Whole-Organism 3D Maps Reveal Parallel Self-Similar Vascular and Neural Networks with Divergent Murray-Like Exponents

Link

Comprehensive, cellular resolution reconstructions of embryonic vasculature and innervation have so far been constrained by the depth limits of optical clearing, the restricted fields-of-view of light-sheet microscopy and the labor-intensity of manual annotation. We present a unified, fully automated pipeline that converts routine serial H&E histology into organism-wide three-dimensional atlases of blood vessels and peripheral nerves.

Convolutional neural networks, iteratively trained on a reduced set of manually attributed labels, achieve ≥93 precision and recall in semantic segmentation and a hybrid rigid-elastic registration restores anatomical continuity across thousands of sections; a TEASAR-based skeletonization yields graphs listing radii, lengths, and branching orders. Applied to four vertebrate embryos - Macaca mulatta at Carnegie stages 33 and 40, Mus Musculus at E11 and Trachemys Scripta at G17, with the last three being matched in development - the method generates vasculature and innervation reconstructions at cellular resolution across whole organisms.

Quantitative analysis shows that both networks are self-similar, with nearly constant Horton ratios for segment number, radius and length. Vascular branch points obey Murray’s cube law, whereas nerves follow a Murray-like relation with an exponent close to 2, consistent with axial-current conservation. The two macaque stages differ only by uniform scaling and mouse and turtle values overlap with the corresponding macaque. These atlases thus offer a quantitative reference for developmental and allometric studies, while the reliance on standard histology allows for prospective multi-omic integration.