Full Breakdown
AI-Driven Mapping of Nucleolar Shapes Opens New Path for Drug Development
6/12/2026, 4:12:38 AM
Biomolecular Condensates and Nucleolar Morphology
Biomolecular condensates are organelles formed by phase separation of proteins and nucleic acids. The nucleolus, a condensate that assembles ribosomal subunits, exhibits shape changes that correlate with cellular stress. Prior work links cap- and necklace-shaped nucleoli to disruptions in RNA processing and to diseases such as Alzheimer’s, ALS, and cancer.
Research Team, Institutions, and Funding
Led by Cliff Brangwynne, June K. Wu ’92 Professor of Chemical & Biological Engineering at Princeton, the study lists co-authors Anita Donlic, Troy Comi, Sofia Quinodoz, Nima Jaberi-Lashkari, Krist Antunes Fernandez, Lifei Jiang, Lennard Wiesner and Ai Ing Lim. Funding came from Howard Hughes Medical Institute, Princeton Center for Complex Materials (NSF MRSEC DMR-2011750), St. Jude Collaborative on Membraneless Organelles, Chan Zuckerberg Initiative Exploratory Cell Network and Princeton Laboratory for Artificial Intelligence.
Imaging and Neural-Network Classification
The team used a microscope to capture nucleolar images from hundreds of cells exposed to a panel of drugs at varying concentrations. Tens of thousands of images—representing healthy spherical nucleoli, cap shapes, and necklace shapes—trained a convolutional neural network to classify morphology into four categories.
Discovery of a Novel “Flower” Nucleolar Shape
The network sorted images into the three known categories—spherical, cap, necklace—and a novel “flower” shape. Two anti-cancer drugs induced caps, an unreported effect. Topotecan, a TOP1 inhibitor, produced the flower morphology, linking TOP1 to nucleolar organization through RNA processing. Dose-response curves showed graded shifts in cap and necklace frequencies with drug concentration.
Official Statements & Interpretation
Brangwynne framed the study as tackling central problem of linking emergent structure to molecular interactions, noting the innovation of learning directly from images. Donlic warned that visual analysis can miss important features, emphasizing the network’s ability to uncover hidden phenotypes.
Verbatim Quotes
“The central problem in biology is how do you get emergent structure from individual molecular interactions,” — Cliff Brangwynne, June K. Wu ’92 Professor of Chemical & Biological Engineering
“No one’s seen this flower morphology before,” — Cliff Brangwynne
“The network flagged it as not fitting neatly into the other three categories.” — Cliff Brangwynne
“You could be missing other important features,” — Anita Donlic, first author
Implications for Drug Development
The neural-network tool links nucleolar shape to functional outcomes, providing a single-cell level assay for monitoring cellular responses to drugs. By quantifying morphology changes, the approach offers a way to evaluate compound effects on RNA-related condensates.
Conflicting Reports & Gaps
The study presents initial observations within cultured human cells; external validation in additional cell types or in vivo models was not reported. Consequently, the generalizability of the flower morphology as a biomarker and its relevance to clinical outcomes remain open questions.
What’s Next: Expanded Applications
The team plans to apply the neural-network pipeline to other RNA-related condensates, including nuclear speckles and viral replication compartments, and to integrate the assay into drug evaluation pipelines. Further collaborations aim to test predictive power in disease models.
