Back

Multiscale volume electron microscopy of the human liver maps vascular-cellular architecture, organelle dynamics and inter-organelle communication

Xing, C.; Xie, R.; Mulcahy, B.; Darbandi, A.; Thoeni, C.; Chandok, I.; Lee, Y.; Ma, J.; Ali, F.; McGilvray, I.; MacParland, S.; Zhen, M.; Bader, G.

2026-04-24 cell biology
10.64898/2026.04.22.719970 bioRxiv
Show abstract

The human liver depends on multiscale structural organization from vasculature to cells to organelles to perform its diverse metabolic functions. A unified three-dimensional view linking these hierarchical scales in intact human tissue would be useful for better understanding these levels and how they relate to each other. We present a high-resolution volume electron microscopy reconstruction of human periportal liver tissue (152 x 140 x 33 {micro}m3 at 8 nm pixel size) acquired by serial block-face scanning electron microscopy. Using a multiscale deep learning approach, we performed automated segmentation across the entire volume, enabling comprehensive annotation of vasculature, cells, and organelles. Quantitative analysis of bile duct architecture revealed coordinated scaling between lumen geometry and cholangiocyte number and size. Sinusoidal capillary branches exhibited distinct structural profiles with differential endothelial coverage. Analysis of 35,790 complete mitochondria identified substantial morphological heterogeneity, with elongated mitochondria displaying preferential endoplasmic reticulum (ER) contacts at narrowing sites, a pattern consistent with ER-mediated fission or fusion. This multiscale reconstruction establishes an ultrastructural reference for the healthy human liver and provides a quantitative framework for investigating hepatic physiology and disease.

Matching journals

The top 6 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.