highlight the benefits of policies that reimagine funding and encourage independent support for young scientists and unexpected collaborations. Notably, the analysis itself is powered by deep learning NLP — a methodological model for AI-assisted science policy research. ScienceUniversity of Chicago

Issue theme in brief: The 7 May issue is one of the strongest in this series for AI and life sciences, anchored by a landmark protein language model paper for directed evolution, a breakthrough cancer immunotherapy nanoparticle, and a transformative neuroscience finding about the dendritic basis of flexible learning — which also speaks directly to AI architecture debates. The neuromorphic ionic computing review and the AI-powered science-of-science policy paper round out a remarkably AI-rich issue.