欧亚国际

来自丹麦奥胡斯大学的科学家，利用Videometer公司为歌本哈根大学大学构建的原位多通道多光谱根系成像系统发表了题为Genomic prediction of yield and root development in wheat under changing water availability的文章，文章发表于Plant Methods上。

丹麦Videometer公司开发的原位根系多光谱表型成像系统，是做根系研究的革新性专业设备，无论对于浅根系园艺蔬菜、作物种质资源、草种质资源还是深根系林木种质资源，都具有现实性研究意义。目前在根系研究尤其是表型研究领域中，对于草类、玉米根系和小麦根系所作的研究比较多，但大多还采用传统不可重复的挖掘方法。植物根系原位多光谱表型成像系统出现，改变了这种情况，使得植物研究人员在对根系进行研究的过程中，可以使用原位的方式、高分辨率、无损伤的进行监测，多光谱成像技术，因具有图谱合一的特点，今年成为植物科学研究的热点。

原位根系多光谱微根管表型成像系统

该系统构建为多通道原位根系多光谱微根管表型成像系统，主要用于设施规划中的高通量根系成像研究。丹麦根本哈根大学科学家等此前已经利用该多光谱成像系统对植物植株、根系进行成像研究，取得了前瞻性的成果。该前研究以深根系大麦为研究对象，将大麦下方埋了有3m长的微根管，使用Videometer公司的Videometer MR多光谱成像系统，定期通过根窗透明面对根系成像分析。原始光谱图像经过Videometer自带软件一系列算法处理后得到目标根系图像，随后进行阈值分割、模糊聚类等模型分析，得到根系的形态学数据。

多通道原位根系多光谱表型成像系统

根系是植物主要吸水、营养物等器官，通过对根系监测和研究，能优化水肥方案，促进农作物、林业等产业增产增效，有利于土地荒漠化治理、土壤修复等。但长期以来，对根系研究主要是采用挖掘法、土钻法、土柱法、容器法、剖面法、传统可见光相机成像法等传统方法，采样破坏性大、工作量大、区分效果不佳，严重阻碍了根系研究的深入开展。《科学》杂志曾出版专辑认为，“人类对自己脚下土壤的了解远远不及对宇宙的了解”，更是佐证了地下根系研究、生态学研究难度之大。因此，对根系研究方法的选择和改进，对科研结果影响巨大。

5个波段下多光谱成像（405、450、590、660、940）

5波段多光谱假彩RGB成像图

四通道5波段多光谱根系微根管成像系统（图片来自歌本哈格大学、禁止盗图、侵权必究）

传统的RGB可见光成像技术目前是业界使用较多的技术，是利用颜色识别根系，前提是根系和土壤之间要有比较明显的色差，但实际根系生长在土壤中，颜色差异并不明显，这样根系识别可能会造成比较大的误差，RGB可见光成像技术使用就会受限。歌本哈根将多光谱成像技术和传统的RGB成像技术进行了对比，显示多光谱成像技术基于光谱特征在根系识别上的明显优势，该系统可对颜色精确定量，符合国际通用的CIE色域空间颜色标准，可以区分异质的物质，如土壤和植物组织，可对土壤和根系分辨进行图像切割，专门对ROI感兴趣区域进行研究，也可区分新根和宿根以及正常根与发生病害的根系，系统分辨率高，可达30um/像素。科学家对多光谱成像的功能进行了探讨-即多光谱特征对于根系生化特性的识别（例如细根发生、成熟、衰老、死亡的周转过程；例如根际分泌物成分的变化等），显示了多光谱成像技术在根系研究领域的巨大潜力。

Construction of a large-scale semi-feld facility to study genotypic diferences in deep root growth and resources acquisition 

Genomic prediction of yield and root development in wheat under changing water availability

Xiangyu Guo 1, Simon F Svane 2, Winnie S Füchtbauer 3, Jeppe R Andersen 4, Just Jensen 1, Kristian Thorup-Kristensen 

Abstract

Background: Deeper roots help plants take up available resources in deep soil ensuring better growth and higher yields under conditions of drought. A large-scale semi-field root phenotyping facility was developed to allow a water availability gradient and detect potential interaction of genotype by water availability gradient. Genotyped winter wheat lines were grown as rows in four beds of this facility, where indirect genetic effects from neighbors could be important to trait variation. The objective was to explore the possibility of genomic prediction for grain-related traits and deep root traits collected via images taken in a minirhizotron tube under each row of winter wheat measured.

Results: The analysis comprised four grain-related traits: grain yield, thousand-kernel weight, protein concentration, and total nitrogen content measured on each half row that were harvested separately. Two root traits, total root length between 1.2 and 2 m depth and root length in four intervals on each tube were also analyzed. Two sets of models with or without the effects of neighbors from both sides of each row were applied. No interaction between genotypes and changing water availability were detected for any trait. Estimated genomic heritabilities ranged from 0.263 to 0.680 for grain-related traits and from 0.030 to 0.055 for root traits. The coefficients of genetic variation were similar for grain-related and root traits. The prediction accuracy of breeding values ranged from 0.440 to 0.598 for grain-related traits and from 0.264 to 0.334 for root traits. Including neighbor effects in the model generally increased the estimated genomic heritabilities and accuracy of predicted breeding values for grain yield and nitrogen content.

Conclusions: Similar relative amounts of additive genetic variance were found for both yield traits and root traits but no interaction between genotypes and water availability were detected. It is possible to obtain accurate genomic prediction of breeding values for grain-related traits and reasonably accurate predicted breeding values for deep root traits using records from the semi-field facility. Including neighbor effects increased the estimated additive genetic variance of grain-related traits and accuracy of predicting breeding values. High prediction accuracy can be obtained although heritability is low.

Keywords: Deep root; Genomic prediction; Grain-related yield; Semi-field; Wheat.

Root length from minirhizotron imaging

Half of the experimental beds were equipped with minirhizotron tubes, and two root traits were analyzed in this study. Root imaging was done by using a multispectral imaging system, in which a portable trolley system and four multispectral camera systems were used to allow for multivariate image analysis using fve wavebands [32]. When taking the image, the portable trolley carried four cameras moved through facility, with a step size of four rows in each movement. Ten the side by side cameras were dropped into four adjacent minirhizotron tubes (one camera in each tube) and images of the root were

taken along the tube with 5 cm intervals. Te root images were made along the upwards facing side of the minirhizotrons, therefore enable photography of roots covering a soil depth interval of 0.7 m to 2.7 m [6]. Te subsequent image analysis delivered an estimate of living root length in each image using the U-

Net Neural Network (CNN) architecture to provide automated image segmentation of root structures [47]. A detailed description of the image

analysis strategy can be found in a previous study [31]. On average, there were 56 images (4×5  cm) used for each minirhizotron tube after editing. Te root data were based on root imaging made on 18th June 2018, where 21,057 root images were recorded by four cameras from the 300 minirhizotrons. Te imaging of roots was done at late fowering early grain flling since it is widely accepted that the cereal root system reaches the maximum extension after anthesis and limited root development gave been observed during grain flling [48]. Total root length between 1.2 and 2.0 m soil depth (TRL, Additional fle 1:

Table S3) was expressed as the total length of living roots found in all the images taken from each minirhizotron tube. Root length in four intervals on each tube (IRL, Additional fle  1: Table  S4) was expressed as the total length of living roots found in all the images taken from each depth interval in each minirhizotron tube. Te root data were edited as follows before the genetic analysis:Te number of records kept in each step can be found in Additional fle 2: Table S5 and the detail depth interval can be found in Additional fle 2: Table S6. After editing by rules from step 1 to 4, 14,270 records (images) were kept for further analysis, i.e. to calculate root length from intervals in each tube and root length for each tube.