NuclearIDTracker resolves intestinal cell identity and lineage dynamics through nuclear phenotypic signatures
Organoid models have transformed our understanding of intestinal renewal. Fluorescent imaging has been extensively used to identify key cell types and their differentiation pathways, but immunofluorescence provides only static readouts, whereas live imaging requires fluorescent-reporter engineering and is constrained by limited multiplexing and spectral overlap. Here, we introduce NuclearIDTracker, an explainable machine-learning framework that infers cell identity directly from 3D nuclear segmentations. Using a single nuclear marker, NuclearIDTracker accurately classifies intestinal cell types and integrates with single-cell tracking to resolve lineages and reconstruct dynamic state transitions during organoid development. We show that TA-like cells, rather than stem cells, drive early crypt formation and generate enterocyte and Paneth lineages, as well as the stem-cell population, which emerges only later and subsequently replenishes the TA-like compartment. Following stem-cell ablation, crypt regeneration was not driven by a single discrete cell type. Instead, multiple epithelial populations converged on a proliferative regenerative state with a nuclear phenotypic signature that resembled, but remained distinct from, that of homeostatic TA-like cells, and a YAP/TAZ-associated fetal-like transcriptional signature. Thus, nuclear phenotypic signatures resolve cell identity and reveal coordinated epithelial plasticity during crypt regeneration. NuclearIDTracker establishes a non-perturbative tool to quantify cell identity and state dynamics at single-cell resolution, revealing previously inaccessible biological dynamics and expanding the toolkit for studying epithelial homeostasis, regeneration, and disease.