This groundbreaking research reveals the physical architecture of memory engrams in the mouse hippocampus, pushing the boundaries of neuroscience through the integration of artificial intelligence with advanced 3D electron microscopy. By combining chemogenetic tagging of neurons with AI-driven image segmentation, the researchers achieved an unprecedented nanoscale reconstruction of hippocampal circuits involved in memory. Contrary to classical Hebbian theory, they found that neurons active during memory formation do not preferentially wire with each other; instead, they expand their connectivity via complex multisynaptic boutons (MSBs), enabling broader and more flexible information encoding. These structural changes were accompanied by synapse-specific enlargement, mitochondrial remodeling, and altered interactions with astrocytes—signatures linked specifically to negative-valence associative learning. The findings challenge traditional models of memory storage and suggest a novel mechanism where synaptic complexity, not just strength or density, enhances memory capacity. Overall, this study provides a structural foundation for understanding how the brain encodes, generalizes, and maintains long-term memories.

Explore and leverage Deep Research in ChatGPT to dramatically speed up and elevate your research work. Deep Research, powered by OpenAI’s o3 reasoning model, autonomously browses, analyzes, and synthesizes hundreds of sources to deliver professional-grade, well-cited reports in minutes. Researchers Isa Fulford and Zhiqing (Edward) Sun emphasized that the best results come from clearly stating your research objective while giving the model creative freedom. OpenAI is expanding access and continuously improving safety, making Deep Research a transformative tool for professionals across science, finance, law, and more.