Three-dimensional visualization of moss rhizoid system by refraction-contrast X-ray micro-computed tomography

Abstract

Land plants have two types of shoot-supporting systems, root system and rhizoid system, in vascular plants and bryophytes. However, since the evolutionary origin of the systems are different, how much they exploit common systems or distinct systems to architect their structures are largely unknown. To understand the regulatory mechanism how bryophytes architect rhizoid system responding to an environmental factor, such as gravity, and compare it with the root system of vascular plants, we have developed the methodology to visualize and quantitatively analyze the rhizoid system of the moss, Physcomitrium patens in 3D. The rhizoids having the diameter of 21.3 m on the average were visualized by refraction-contrast X-ray micro-CT using coherent X-ray optics available at synchrotron radiation facility SPring-8. Three types of shape (ring-shape, line, black circle) observed in tomographic slices of specimens embedded in paraffin were confirmed to be the rhizoids by optical and electron microscopy. Comprehensive automatic segmentation of the rhizoids which appeared in different three form types in tomograms was tested by a method using Canny edge detector or machine learning. Accuracy of output images was evaluated by comparing with the manually-segmented ground truth images using measures such as F1 score and IoU, revealing that the automatic segmentation using the machine learning was more effective than that using Canny edge detector. Thus, machine learning-based skeletonized 3D model revealed quite dense distribution of rhizoids, which was similar to root system architecture in vascular plants. We successfully visualized the moss rhizoid system in 3D for the first time.

Plant phenotyping relevance

コケの根茎系を3D可視化・定量化するX線マイクロCTと自動セグメンテーション手法を開発し、教師データとの比較で精度検証しているため、植物表現型取得法が中心である。

Code and data availability

The paper describes X-ray micro-CT imaging of moss rhizoids and segmentation via Fiji plugins and a custom Python program, but no public deposit of tomograms, ground-truth annotations, trained classifiers, or the authors' analysis code is stated. The URLs mentioned (SP-µCT software package, Fiji, Canny plugin) are pre-

No evidence-backed public reproduction asset is currently recorded.