Back

A 3D image atlas chronicling cellular and structural dynamics following lung injury identifies the aberrant expansion of endothelial cells that fail to form perfused vasculature

Xia, J.; Gupta, A.; Poupard, E.; Helms, H.; Baker, B.

2026-07-23 bioengineering
10.64898/2026.07.22.740063 bioRxiv
Show abstract

Intratracheally delivered bleomycin in mice is the most widely used in vivo model of pulmonary fibrosis, yet key aspects remain poorly defined, including sex-dependent responses and the temporal peak of injury. Standard 2D histology further overlooks regional heterogeneity and cannot resolve the 3D architecture or connectivity of endothelial cells (ECs). Here, we established a multi-scale 3D imaging pipeline integrating precision-cut lung slices from EC lineage-tracing mice, optical clearing, and AI-driven 3D segmentation to map cellular and structural dynamics from whole-lobe tile scans to single-cell resolution. We identified sex as a critical biological variable, with males exhibiting a delayed but more severe fibroproliferative response. Unsupervised K-means clustering identified three distinct tissue microenvironments: healthy parenchyma(KMC1), a myofibroblast-rich fibrotic core (KMC2), and a previously uncharacterized EC-dense perilesional region (KMC3) defined by massively expanded but non-perfused ECs that acquire a pro-inflammatory phenotype. This aberrant endothelial response precedes peak myofibroblast accumulation and persists beyond fibrotic resolution, leaving a vascular scar that extends into the large-vessel hierarchy. Together, this 3D image atlas, made publicly available as an interactive resource [https://mosaic-lung.com/], reveals the activated endothelium as an underexplored therapeutic target in pulmonary fibrosis.

Matching journals

The top 5 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.